The BCPL Cintsys and Cintpos

User Guide
by
Martin Richards
[email protected]

http://www.cl.cam.ac.uk/users/mr10/

Computer Laboratory
University of Cambridge
Revision date: Sat 30 Oct 16:32:25 BST 2021

Abstract
BCPL is a simple systems programming language with a small fast compiler which
is easily ported to new machines. The language was first implemented in 1967
and has been in continuous use since then. It is a typeless and provides machine
independent pointer arithmetic allowing a simple way to represent vectors and
structures. BCPL functions are recursive and variadic but, like C, do not allow
dynamic free variables, and so can be represented by just their entry addresses.
There is no built-in garbage collector and all input-output is done using library
calls.
This document describes both the single threaded BCPL Cintcode System
(called Cintsys) and the Cintcode version of the Tripos portable operating system
(called Cintpos). It gives the definition of standard BCPL including the recently
added features such as floating point expressions and constructs involving oper-
ators such as <> and op:=. The language has recently been extended to include
some of the pattern matching features of MCPL. This manual also describes the
standard library and running environment and the native code version of the
system based on Sial. Installation instructions are included. Since May 2013,
the standard BCPL distribution supports both 32 and 64 bit Cintcode versions.
Since August 2014, standard Cintcode BCPL includes floating point constants
and operators, and since March 2018 it includes the FLT feature to make it easier
to perform floating point calculations. Pattern matching was added in September
2021. These extensions are now in the standard BCPL distribution.
Keywords: Systems programming language, BCPL, Cintcode, Cintpos
2
Contents

Preface vii

1 The System Overview 1

1.1 A Cintsys Console Session . . . . . . . . . . . . . . . . . . . . . . 1
1.2 A Cintpos Console Session . . . . . . . . . . . . . . . . . . . . . . 7

2 The BCPL Language 13

2.1 Language Overview . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.1 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.2 The GET Directive . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.3 Conditional Compilation . . . . . . . . . . . . . . . . . . . 15
2.1.4 Section Brackets . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.1 Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.2 Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.3 Function Calls . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.4 Method Calls . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.5 Prefixed Expression Operators . . . . . . . . . . . . . . . . 21
2.2.6 Infixed Expression Operators . . . . . . . . . . . . . . . . 21
2.2.7 Boolean Evaluation . . . . . . . . . . . . . . . . . . . . . . 23
2.2.8 MATCH Expression . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.9 EVERY Expression . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.10 VALOF Expressions . . . . . . . . . . . . . . . . . . . . . . 24
2.2.11 Expression Precedence . . . . . . . . . . . . . . . . . . . . 24
2.2.12 Manifest Constant Expressions . . . . . . . . . . . . . . . 24
2.3 Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.1 Assignments . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.2 Routine Calls . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.3 Conditional Commands . . . . . . . . . . . . . . . . . . . . 26
2.3.4 Repetitive Commands . . . . . . . . . . . . . . . . . . . . 27
2.3.5 SWITCHON command . . . . . . . . . . . . . . . . . . . . . . 27
2.3.6 MATCH Command . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.7 EVERY Command . . . . . . . . . . . . . . . . . . . . . . . 28

i
ii CONTENTS

2.3.8 Flow of Control . . . . . . . . . . . . . . . . . . . . . . . . 28

2.3.9 Compound Commands . . . . . . . . . . . . . . . . . . . . 29
2.3.10 Blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Declarations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.1 Labels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.2 Manifest Declarations . . . . . . . . . . . . . . . . . . . . . 31
2.4.3 Global Declarations . . . . . . . . . . . . . . . . . . . . . . 31
2.4.4 Static Declarations . . . . . . . . . . . . . . . . . . . . . . 32
2.4.5 LET Declarations . . . . . . . . . . . . . . . . . . . . . . . 32
2.4.6 Dynamic Free Variables . . . . . . . . . . . . . . . . . . . 35
2.5 Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.6 Separate Compilation . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.7 The FLT Feature . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.8 The objline1 Feature . . . . . . . . . . . . . . . . . . . . . . . . 41

3 The Library 43
3.1 Manifest constants . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2 Global Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3 Global Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3.1 Streams . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.3.2 The Filing System . . . . . . . . . . . . . . . . . . . . . . 92
3.4 Random Access . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.5 RAM streams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.6 Environment Variables . . . . . . . . . . . . . . . . . . . . . . . . 95
3.7 Coroutine examples . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.7.1 A square wave generator . . . . . . . . . . . . . . . . . . . 97
3.7.2 Hamming’s Problem . . . . . . . . . . . . . . . . . . . . . 98
3.7.3 A Discrete Event Simulator . . . . . . . . . . . . . . . . . 100
3.8 The BMP Graphics Library . . . . . . . . . . . . . . . . . . . . . 106
3.8.1 The Graphics Functions . . . . . . . . . . . . . . . . . . . 107
3.9 The SDL Graphics Library . . . . . . . . . . . . . . . . . . . . . . 110
3.9.1 sdl.h details . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.9.2 Functions defined in sdl.b . . . . . . . . . . . . . . . . . . 112
3.9.3 sys(Sys sdl,...) calls . . . . . . . . . . . . . . . . . . . 117
3.10 The GL Graphics Library . . . . . . . . . . . . . . . . . . . . . . 120
3.11 The Sound Library . . . . . . . . . . . . . . . . . . . . . . . . . . 121
3.11.1 The Sound Constants . . . . . . . . . . . . . . . . . . . . . 121
3.11.2 The Sound Global Variables . . . . . . . . . . . . . . . . . 122
3.11.3 The Sound Functions . . . . . . . . . . . . . . . . . . . . . 122
3.12 The EXT Library . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
CONTENTS iii

4 The Command Language 123

4.1 Bootstrapping Cintsys . . . . . . . . . . . . . . . . . . . . . . . . 123
4.2 Bootstrapping Cintpos . . . . . . . . . . . . . . . . . . . . . . . . 125
4.2.1 The Cintpos BOOT module . . . . . . . . . . . . . . . . . 125
4.2.2 startroot . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
4.3 Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.4 cli.b and cli init.b . . . . . . . . . . . . . . . . . . . . . . . . 155

5 Console Input and Output 159

5.1 Cintsys console streams . . . . . . . . . . . . . . . . . . . . . . . . 159
5.2 Cintpos console streams . . . . . . . . . . . . . . . . . . . . . . . 160
5.2.1 Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5.2.2 Exclusive Input . . . . . . . . . . . . . . . . . . . . . . . . 161
5.2.3 Direct access to the screen and keyboard . . . . . . . . . . 162

6 Cintpos Devices 163

6.0.1 The Clock Device . . . . . . . . . . . . . . . . . . . . . . . 163
6.0.2 The Keyboard Device . . . . . . . . . . . . . . . . . . . . 164
6.0.3 The Screen Device . . . . . . . . . . . . . . . . . . . . . . 164
6.0.4 TCP/IP Devices . . . . . . . . . . . . . . . . . . . . . . . 164

7 The Debugger 167

7.1 The Cintsys Debugger . . . . . . . . . . . . . . . . . . . . . . . . 167
7.2 The Cintpos Debugger . . . . . . . . . . . . . . . . . . . . . . . . 171
7.3 Debugging Techniques . . . . . . . . . . . . . . . . . . . . . . . . 172

8 The Design of OCODE 179

8.1 Representation of OCODE . . . . . . . . . . . . . . . . . . . . . . 179
8.2 The OCODE Abstract Machine . . . . . . . . . . . . . . . . . . . 180
8.3 Loading and Storing values . . . . . . . . . . . . . . . . . . . . . . 181
8.4 Field Selection Operators . . . . . . . . . . . . . . . . . . . . . . . 182
8.5 Expression Operators . . . . . . . . . . . . . . . . . . . . . . . . . 183
8.6 Functions and Routines . . . . . . . . . . . . . . . . . . . . . . . . 184
8.7 Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
8.8 Directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
8.9 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

9 The Design of Cintcode 189

9.1 Designing for Compactness . . . . . . . . . . . . . . . . . . . . . . 190
9.1.1 Global Variables . . . . . . . . . . . . . . . . . . . . . . . 191
9.1.2 Composite Instructions . . . . . . . . . . . . . . . . . . . . 192
9.1.3 Relative Addressing . . . . . . . . . . . . . . . . . . . . . . 192
9.2 The Cintcode Instruction Set . . . . . . . . . . . . . . . . . . . . 193
iv CONTENTS

9.2.1 Byte Ordering and Alignment . . . . . . . . . . . . . . . . 193

9.2.2 Loading Values . . . . . . . . . . . . . . . . . . . . . . . . 195
9.2.3 Indirect Load . . . . . . . . . . . . . . . . . . . . . . . . . 196
9.2.4 Expression Operators . . . . . . . . . . . . . . . . . . . . . 196
9.2.5 Simple Assignment . . . . . . . . . . . . . . . . . . . . . . 197
9.2.6 Indirect Assignment . . . . . . . . . . . . . . . . . . . . . 198
9.2.7 Function and Routine Calls . . . . . . . . . . . . . . . . . 198
9.2.8 Flow of Control and Relations . . . . . . . . . . . . . . . . 200
9.2.9 Switch Instructions . . . . . . . . . . . . . . . . . . . . . . 200
9.2.10 Miscellaneous . . . . . . . . . . . . . . . . . . . . . . . . . 201
9.2.11 Floating-point Instructions . . . . . . . . . . . . . . . . . . 203
9.2.12 Select Instructions . . . . . . . . . . . . . . . . . . . . . . 203
9.2.13 Undefined Instructions . . . . . . . . . . . . . . . . . . . . 203
9.2.14 Corruption of B . . . . . . . . . . . . . . . . . . . . . . . . 204
9.2.15 Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 204

10 The Design of Sial 205

10.1 The Sial Specification . . . . . . . . . . . . . . . . . . . . . . . . . 207
10.2 Compaction of Sial . . . . . . . . . . . . . . . . . . . . . . . . . . 219

11 The MC Package 221

11.1 MC Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
11.2 MC Library Functions . . . . . . . . . . . . . . . . . . . . . . . . 225
11.3 The MC Language . . . . . . . . . . . . . . . . . . . . . . . . . . 226
11.4 MC Debugging Aids . . . . . . . . . . . . . . . . . . . . . . . . . 234
11.5 The n-queens Demonstration . . . . . . . . . . . . . . . . . . . . . 234

12 Installation 239
12.1 Linux Installation . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
12.2 Command Line Arguments . . . . . . . . . . . . . . . . . . . . . . 243
12.3 Installation on Other Machines . . . . . . . . . . . . . . . . . . . 243
12.4 Installation for Windows XP . . . . . . . . . . . . . . . . . . . . . 243
12.5 Installation using Cygwin . . . . . . . . . . . . . . . . . . . . . . 244
12.6 Installation for Windows CE2.0 . . . . . . . . . . . . . . . . . . . 245
12.7 The Native Code Version . . . . . . . . . . . . . . . . . . . . . . . 245

13 Example Programs 247

13.1 Coins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
13.2 Primes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
13.3 Queens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
13.4 Fridays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
13.5 Lambda Evaluator . . . . . . . . . . . . . . . . . . . . . . . . . . 250
13.6 Fast Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . 254
CONTENTS v

Bibliography 257

A BCPL Syntax Diagrams 259

vi CONTENTS
Preface

The concept for BCPL originated in 1966 and was first outlined in my PhD
thesis [4]. Its was first implemented early in 1967 when I was working at M.I.T.
Its heyday was perhaps from the mid 70s to the mid 80s, but even now it is still
continues to be used at some universities, in industry and by private individuals.
It is a useful language for experimenting with algorithms and for research in
optimizing compilers. Cintpos is the multi-tasking version of the system based
on the Tripos [5]. It is simple and easy to maintain and can be used for real-time
applications such as process control. BCPL was designed many years ago but
is still useful in areas where small size, simplicity and portability are important.
Recently I have decided to augment BCPL with some of the features of MCPL
including particularly the pattern matching mechanism used in the definition of
functions.
This document is intended to provide a record of the main features of the
BCPL in sufficient depth to allow a serious reader to obtain a proper understand-
ing of philosophy behind the language. An efficient interpretive implementation
is presented, the source of which is freely available via my home page [3]. The
implementation is machine independent and should be easy to transfer to almost
any architecture both now and in the future.
The main topics covered by this report are:
ˆ A specification of the BCPL language.
ˆ A description of its runtime library and the extensions used in the Cintpos
system.
ˆ The design and implementation of command language interpreters for both
the single and multi-threaded versions of the system.
ˆ A description of OCODE, the intermediate code used in the compiler, and
Cintcode, the compact byte stream target code used by the interpreter.
ˆ A description of the single and multi-threaded interactive debugger and
other debugging aids.
ˆ The efficient implementation of the Cintcode interpreter for several proces-
sors including both RISC and i386/Pentium based machines.

vii
viii CONTENTS

ˆ The profiling and statistics gathering facilities offered by the system.

ˆ The SIAL intermediate code that allows easy translation of BCPL in native
code for most architectures, including, for instance, the Raspberry Pi.

ˆ The MC package that allows machine independent dynamic compilation

and execution of native machine code.

For many example BCPL programs see bcpl4raspi.pdf available from my

home page.

MR
Chapter 1

The System Overview

This document contains a full description of an interpretive implementation of

BCPL that supports a command language and low level interactive debugger. As
an introduction, two example console sessions are presented to exhibit some of
the key features of both the single threaded version of the system (Cintsys) and
the interpretive version of Tripos (Cintpos).

1.1 A Cintsys Console Session

The BCPL Cintcode system can be entered using the cintsys shell command
under the host operating system. If cintsys is called with the -h option it will
output the following information about other possible options.

Valid arguments:

-h Output this help information

-m n Set Cintcode memory size to n words
-t n Set Tally vector size to n words
-g n Set the default global vector upb to n
-c args Pass args to interpreter as CLI input
-- args Pass args to interpreter as CLI input,
then re-attach stdin
-s file args Invoke the interpreter with this file as CLI input
-cin name Set the pathvar environment variable name
-f Trace use of environment variables in pathinput
-v Trace the bootstrapping process
-vv As -v, but include some Cincode level tracing
-d Cause a dump of the Cintcode memory to DUMP.mem
if a fault/error is encountered
-slow Force the slow interpreter to always be selected

1
2 CHAPTER 1. THE SYSTEM OVERVIEW

The Cintsys system is normally started using the cintsys shell command. If
this demonstration was run on a machine called c223 in the directory cintcode,
its opening message was as follows.
c223$ cintsys
BCPL 32-bit Cintcode System (6 Sept 2019)
0.000>

The characters 0.000> are followed by a space character and is the command
language prompt string inviting the user to type a command. The number gives
the execution time in seconds of the preceeding command. A program called
fact.b in directory cintcode/com to compute factorials can be displayed using
the type command as follows:
0.000> type com/fact.b
GET "libhdr"
LET start() = VALOF
{ FOR i = 1 TO 5 DO writef("fact(%n) = %i4*n", i, fact(i))
RESULTIS 0
}
AND fact(n) = n=0 -> 1, n*fact(n-1)
0.000>

The directive GET "libhdr" causes the standard library declarations to be

inserted at that position. The text:
LET start() = VALOF

is the heading for the declaration of the function start which, by convention, is
the first function to be called when a program is run. The empty parentheses ()
indicate that the function expects no arguments. The text

FOR i = 1 TO 5 DO

introduces a for-loop whose control variable i successively takes the values from
1 to 5. The body of the for-loop is a call of the library function writef whose
effect is to output the format string after replacing the substitution items %n
and %i4 by appropriately formatted representations of i and fact(i). Within
the string *n represents the newline character. The statement RESULTIS 0 exits
from the VALOF construct providing the result of start that indicates the program
completed successfully. The text:

AND fact(n) =
1.1. A CINTSYS CONSOLE SESSION 3

introduces the definition of the function fact which take one argument (n) and
yields n factorial. The word AND causes fact to available to the previously defined
function. This program can be compiled by using the following command:
0.000> bcpl com/fact.b to fact
BCPL (3 Sep 2019) 32 bit with the FLT feature
Code size = 104 bytes 0f 32-bit little ender Cintcode
0.030>

This command compiles the source file fact.b creating an executable object
module in the file called fact. The program can then be run by simply typing
the name of this file.
0.030> fact
fact(1) = 1
fact(2) = 2
fact(3) = 6
fact(4) = 24
fact(5) = 120
0.011>

When the BCPL compiler is invoked, it can be given additional arguments

that control the compiler options. One of these (d1) directs the compiler to
output the compiled code in a readable form, as follows:
0.010> bcpl com/fact.b to fact d1
BCPL (3 Sep 2019) 32 bit with the FLT feature
0: DATAW 0x00000000
4: DATAW 0x0000DFDF
8: DATAW 0x6174730B
12: DATAW 0x20207472
16: DATAW 0x20202020
// Entry to: start
20: L1:
20: L1
21: SP3
22: L4:
22: LP3
23: LF L2
25: K9
26: SP9
27: LP3
28: SP8
29: LLL L9920
31: K4G 94
33: L1
34: AP3
35: SP3
36: L5
37: JLE L4
39: L0
4 CHAPTER 1. THE SYSTEM OVERVIEW

40: RTN
44: L9920:
44: DATAW 0x6361660F
48: DATAW 0x6E252874
52: DATAW 0x203D2029
56: DATAW 0x0A346925
60: DATAW 0x0000DFDF
64: DATAW 0x6361660B
68: DATAW 0x20202074
72: DATAW 0x20202020
// Entry to: fact
76: L2:
76: JNE0 L5
78: L1
79: RTN
80: L5:
80: LM1
81: AP3
82: LF L2
84: K4
85: LP3
86: MUL
87: RTN
88: L3:
88: DATAW 0x00000000
92: DATAW 0x00000001
96: DATAW 0x00000014
100: DATAW 0x0000005E
Code size = 104 bytes 0f 32-bit little ender Cintcode
0.072>

This output shows the sequence of Cintcode instructions compiled for the func-
tions start and fact. In addition there are some data words holding the string
constant, initialisation data and symbolic information for the debugger. The
data word at location 4 holds a special bit pattern indicating the presence of a
function name placed just before the entry point. As can be seen the name in
this case is start. Similar information is packed at location 60 for the function
fact. Most Cintcode instructions occupy one byte and perform simple opera-
tions on the registers and memory of the Cintcode machine. For instance, the
first two instructions of start (L1 and SP3 at locations 20 and 21) load the
constant 1 into the Cintcode A register and then stores it at word 3 of the cur-
rent stack frame (pointed to by P). This corresponds to the initialisation of the
for-loop control variable i. The start of the for-loop body has label L4 corre-
sponding to location 22. The compilation of fact(i) is LP3 LF L2 K9 which
loads i and the entry address of fact and enters the function incrementing P
by 9 locations. The result of this function is returned in A which is stored in
the stack using SP9 in the appropriate position for the third argument of the
call of writef. The second argument, i, is setup using LP3 SP8, and the first
argument which is the format string is loaded by LLL L9920. The next instruc-
tion (K4G 94) causes the routine writef, whose entry point is in global variable
1.1. A CINTSYS CONSOLE SESSION 5

94, to be called incrementing P by 4 words as it does so. Thus the compilation

of the call writef("fact(%n) = %i5*n", i, f(i)) occupies just 11 bytes from
location 22 to 32, plus the 16 bytes at location 44 where the string is packed. The
next three instructions (L1 AP3 SP3) increment i, and L5 JNE L4 jumps to label
L4 if i is still less than 5. If the jump is not taken, control falls through to the
instructions L0 RTN causing start to return with result 0. Each instruction of
this function occupies one byte except for the LF, LLL, K4G and JNE instructions
which each occupy two. The body of the function fact is equally easy to under-
stand. It first tests whether its argument is zero (JNE0 L5). If it is, it returns
one (L1 RTN). Otherwise, it computes n-1 by loading -1 and adding n (LM1 AP3)
before calling fact (LF L2 K4). The result is then multiplied by n (LP3 MUL)
and returning (RTN). The space occupied by this code is just 12 bytes.
The debugger can be entered using the abort command.
0.030> abort
!! ABORT 99: User requested
*

The asterisk is the prompt inviting the user to enter a debugging command. The
debugger provides facilities for inspecting and changing memory as well as setting
breakpoints and performing single step execution. As an example, a breakpoint
is placed at the first instruction of the routine clihook which is used by the
command language interpreter (CLI) to transfer control to a command. Consider
the following commands:
* g4 b1
* b
1: clihook
*

This first loads the entry point of clihook (held in global variable 4) and sets
(b1) a breakpoint numbered 1 at this position. The command b, without an
argument, lists the current breakpoints confirming that the correct one has been
set. Normal execution is continued using the c command.
* c
0.010>

If we now try to execute the factorial program, we immediately hit the break-
point.
0.000> fact
!! BPT 1: clihook
A= 0 B= 0 17940: K4G 1
*
6 CHAPTER 1. THE SYSTEM OVERVIEW

This indicates that the breakpoint occurred when the Cintcode registers A and
B were both zero, and that the program counter is set to 17940 where the next
instruction to be obeyed is K4G 1. Single step exection can now be performed
using the \ command.
* \ A= 0 B= 0 46276: L1
* \ A= 1 B= 0 46277: SP3
* \ A= 1 B= 0 46278: LP3
*

After each single step execution a summary of the current state is printed. In the
above sequence we see that the execution of the instruction L1 loading 1 into the
A register. The execution of SP3 does not have an immediately observable effect
since it updates a local variable held in the current stack frame, but the stack
frame can be displayed using the t command.
* p t4
P 0: 46420 17942 start 1
*

This confirms that location P3 contains the value 1 corresponding to the initial
value of the for-loop control variable i. At this stage it is possible to change its
value to 3, say.
* 3 sp3
* p t4
P 0: 46420 17942 start 3
*

If single stepping is continued for a while we observe the evaluation of the

recursive call fact(3).
* \ A= 3 B= 1 46279: LF 46332
* \ A= fact B= 3 46281: K9
* \ A= 3 B= 3 46332: JNE0 46336
* \ A= 3 B= 3 46336: LM1
* \ A= -1 B= 3 46337: AP3
* \ A= 2 B= 3 46338: LF 46332
* \ A= fact B= 2 46340: K4
* \ A= 2 B= 2 46332: JNE0 46336
* \ A= 2 B= 2 46336: LM1
* \ A= -1 B= 2 46337: AP3
* \ A= 1 B= 2 46338: LF 46332
* \ A= fact B= 1 46340: K4
* \ A= 1 B= 1 46332: JNE0 46336
* \ A= 1 B= 1 46336: LM1
* \ A= -1 B= 1 46337: AP3
* \ A= 0 B= 1 46338: LF 46332
* \ A= fact B= 0 46340: K4
* \ A= 0 B= 0 46332: JNE0 46336
1.2. A CINTPOS CONSOLE SESSION 7

* \ A= 0 B= 0 46334: L1
* \ A= 1 B= 0 46335: RTN
* \ A= 1 B= 0 46341: LP3
* \ A= 1 B= 1 46342: MUL
* \ A= 1 B= 1 46343: RTN
* \ A= 1 B= 1 46341: LP3
* \ A= 2 B= 1 46342: MUL
* \ A= 2 B= 1 46343: RTN
* \ A= 2 B= 1 46341: LP3
* \ A= 3 B= 2 46342: MUL
* \ A= 6 B= 2 46343: RTN
* \ A= 6 B= 2 46282: SP9
* \ A= 6 B= 2 46283: LP3
* \ A= 3 B= 6 46284: SP8
* \ A= 3 B= 6 46285: LLL 46300
* \ A= 11575 B= 3 46287: K4G 94
*

At this moment the routine writef is just about to be entered to print an message
about factorial 3. We can unset breakpoint 1 and continue normal execution by
typing 0b1 c.
* 0b1 c
fact(3) = 6
fact(4) = 24
fact(5) = 120
0.010>

As one final example in this session we will re-compile the BCPL compiler.
0.010> bcpl com/bcpl.b to junk
BCPL (3 Sep 2019) 32 bit with the FLT feature
Code size = 32044 bytes of 32-bit little ender Cintcode
Code size = 14968 bytes of 32-bit little ender Cintcode
0.478>

This shows that the total size of the compiler is 47,012 bytes and that it can
be compiled (on a 1.6GHz CPU) in 0.478 seconds. Since this involves executing
41,273,342 Cintcode instructions, the rate is about 86 million Cintcode instruc-
tions per second with the current interpreter.

1.2 A Cintpos Console Session

When the Cintpos system is started (on a machine called meopham) in the di-
rectory Cintpos/cintpos, its opening message is as follows:
meopham$ cintpos
Cintpos System (09 Mar 2010)
0.000 1>
8 CHAPTER 1. THE SYSTEM OVERVIEW

There is a directory called com that holds the BCPL source code of several
Cintpos commands, such as bcpl.b, bench100.b and fact.b. We can inspect
fact.b using the type command as follows.
0.000 1> type com/fact.b
SECTION "fact"
GET "libhdr"
LET f(n) = n=0 -> 1, n*f(n-1)
LET start() = VALOF
{ FOR i = 1 TO 10 DO
writef("f(%i2) = %i8*n", i, f(i))
RESULTIS 0
}
0.000 1>
It can be compiled and run as follows.
0.000 1> c bc fact
bcpl com/fact.b to cin/fact hdrs POSHDRS
BCPL (20 Oct 2009)
Code size = 120 bytes
0.020 1> fact
f( 1) = 1
f( 2) = 2
f( 3) = 6
f( 4) = 24
f( 5) = 120
f( 6) = 720
f( 7) = 5040
f( 8) = 40320
f( 9) = 362880
f(10) = 3628800
0.000 1>
There is a benchmark program called bench100.b which can be compiled and
run as follows.
0.000 1> c bc bench100
bcpl com/bench100.b to cin/bench100 hdrs POSHDRS
BCPL (20 Oct 2009)
Code size = 1444 bytes
0.040 1> bench100
bench mark starting, Count=1000000
starting
finished
qpkt count = 2326410 holdcount = 930563
these results are correct
end of run
9.170 1>
1.2. A CINTPOS CONSOLE SESSION 9

The latest prompt (9.170 1>) indicates that the benchmark program took 9.17
seconds to run and that we are connected to the root command language inter-
preter running as task one.
When Cintpos starts these are six resident tasks which can be seen using the
status command as follows.
0.000 1> status
Task 1: Root_Cli running CLI Loaded command: status
Task 2: Debug_Task waiting DEBUG
Task 3: Console_Handler waiting COHAND
Task 4: File_Handler waiting FH0
Task 5: MBX_Handler waiting MBXHAND
Task 6: TCP_Handler waiting TCPHAND
0.010 1>

Task 2 is an interactive debugging aid, task 3 handles communication between

tasks and the keyboard and display devices, task 4 handles communication be-
tween tasks and the filing system, task 5 provides a mailbox facility that allows
communication of short text messages between tasks and, finally, task 6 handles
TCP/IP communication between tasks and the internet.
Tasks may be dynamically created and destoyed. For instance, the run com-
mand will create a new CLI task giving it a command to run.
0.010 1> run status
0.000 1> Task 1: Root_Cli waiting CLI No command loaded
Task 2: Debug_Task waiting DEBUG
Task 3: Console_Handler waiting COHAND
Task 4: File_Handler waiting FH0
Task 5: MBX_Handler waiting MBXHAND
Task 6: TCP_Handler waiting TCPHAND
Task 7: Run_Cli running CLI Loaded command: status

Notice that the root CLI (task 1) completes the execution of the run command
and issues a prompt (0.000 1>) before the newly created CLI (task 7) has had
time to load and run the status command. As soon as task 7 finishes running
the status command it commits suicide leaving the original 6 tasks.
The bounce.b program provides a demonstration of how communication be-
tween Cintpos tasks works.
0.000 1> type com/bounce.b
SECTION "bounce"
GET "libhdr"
LET start() BE qpkt(taskwait()) REPEAT
0.000 1>

It can be compiled and run as follows.

10 CHAPTER 1. THE SYSTEM OVERVIEW

0.000 1> c bc bounce

bcpl com/bounce.b to cin/bounce hdrs POSHDRS
BCPL (20 Oct 2009)
Code size = 60 bytes
0.010 1> run bounce
0.000 1> status
Task 1: Root_Cli running CLI Loaded command: status
Task 2: Debug_Task waiting DEBUG
Task 3: Console_Handler waiting COHAND
Task 4: File_Handler waiting FH0
Task 5: MBX_Handler waiting MBXHAND
Task 6: TCP_Handler waiting TCPHAND
Task 7: Run_Cli waiting CLI Loaded command: bounce
0.000 1>

The status output shows that the bounce program is running as task 7 and is
suspended in taskwait waiting for another task to send it a packet. When it
receives a packet it immediately returns it to the sender and waits for another to
arrive. We can send a suitable packet to bounce using the send command whose
source code is as follows.
0.000 1> type com/send.b
SECTION "send"
GET "libhdr"
GLOBAL { task: 200; count: 201 }
LET start() BE
{ LET pkt = VEC 2
LET argv = VEC 50
UNLESS rdargs("TASK/n,COUNT/n", argv, 50) DO
{ writef("Bad arguments for SEND*n")
stop(20)
}
task, count := 7, 1_000_000
IF argv!0 DO task := !argv!0
IF argv!1 DO count := !argv!1
pkt!0, pkt!1, pkt!2 := notinuse, task, count
writef("*nSending a packet to task %n, %n times*n", task, count)
{ LET k = pkt!2
UNLESS k BREAK
pkt!2 := k-1
qpkt(pkt)
pkt := taskwait()
} REPEAT
writes("Done*n")
}
0.010 1>
1.2. A CINTPOS CONSOLE SESSION 11

This program creates a packet consisting of a vector (one dimensional array) of

three elements. The first is used by the system for chaining packets together
and must be initialised the the special value notinuse. The next element of the
packet (pkt!1) holds the destination task number and the final element (pkt!2)
holds a value (initially 1000000) which is going to be used as a counter. The
REPEAT loop decrements this counter field and sends the packet using qpkt to
the bounce task suspending itself in taskwait until the packet returns. Control
leaves the REPEAT loop when the counter reaches zero, causing send to output
the message Done. We can compile and run send as follows.
0.010 1> c bc send
bcpl com/send.b to cin/send hdrs POSHDRS
BCPL (20 Oct 2009)
Code size = 252 bytes
0.020 1> send
Sending a packet to task 7, 1000000 times
Done
3.940 1>

This demonstration shows that a packet may be sent from one task to another
2 million times in 3.94 seconds. This corresponds to a rate of just over half a
million times per second.
12 CHAPTER 1. THE SYSTEM OVERVIEW
Chapter 2

The BCPL Language

The design of BCPL owes much to the work done on CPL (originally Cambridge
Programming Language) which was conceived at Cambridge to be the main lan-
guage to run on the new and powerful Ferranti Atlas computer to be installed
in 1963. At that time there was another Atlas computer in London and it was
decided to make the development of CPL a joint project between the two Uni-
versities. As a result the name changed to Combined Programming Language. It
could reasonably be called Christopher’s Programming Language in recognition
of Christpher Strachey whose bubbling enthusiasm and talent steered the course
of its development.
CPL was an ambitious language in the ALGOL tradition but with many novel
and significant extensions intended to make its area of application more general.
These included a greater richness in control constructs such as the now well known
IF, UNLESS, WHILE, UNTIL, REPEATWHILE, SWITCHON statements. It could handle
a wide variety of data types including string and bit patterns and was one of the
first strictly typed languages to provided a structure mechanism that permitted
convenient handling of lists, trees and directed graphs. Work on CPL ran from
about 1961 to 1967, but was hampered by a number of factors that eventually
killed it. It was, for instance, too large and complicated for the machines available
at the time, and the desire for elegance and mathematical cleanliness outweighed
the more pragmatic arguments for efficiency and implementability. Much of the
implementation was done by research students who came and left during the
lifetime of the project. As soon as they knew enough to be useful they had
to transfer their attention to writing theses. Another problem (that became of
particular interest to me) was that the implementation at Cambridge had to
move from EDSAC II to the Atlas computer about halfway through the project.
The CPL compiler thus needed to be portable. This was achieved by writing it
in a simple subset of CPL which was then hand translated into a sequence of
low level macro calls that could be expanded into the assembly language of either
machine. The macrogenerator used was GPM[6] designed by Strachey specifically
for this task. It was a delightfully elegant work of art in its own right it is well

13
14 CHAPTER 2. THE BCPL LANGUAGE

worth study. A variant of GPM, called BGPM, is included in the standard BCPL
distribution.
BCPL was initially similar to this subset of CPL used in the encoding of
the CPL compiler. An outline of BCPL’s main features first appeared in my
PhD thesis [4] in 1966 but it was not fully designed and implemented until early
the following year when I was working at Project MAC of the Massachussetts
Institute of Technology. Its first implementation was written in Ross’s Algol
Extended for Design (AED-0)[1] which was the only language then available on
CTSS, the time sharing system at Project MAC, other than LISP that allowed
recursion.

2.1 Language Overview

A BCPL program is made up of separately compiled sections, each consisting of a
list of declarations that define the constants, static data and functions belonging
to the section. Within functions it is possible to declare dynamic variables and
vectors that exist only as long as they are required. The language is designed so
that these dynamic quantities can be allocated space on a runtime stack. The
addressing of these quantities is relative to the base of the stack frame belonging
to the current function activation. For this to be efficient, dynamic vectors have
sizes that are known at compile time. Functions may be called recursively and
their arguments are called by value. The effect of call by reference can be achieved
by passing pointers. Input and output and other system operations are provided
by means of library functions.
The main syntactic components of BCPL are: expressions, commands, and
declarations. These are described in the next few sections. In general, the pur-
pose of an expression is to compute a value, while the purpose of a command is
normally to change the value of one or more variables or to perform input/output.

2.1.1 Comments
There are two form of comments. One starts with the symbol // and extends
up to but not including the end-of-line character, and the other starts with the
symbol /* and ends at a matching occurrence of */. Comment brackets (/* and
*/ may be nested, and within such a comments the lexical analyser is only looking
for /* and */ and so special care is needed when commenting out fragments of
program containing // comments and string constants. Comments are equivalent
to white space and so may not occur within of multi-character symbols such as
identifiers or constants.
2.1. LANGUAGE OVERVIEW 15

2.1.2 The GET Directive

A directives of the form GET "filename" is replaced by the contents of the named
file. Early versions of the compiler only inserted the file up to the first occurring
dot but now the entire file is inserted. By convention, GET directives normally
appear on separate lines. If the filename does not end in .h or .b the extension
.h is added.
The name is looked up by first searching the current directory and then
the directories specified by the environment variable whose name is held in the
rtn hdrsvar of the rootnode, but this can be overridden using the hdrs com-
piler option. The default environment variable for BCPL headers is BCPLHDRS
under Cintsys and POSHDRS under Cintpos. Header files are normally in the g/
directory in the root directory of the current system. To check whether the envi-
ronment variables are set correctly, enter cintsys or cintpos with the -f option
as suggested in Section 3.6.

2.1.3 Conditional Compilation

A simple mechanism, whose implementation takes fewer than 20 lines of code
in the lexical analyser allows conditional skipping of lexical symbols. It uses
directives of the following form:

$$tag
$<tag
$~tag
$>tag

where tag is conditional compilation tag composed of letters, digits, dots and
underlines. All tags are initially unset, but may be complemented using the $$tag
directive. All the lexical tokens between $<tag and $>tag are skipped (treated as
comments) unless the specified tag is set. All the lexical tokens between $~tag
and $>tag are skipped unless the specified tag is not set.
The following example shows how this conditional compilation feature can be
used.
$$Linux // Set the Linux conditional compilation tag
$<Linux // Include if the Linux tag is set
$<WinXP $$WinXP $>WinXP // Unset the WinXP tag if set
writef("This was compiled for Linux")
$>Linux
$<WinXP // Include if the WinXP tag is set
writef("This was compiled for Windows XP")
$>WinXP
16 CHAPTER 2. THE BCPL LANGUAGE

2.1.4 Section Brackets

Historically BCPL used the symbols $( and $) to bracket commands and decla-
rations. These symbols are called section brackets and are allowed to be followed
by tags composed of letters, digits, dots and underlines. A tagged closing sec-
tion bracket is forced to match with its corresponding open section bracket by
the automatic insertion of extra closing brackets. Use of this mechanism is no
longer recommended since it often leads to obscure programming errors. BCPL
has been extended to allow all untagged section brackets to be replaced by { and
} as appropriate.

2.2 Expressions
Expressions are composed of names, constants and expression operators and may
be grouped using parentheses. The precedence and associativity of the different
expression constructs is given in Section 2.2.11. BCPL expressions always vield
values of the same bit length, normall 32 or 64 bits.

2.2.1 Names
Syntactically a name is of a sequence of letters, digits, dots and underlines starting
with a letter, but it must not one of the reserved words (such as IF, WHILE or
RESULTIS). The use of dots in names is no longer recommended, and should be
replaced by underscores. Double dots are no longer permitted in names because
.. is the range operator used in the pattern matching extension.
A name may be declared as a local variable, a static variable, a global variable,
a manifest constant, a label or a function or routine. Since the language is
typeless, the value of a name is just a bit pattern whose interpretation depends
on how it is used. This is similar to the way values in central registers of most
computers are used.

2.2.2 Constants
Decimal numbers consist of a sequence of digits, while binary, octal or hexadeci-
mal are represented, respectively, by #B, #O or #X followed by digits of the appro-
priate sort. Letters in hexadecimal numbers may use both upper and lower case
and the case of the letters B, O or X after #. The O may be omitted in octal num-
bers. Underlines may be inserted within numbers to improve their readability.
2.2. EXPRESSIONS 17

The following are examples of valid numbers:

1234
1_234_456
#B_1011_1100_0110
#o377
#X3fff
#x_DEADC0DE

Since August 2014, floating point constants are now allowed, such as the
following:
1234.0
1.234_456e-5
10e6

Note that 12.34 is a floating point number, but 12..34 is 12 followed by the
range operator .. and 34. A floating point constant must start with a digit and
contain a decimal point (.) or an exponent sign (e or E).
BCPL uses the standard IEEE representation for floating point numbers using
the same word length as other BCPL values For 32-bit BCPL the format is as
follows. The left most bit is the sign with 1 representing negative. The next 8
bits hold an unsigned number e in the range 0 to 255. e = 0 and e = 255 are used
to specify in the representation of some special values such as zero, infinity or
various error values. The values between 1 and 254 specify binary exponents in
the range -126 to +127 equal to e − 127. The remaining 23 bits are the fractional
bits of the significand. For non zero values, the significand has 24 bits with its
left most bit being 1 followed by these 23 fractional bits. This represents a value
greater than or equal to 1.0 and less than 2.0. Note that 1+8+23=32. The value
of the floating point number is the significand multiplied by 2e−127 . As a special
case, the number 0.0 is represented by a bit pattern of 32 zeroes.
For 64-bit numbers the exponent has 11 bits and the significand has 53. Note
that 1+11+52=64.
The compiler does not use any floating point operators or constants using,
where necessary, calls of the form sys(Sys flt,...) to perform any floating
point calculations needed. This allows the compiler to be compiled using older
versions of the compiler. Floating point constants are currently only compiled
correctly if the BCPL word length of the compiler is the same as that of the
target code.
TRUE and FALSE are reserved words that have values -1 and 0, respectively,
representing the two truth values. They can be used in manifest constant ex-
pressions. Whenever a boolean test is made, the value is compared with with
FALSE (=0). BITSPERBCPLWORD is also a reserved word whose value is 32 or 64
giving the BCPL word length currently being used. This constant was added on
16 May 2013 to allow the same header file to be used on both 32- and 64-bit
BCPL systems. It is used in the MANIFEST declarations of constants such as
18 CHAPTER 2. THE BCPL LANGUAGE

bytesperword and minint that are word length dependent. If you are using an
older BCPL compiler with the latest version of libhdr.h you will need to un-
comment a line that declares BITSPERBCPLWORD as a MANIFEST constant with
the appropriate value for the system you are using.
The commands BREAK, LOOP, NEXT, EXIT and ENDCASE are permitted within
expressions and have the same effect as the corresponding commands described
in Section 2.3.8.
A question mark (?) may be used as a constant with undefined value. It can
also be used in statements such as:
LET a, b, count = ?, ?, 0
sendpkt(notinuse, rdtask, ?, ?, Read, buf, size)

Constants of the form: SLCT len:shift:offset pack the three constants len,
shift and offset into a word. Such packed constants are used by the field selection
operator OF to access fields of given length, shift and offset relative to a pointer
as described in Section 2.2.6. The len and shift components are optional. Their
omission has the following effect.

SLCT shift:offset means SLCT 0:shift:offset

SLCT offset means SLCT 0:0:offset

Character constants consist of a single character enclosed in single quotes

(’). Character constants behave like integers typically in the range 0 to 255
corresponding to its normal ASCII encoding, but can be larger using unicode
characters as describer below.
Character (and string) constants may use the following escape sequences.
2.2. EXPRESSIONS 19

Escape Replacement
*n A newline (end-of-line) character.
*c A carriage return character.
*p A newpage (form-feed) character.
*s A space character.
*b A backspace character.
*t A tab character.
*e An escape character.
*" "
*’ ’
** *
*xhh The single character with number hh (two hexadecimal
digits denoting an integer in the range [0,255]).
*ddd The single character with number ddd (three octal digits
denoting an integer in the range [0,255]).
*#g Set the encoding mode to GB2312 for the rest of this
string or character constant. The default encoding is
UTF8 unless speified by the GB2312 compiler option,
See the specification of the bcpl command on page 131.
*#u Set the encoding mode explicitly to UTF8 for the rest
of this string or character constant.
*#hhhh In UTF8 mode, this specifies a single Unicode character
with up to four hexadecimal digits. In string constants,
this is converted to a sequence of bytes giving its UTF-
8 representation. In character constants, it yields the
integer hhhh. Thus ’*#C13F’=#xC13F.
*##h..h In UTF8 mode, this specifies a Unicode character with
up to eight hexadecimal digits, but is otherwise treated
as the *#hhhh escape.
*#dddd In GB2312 mode, this specifies the GB2312 decimal code
(dddd) for an extended character. In string constants,
this is converted to a sequence of bytes giving its GB2312
representation. In character constants, it yields the in-
teger dddd. Thus ’*#g*#4566’=4566.
*f..f * This sequence is ignored, where f..f stands for a se-
quence of white space characters. In this context, com-
ments introduced by ’//’ are treated as white space,
but those introduced by ’/*’ are not.

A string constant consists of a sequence of zero or more characters enclosed

within quotes ("). Both string and character constants use the same character
escape mechanism described above. The value of a string is a pointer where the
20 CHAPTER 2. THE BCPL LANGUAGE

length and bytes of the string are packed. If s is a string then s%0 is its length
and s%1 is its first character, see Section 2.2.6. The *# escapes allow Unicode
and GB2312 characters to be handled. For instance, if the following statements
output to a suitable UTF8 configured device:
writef("*#uUnicode hex 2200 prints as: ’*#2200’*n")
writef("%%# in writef can also be used: ’%#’*n", #x2200)

the result is as follows

Unicode hex 2200 prints as: ’∀’ %# in writef can also be used: ’∀’
A static vector can be created using an expression of the following form:
TABLE K0 , . . . , Kn where K0 , . . . , Kn are manifest constant expressions, see Sec-
tion 2.2.12. The space for a static vector is allocated for the life time of the
program and its elements are updateable.

2.2.3 Function Calls

Syntactically, a function call is an expression followed by a, possibly empty, ar-
gument list enclosed in paretheses as in the following examples.
newline()
mk3(Mult, x, y)
writef("f(%n) = %n*n", i, f(i))
f(1,2,3)
(fntab!i)(p, @a)

The parentheses are required even if no arguments are given. The last example
above illustrates a call in which the function is specified by an expression. If
the function being called was defined by a routine definition, the result of the
call will be undefined. The arguments are evaluated and laid out in consective
stacklocations where they become the initial values of the formal parameters of
the called function or routine. There is no need for the number of arguments
to be the same as the number of formal parameters. See Section 2.4.5 for more
details.

2.2.4 Method Calls

Method calls are designed to make an object oriented style of programming more
convenient. They are syntactically similar to a function calls but uses a hash
symbol (#) to separate the function specifier from its arguments. The expression:

E#(E1 ,..,En )

is defined to be equivalent to:

2.2. EXPRESSIONS 21

(E1 !0!E)(E1 ,..,En )

Here, E1 points to the fields of an object, with the convention that its zeroth
field (E1 !0) is a pointer to the methods vector containing the possible functions
to call. Element E of this vector is applied to the given set of arguments. E is
normally a manifest constant. An example program illustrating method calls can
be found in BCPL/bcplprogs/demos/objdemo.b in the BCPL distribution.

2.2.5 Prefixed Expression Operators

An expression of the form !E returns the contents of the memory word pointed
to by the value of E.
An expression of the form @E returns a pointer to the BCPL word sized
location specified by E. E can only be a variable name or an expression with
leading operator !. Pointers to consecutive locations are consecutive integers.
Expressions of the form: +E, -E, ABS E, ~E and NOT E return the result
of applying the given prefixed operator to the value of the expression E. The
operator + returns the value unchanged, - returns the integer negation, ABS
returns the absolute value, ~ return the bitwise complement of the value.
FLOAT E converts the integer E to its corresponding floating point value. FIX
E converts the floating point value E to its closest integer representation. #ABS E
returns the absolute value of the floating point number E, and #+ and #- perform
monadic plus and minus on floating point values.

2.2.6 Infixed Expression Operators

An expression of the form E1 !E2 evaluates E1 and E2 to yield respectively a
pointer, p say, and an integer, n say. The value returned is the contents of the
nth word relative to p. Since words of memory have consecutive integer addresses,
the expression E1 !E2 is exactly equivalent to !(E1 +E2 ).
An expression of the form E1 [E2 ] has recently been added. It is syntactically
like a function call but is equivalent to E1 !E2 .
An expression of the form E1 %E2 evaluates E1 and E2 to yield a pointer, p
say, and an integer, n say. The expression returns the unsigned byte at position
n relative to p.
An expression of the form K OF E accesses a field of consecutive bits
in memory. K must be a manifest constant (see section 2.2.12) equal to
SLCT length:shift:offset and E must yield a pointer, p say. The field is con-
tained entirely in the word at position p+offset. It has a bit length of length and
is shift bits from the right hand end of the word. A length of zero is interpreted
as the longest length possible consitent with shift and the word length of the
implementation.
22 CHAPTER 2. THE BCPL LANGUAGE

An :: expression can be used on both right and left hand sides of assignments
but not as the operand of @. When used in a right hand context the selected field
is shifted to the right hand end of the result with vacated positions filled with
zeros. A shift to the left is performed when a field is updated. Suppose p!3
holds #x12345678, the expression (SLCT 12:8:3)::p yields #x456 and after the
assignment:

(SLCT 12:8:3)::p := #xABC

the value of p!3 will be #x123ABC78.

OF can be used a synonym of ::.
An expression of the form E1 <<E2 (or E1 >>E2 ) evaluates E1 and E2 to yield a
bit pattern, w say, and an integer, n say, and returns the result of shifting w to the
left (or right) by n bit positions. Vacated positions are filled with zeroes. Shifts of
the word length or more return 0, and negative shifts return undefined typically
zero results, although on some versions of BCPL they reverse the direction of the
shift.
Expressions of the form: E1 *E2 , E1 /E2 , E1 MOD E2 , E1 +E2 , E1 -E2 . E1 EQV E2
and E1 XOR E2 return the result of applying the given operator to the two
operands. The operators are, respectively, integer multiplication, integer divi-
sion, remainder after integer division, integer addition, integer subtraction, bit-
wise equivalent and bitwise not equivalent (exclusive OR).
Expressions of the form: E1 &E2 and E1 |E2 return, respectively, the bitwise
AND or OR of their operands unless the expression is being evaluated in a boolean
context such as the condition in a while command, in which case the operands
are tested from from left to right until the value of the condition is known.
An expression of the form: E relop E relop . . . relop E where each relop is one
of =, ~=, <=, >=, < or > returns TRUE if all the individual relations are satisfied and
FALSE, otherwise. The operands are evaluated from left to right, and evaluation
stops as soon as the result can be determined. Operands may be evaluated more
than once, so don’t try ’0’<=rdch()<=’9’.
An expression of the form: E1 ->E2 ,E3 evaluates E1 in a boolean context,
and, if this yields FALSE, it returns the value of E3 , otherwise it returns the value
of E2 .
The floating point operators #*, #/, #+, #-, #=, \#~=, #<, #>, #<=, #>= and #->
have recently been added to standard BCPL. They have the same binding power
as the corresponding integer operators. Beware that, with older versions of the
BCPL compiler that do not implement the FLT feature, it is easy make mistakes
such as -1.2 which performs integer negation of the bit patterns representing
1.2. The expression should have been written #-1.2. With the FLT feature -
would have been automatically replaced by #- in this situation. See Section 2.7
for details.
2.2. EXPRESSIONS 23

2.2.7 Boolean Evaluation

Expressions used to control the flow of execution in coditional constructs, as in
IF and WHILE commands, are evaluated in a Boolean context. This effects the
treatment of the operators ~ & and | whose operands are evaluated in Boolean
contexts. In a Boolean context, the operands of & and | are evaluated from left
to right until the value of the condition is known, and ~ negates the condition.

2.2.8 MATCH Expression

A MATCH expression has the following form:
MATCH ( args )
: P ,.., P => E
...
: P ,.., P => E
.
It consists of the word MATCH followed by a list of arguments enclosed in parenthe-
ses, followed by a squence of one or more match items terminated by an optional
dot. The match items are applied to the arguments as described in Section 2.5,
yielding the value of the expression in the first match item to be satisfied. If all
match item fail the result is zero.
Within a match item the command NEXT causes control to pass to the next
match item, and EXIT causes the MATCH expression to terminate yielding the value
zero.

2.2.9 EVERY Expression

A EVERY expression has the following form:
EVERY ( args )
: P ,.., P => E
...
: P ,.., P => E
.
It consists of the word EVERY followed by a list of arguments enclosed in
parentheses, followed by a squence of one or more match items terminated by
an optional dot. The match items are applied to the arguments as described in
Section 2.5, yielding the integer sum of the values of the expressions of successful
match items. If none match the result is zero.
Within a match item the command NEXT causes control to pass to the next
match item, and EXIT causes the EVERY expression to terminate yielding the sum
accumulated so far.
24 CHAPTER 2. THE BCPL LANGUAGE

2.2.10 VALOF Expressions

An expression of the form VALOF C, where C is a command, is evaluated by
executing the command C. On encountering a command of the form RESULTIS E
within C, execution terminates, returning the value of E as the result. VALOF
expressions are often used as the bodies of functions.

2.2.11 Expression Precedence

So that the separator semicolon (;) can be omitted at the end of lines, there is
the restriction that infixed operators may not occur as the first token of a line. If
the first token on a line is !, + or -, these must be regarded as prefixed operators.
The syntax of BCPL is specified by the diagrams in Appendix A, but a sum-
many of the precendence of expression operators is given in table 2.1. The prece-
dence values are in the range 0 to 9, with the higher values signifying greater
binding power. The letters L and R denote the associativity of the operators.
For instance, the dyadic operator - is left associative and so a-b-c is equiva-
lent to (v-i)-j, while a->x,b->y,z is right associative and so is equivalent to
a->x,(b->y,z).
Notice that these precedence values imply that

! f x means ! (f x)
! @ x means ! (@ x)
! v ! i ! j means ! ((v!i)!j)
@ v ! i ! j means @ ((v!i)!j)
x << 1+y >> 1 means (x<<(1+y))>>1)
~ x!y means ~ (x!y)
~ x=y means ~ (x=y)
NOT x=y means NOT (x=y)
b1-> x, b2 -> y,z means b1 -> x, (b2 -> y, z)

2.2.12 Manifest Constant Expressions

Manifest constant expressions are expressions whose values can be determined
before the program is run. They may only consist of manifest constant names,
numbers and character constants, TRUE, FALSE, BITSPERBCPLWORD, ?, the opera-
tors MOD, SLCT, FIX, FLOAT, *, /, +, -, ABS, the relational operators, #*, #/, #+,
#-, #ABS, #MOD, the floating point relational operators, <<, >>, NOT, ~, &, |, EQV,
XOR, and conditional expressions. Manifest expressions are used in MANIFEST,
GLOBAL and STATIC declarations, the upper bound in vector declarations and the
step length in FOR commands, and as the left hand operand of OF. Floating point
manifest expressions are undefined if the word length of the compiler is different
from that of the target code.
2.3. COMMANDS 25

9 Names, Literals, ?,
TRUE, FALSE, BITSPERBCPLWORD,
BREAK, LOOP, ENDCASE,
NEXT, EXIT
(E),
9L SLCT : Field selectors
Function and method calls
Subscripted expressions using [ and ]
8L ! % OF Dyadic
7 ! @ Prefixed
6L * / MOD #* #/ #MOD Dyadic and
5 + - ABS #+ #- #ABS monadic
4 = ~= <= >= < > Extended Relations
#= #~= #<= #>= #< #>
4L << >>
3 ~ Bitwise and Boolean operators
3L &
2L |
1L EQV XOR
1R -> , Conditional expressions
0 MATCH EVERY VALOF TABLE

Table 2.1: Operator precedence

2.3 Commands
The primary purpose of commands is to update variables, to perform in-
put/output operations, and fto control the flow of control. They are described in
the following sections.

2.3.1 Assignments
Simple assignments have the following possible forms:
L:=E
L#:=E
Lop:=E
where op is one of the following operators: !, *, /, MOD, +, -, #*, #/, #MOD, #*,
#-, <<, >>, &, |, EQV or XOR and L is a variable name or an expression of one
of the following forms: E1 !E2 , !E, E1 %E2 , %%E or K OF E. K is normally a
selector of the form SLCT length:shift:offset. For := and #:= assignments,
the right hand side is evaluated and used to update the location specified by
the left hand side. For op:= assignments, the value to assign is Lop:=R. If # is
26 CHAPTER 2. THE BCPL LANGUAGE

present a floating point assignment is performed. This causes the right hand side
to be evaluated in FLT mode with the restriction that the left hand side must
refer to a full word. See Section 2.7 for details. Typical simple assignments are
as follows:

cg_x := 1000
v!i := x+1
!ptr := mk3(op, a, b)
str%k := ch
%strp := ’A’
SLCT 8:10:1 OF p +:= 1
p &:= #x7F
w!p #*:= a

Assignments are not permitted to start with any of the following keywords:
MATCH, EVERY, BREAK, LOOP, ENDCASE, NEXT or EXIT.

A multiple assignment has the following possible forms:

L1 ,..,Ln := E1 ,..,En
L1 ,..,Ln #:= E1 ,..,En
L1 ,..,Ln op:= E1 ,..,En

These constructs allows a single command to make several assignments without

needing having to enclose then in section brackets. The assignments are done
strictly from left to right and are exactly eqivalent to:

L1 :=E1 ;. . . ; Ln := En
L1 #:=E1 ;. . . ; Ln #:= En
L1 op:=E1 ;. . . ; Ln op:= En

This conversion is performed before the application of the rules of the FLT
feature. See Section 2.7 for details.

2.3.2 Routine Calls

Both function calls and method calls as described in sections 2.2.3 and 2.2.4 are
allowed to be executed as commands. Any results produced are discarded.

2.3.3 Conditional Commands

The syntax of the three conditional commands is as follows:

IF E DO C1
UNLESS E DO C2
TEST E THEN C1 ELSE C2
2.3. COMMANDS 27

where E denotes an expression and C1 and C2 denote commands. The symbols

DO and THEN may be omitted whenever they are followed by a command keyword.
To execute a conditional command, the expression E is evaluated in a Boolean
context. If it yields a non zero value and C1 is present then C1 is executed. If it
yields zero and C2 is present, C2 is executed.

2.3.4 Repetitive Commands

The syntax of the repetitive commands is as follows:

WHILE E DO C
UNTIL E DO C
C REPEAT
C REPEATWHILE E
C REPEATUNTIL E
FOR N = E1 TO E2 BY K DO C
FOR N = E1 TO E2 DO C
FOR N = E1 BY K DO C
FOR N = E1 DO C

The symbol DO may be omitted whenever it is followed by a command key-

word. The WHILE command repeatedly executes the command C as long as E
is non-zero. The UNTIL command executes C until E is zero. The REPEAT com-
mand executes C indefinitely. The REPEATWHILE and REPEATUNTIL commands
first execute C then behave like WHILE E DO C or UNTIL E DO C, respectively.
A FOR command declares the control variable N as a new local variable ini-
tialised with the value of E1 . The scope of the control variable is the body of
the FOR command. The control variable may not be given the FLT tag. If BY is
present, the step length is K which must be a manifest constants (see Section
2.2.12), but if omitted BY 1 is assumed. If TO is present, the end limit is E2 ,
but if omitted an infinite end limit with the same sign as the step length is as-
sumed. Until N moves beyond the end limit, the command C is executed and N
increment by the step length (which can be negative).

2.3.5 SWITCHON command

A SWITCHON command has the following form:

SWITCHON E INTO { C1 ;...; Cn }

Labels of the form DEFAULT: or CASE K: are permitted in the command se-
quence. E is evaluated and control is passed to the matching case label if it
28 CHAPTER 2. THE BCPL LANGUAGE

exists, otherwise a jump is made to the default label but, if that is not given,
control passes to the point just after the switchon command.

2.3.6 MATCH Command

A MATCH command has the following form:
MATCH ( args )
: P ,.., P BE C
...
: P ,.., P BE C
.
It consists of the word MATCH followed by a list of zero or more arguments en-
closed in parentheses. This is followed by followed by one or more match items
and optionally terminated by a dot. The match items are applied in turn to
the arguments as described in Section 2.5 executing the command in the first
match item to be satisfied. If none are satisfied the match command has no ef-
fect. Within a match item the the command NEXT causes control to pass to the
next match item, if any, and the command EXIT causes the MATCH command to
terminate.

2.3.7 EVERY Command

An EVERY command has the following form:
EVERY ( args )
: P ,.., P BE C
...
: P ,.., P BE C
.
It consists of the word EVERY followed by a list of arguments enclosed in paren-
theses, followed by a sequence of one or more match items optionally terminated
by a dot. The match items are applied in turn to the arguments as described in
Section 2.5 executing the commands of every match item that is satisfied.
Within a match item the commands NEXT causes control to pass to the next
match item, if any, and EXIT causes termination of the entire match list. If in
a MATCH expression or a function definition the result is zero, but if in an EVERY
expression the result is the sum accumulated so far. MATCH and EVERY commands
and routines do not return results.

2.3.8 Flow of Control

The commands in this section affect the flow of control.
2.3. COMMANDS 29

RESULTIS E causes E to be evaluated and returned as the result of the smallest

textually enclosing VALOF expression which must be within the current function
or routine.
RETURN causes a return from the current routine.
LOOP causes a jump to the point just after the end of the body of the smallest
textually enclosing repetitive command (see Section 2.3.4). The destination of the
jump must be within the current function or routine. For a REPEAT command,
LOOP causes the body to be executed again. For a FOR command, it causes a
jump to where the control variable is incremented, and for the REPEATWHILE and
REPEATUNTIL commands, it causes a jump to the place where the controlling
expression is re-evaluated.
BREAK causes a jump to the point just after the smallest enclosing repetitive
command which must be within the current function or routine.
ENDCASE causes execution to jump to the point just after the end of the small-
est enclosing SWITCHON command which must be within the current function or
routine.
GOTO E command jumps to the point whose address is the value of E. E is
typically a label. See Section 2.4.1 for details on how labels are declared. The
destination of a GOTO must be within the current function or routine.
NEXT is a newly added command that can only be used inside a the pattern or
body of a match item. It causes control to pass to the start of the next match
item, if any.
EXIT is a newly added command that can only be used inside a the pattern or
body of a match item. It causes termination of the match construct.If in a Match
expression or function body, it returns a zero result. If in an EVERY expresion, it
returns the sum acccumulated so far. MATCH and EVERY commands and routine
do not return results.
FINISH only remains in BCPL for historical reasons and should not be used. It is
equivalent to the call stop(0, 0) which causes the current program to terminate.
See the description of stop(code, res) page 73.

2.3.9 Compound Commands

It is often useful to be able to execute commands in a sequence, and this can be
done by writing the commands one after another, separated by semicolons and
enclosed in section brackets. The syntax is as follows:

{ C1 ;...; Cm }
30 CHAPTER 2. THE BCPL LANGUAGE

where C1 to Cm are commands. It is permissible to have no commands in a

command sequence, thus {} is allowed and performs no commands.
Any semicolon occurring at the end of a line may be omitted. For this rule
to work, infixed expression operators may never start a line (see Section 2.2.11).
A command sequence can also be formed using the symbol <> which be-
haves like semicolon but is more binding than DO, THEN, ELSE, REPEATWHILE,
REPEATUNTIL and REPEAT. It purpose is to reduce the need for section brackets
({ and }) as in
IF x<y DO t:=x <> x:=y <> y:=t

which is equivalent to:

IF x<y DO { t:=x; x:=y; y:=t }

This sequencing operator has been included since it was in the extended non
standard version of BCPL at MIT and extensively used in the PAL compiler.
See for instance com/pal75.b.

2.3.10 Blocks
A block is similar to a compound command but may start with some declarations.
The syntax is as follows:

{ D1 ;...; Dn ; C1 ;...; Cm }

where D1 to Dn are delarations and C1 to Cm are commands. The declarations

are executed in sequence to initialise any variables declared. A name may be used
on the right hand side of its own and succeeding declarations and the commands
(the body) of the block.

2.4 Declarations
Each name used in BCPL program must in the scope of its declaration. The
scope of names declared at the outermost level of a program include the right
hand side of its own declaration and all the remaining declarations in the section.
The scope of names declared at the head of a block include the right hand side of
its own declaration, the succeeding declarations and the body of the block. Such
declarations are introduced by the keywords MANIFEST, STATIC, GLOBAL and LET.
A name is also declared when it occurs as the control variable of a for loop. The
scope of such a name is the body of the for loop.
2.4. DECLARATIONS 31

2.4.1 Labels
The only other way to declare a name is as a label of the form N :. This may
prefix a command or occur just before the closing section bracket of a compound
command or block. The scope of a label is the body of the block or compound
command in which it was declared.

2.4.2 Manifest Declarations

A MANIFEST declaration has the following form:

MANIFEST { N1 = K1 ;...; Nn = Kn }

where N1 ,...,Nn are names (see Section 2.2.1) and K1 ,...,Kn are manifest
constant expressions (see Section 2.2.12). Each name is declared to have the
constant value specified by the corresponding manifest expression. The details
have recently changed due to the introduction of the FLT feature, see Section 2.7
on page 40. Manifest names with the FLT tag have floating point values otherwise
they are integers. If a value specification (=K1 ) is omitted in the declaration of the
first name, the value 1 or 1.0 is assumed. If a value specification (=Ki ) is omitted
in later declarations a value 1 or 1.0 greater than the value of the previous name
is used. An automatic conversion between integer and floating point is performed
if necessary. Thus, the declaration:

MANIFEST { A; B; FLT C=10; D; E=C+100 }

declares A, B, C, D and E to have manifest values 0, 1, 10.0, 11 and 110, respec-

tively.

2.4.3 Global Declarations

The global vector is a permanently allocated region of memory that may be
directly accessed by any (separately compiled) section of a program (see Sec-
tion 2.6). It provides a mechanism for linking together separately compiled sec-
tions. A GLOBAL declaration allows a names to be explicitly associated with
elements of the global vector. The syntax is as follows:

GLOBAL { N1 :K1 ;...; Nn :Kn }

where N1 ,...,Nn are names possibly prefixed by FLT (see Section 2.2.1) and
K1 ,...,Kn are manifest constants (see Section 2.2.12). Each constant specifies
which global vector element is associated with each variable, and if FLT is specified
the variable is assumed to hold a floating point number.
If a global number (:Ki ) is omitted, the next global variable element is im-
plied. If :K1 is omitted, then :0 is assumed. Thus, the declaration:
32 CHAPTER 2. THE BCPL LANGUAGE

GLOBAL { a; b:200; c; d:251 }

declares the variables a, b, c and d occupy positions 0, 200, 201 and 251 of the
global vector, respectively.

2.4.4 Static Declarations

A STATIC declaration has the following form:

STATIC { N1 =K1 ;...; Nn =Kn }

where N1 ,...,Nn are names possibly prefixed by FLT (see Section 2.2.1) and
K1 ,...,Kn are manifest constant expressions (see Section 2.2.12). Each name
is declared to be a statically allocated variable initialised to the corresponding
manifest expression. If a value specification (=Ki ) is omitted, the a value 0 or 0.0
is implied. Thus, the declaration:

STATIC { A; B; FLT C=10; D; E=100 }

declares A, B, C, D and E to be static variables having initial values 0, 0, 10.0, 0

and 100, respectively.

2.4.5 LET Declarations

LET declarations are used to define local variables, local vectors, and functions.
The textual scope of names declared in a LET declaration is the right hand side
of its own definition (to allow recursive functions), and subsequent definitions,
declarations and commands.
Local variable, local vector, function definitions can be combined using the
word AND. The only effect of this is to extend the scope of names defined back to
the word LET, thus allowing the definition of mutually recursive functions.

Local Variable Definitions

A local variable definition has the following form:
N1 ,..., Nn = E1 ,..., En
where N1 ,...,Nn are names possibly prefixed by FLT (see Section 2.2.1) and
E1 ,...,En are expressions. The names, Ni , are allocated space in the current
stack frame and are initialized with the corresponding values of Ei . Such vari-
ables are called dynamic variables since they are allocated when the definition is
executed and cease to exist when control leaves their scope.
The variables N1 ,...,Nn are allocated consecutive locations in the stack
frame of the current function and so, for instance, the variable Ni may be accessed
by the expression (@N1 )!(i − 1). This feature is a recent addition to the language.
2.4. DECLARATIONS 33

When a local variable has the FLT tag it initial value expression is evaluated in
FLT mode. For details see Section 2.7 on page 40.
The query expression (?) should be used on the right hand side when a
variable does not need a specified initial value.

Local Vector Definitions

N = VEC K

where N is a name which may not be qualified by the FLT tag and K is a manifest
constant. A location is allocated for N and initialized to point to a vector whose
lower bound is 0 and whose upper bound is K. The variable N and the vector
elements (N !0 to N !K) reside in the stack frame of the current function and
only continue to exist while control remains within the function.

Function Definitions
These definitions have the following form:
N ( N1 ,..., Nn ) = E
N ( N1 ,..., Nn ) BE C
where N is the name of the function being defined, and N1 ,...,Nn are its formal
parameters, each of which may be prefixed by FLT. A function defined using =
returns E as result, but one defined using BE and executes the command C and
does not return a defined a result. Functions defined using BE are often called
routines.
Some example functions definitions are as follows.

LET wrpn(n) BE { IF n>9 DO wrpn(n/10)

wrch(n MOD 10 + ’0’)
}
LET gray(n) = n XOR n>>1
LET next() = VALOF { c := c+1
RESULTIS !c
}

A function can be defined using pattern matching by giving its name followed
a one or more match items of the form:
: Plist => E.
optionally followed by a dot. The way patterns work is described in on page 35.
A routine can be defined using pattern matching by giving its name followed
a one or more match items of the form:
: Plist BE C.
34 CHAPTER 2. THE BCPL LANGUAGE

optinally followed by a dot. The way patterns work is described in on page 35.
Example functions defined using pattern matching are as follows.

LET ways
: 0, ? => 1
: ?, [0] => 0
: s, coins[>s] => ways(s, coins+1)
: s, coins[v ] => ways(s, coins+1) + ways(s-v, coins)

LET eval
: [Id, x], e => lookup(x, e)
: [Num, k], ? => k
: [Mul, x, y], e => eval(x, e) * eval(y, e)
: [Div, x, y], e => eval(x, e) / eval(y, e)
: [Add, x, y], e => eval(x, e) + eval(y, e)
: [Sub, x, y], e => eval(x, e) - eval(y, e)

If a function or routine is defined in the scope of a global variable with the

same name, the global variable is given an initial value representing that function
or routine (see section 2.6).
If a function is called as a command its result is thrown away, and if a routine
is called when a result is required its result is undefined. See section 2.2.3 for
information about the syntax of function and routine calls.
The arguments of a functions and routines behave like named elements of a
dynamic vector and so exist only for the lifetime of the call. This vector has as
many elements as there are formal parameters and they receive their initial values
from the actual parameters of the call. Functions and routines are variadic; that
is, the number of actual parameters need not equal the number of formals. If
there are too few actual parameters, the missing ones are left uninitialized, and
if there are too many actual parameters, the extra ones are evaluated and then
discarded. Notice that arguments can be accessed by the expressions (@x)!0,
(@x)!1, (@x)!2,. . . where x is the first argument. This feature is useful in the
definition of functions, such as writef, having a variable number of arguments.
The scope of the formal parameters is the body of the function or routine.
Function and routine calls are cheap in both space and execution time, with
a typical space overhead of three words of stack per call plus one word for each
formal parameter. In the Cintcode implementation, the execution overhead is
typically just one Cintcode instruction for the call and one for the return.
There are two important restrictions concerning functions and routines. One
is that a GOTO command cannot make a jump to a label not declared within the
current function or routine, although such non local jumps can be made using
level and longjump, described on page 65. The other is that dynamic free
variables are not permitted.
2.5. PATTERNS 35

2.4.6 Dynamic Free Variables

Free variables of a function or routine are those that are used but not declared in
the function or routine, and they are restricted to be either manifest constants,
static variables, global variables, functions, routines or labels. This implies that
they are not permitted to be dynamic variables (ie local variables) of another
function or routine. There are several reasons for this restriction, including the
ability to represent a function or routine by a single BCPL word, the ability to
provide a safe separate compilation with the related ability to assign functions
and routines to variables. It also allows calls to be efficient. Programmers used
to languages such as Algol or Pascal will find that they need to change their
programming style somewhat; however, most experienced BCPL users agree that
the restriction is well worthwhile. Note that C adopted the same restriction,
although in that language it is imposed by the simple expedient of insisting
that all functions are declared at the outermost level, thus making dynamic free
variables syntactically impossible.
A style of programming that is often be used to avoid the dynamic free variable
restriction is exemplified below.

GLOBAL { var:200 }

LET f1(...) BE
{ LET oldvar = var // Save the current value of var
var := ... // Use var during the call of f1
...
f2(...) // var may be used in f2
...
IF ... DO f1(...) // f1 may be called recursively
var := oldvar // restore the original value of var
}

AND f2(...) BE // f2 uses var as a free variable

{ ... var ... }

2.5 Patterns
This section describes the MCPL style pattern match features that is now avail-
able in BCPL.
Pattern matching is an important feature since it provides a mechanism to
select an outcome based on the values of locations in a structure referenced di-
rectly or indirectly from a set of arguments. Names, called pattern variables,
can be associated with locations in the structure. Such variables have much in
common with ordinary local variables. They can only be used in the pattern and
the expression or command in the match item.
36 CHAPTER 2. THE BCPL LANGUAGE

Patterns are used in function and routine definitions and in MATCH and EVERY
constructs. They are applied to the arguments given by the function or routine
calls or the arguments of MATCH and EVERY constructs. Each matxh construct has
a sequence of one or more match items, consisting of a pattern and an expression
or command which must be of one of the following forms.
: Plist => E
: Plist BE C
where Plist is a list of zero or more pattern terms separated by commas. E is
an expression and C is a command. The sequence of match items is optionally
terminated by a dot. If the first match item in a sequence used the operator =>
then all the subsequent items must also use =>, otherwise all the match items
must use BE.
For functions routine and MATCH expressions and commands, the patterns
are tested in turn and the first to be successful causes the related expression or
command to be evaluated.
For an EVERY command, the commands in all successful match items are exe-
cuted, and for an EVERY expression the values of the expressions of all successful
match items are summed and returned as the result.
A pattern term tests an individual location. It may test the numerical value
contained in the location or, if the location holds a pointer, it can apply a pattern
list to the consecutive memory locations pointed to. Pattern terms in a list test
consective locations in left to right order. A pattern can only successfully match
its arguments if every pattern term it contains is suceessful.
There are many kinds of term and, just as with ordinary expressions, it is
convenient to define the syntax using precedence. The grammar will use syntactic
categories for the different precedence levels. Plist, P1, P2, P3. P4, P5 and P6
are the term categories from lowest to highest precedence.
The syntax of P6 and P5 is as follows:
P6 -> an integer constant
| a floating point constant

P5 -> manifest constant name

| a character constant
| FALSE
| TRUE
| + T6
| - T6
| T6
Thus P5 can be a possibly signed numeric constant, a manifest name, a character
costant or a truth value. A character constant is just the integer code for the
character, and TRUE and FALSE are just the integers -1 and zero respectively.
Some example terms of category P5 are: 123, -1.5, s sub and ’A’.
2.5. PATTERNS 37

A P5 term successfully matched if its numeric value is equal to the contents

of the location being tested.
P4 has the following syntax:
T4 -> T5 .. T5
| T5
Thus P4 includes all terms belonging to P5 and also a term using the range
operator ... A typical range term is 10 .. 12. This will match a location con-
taining 10, 11 or 12, ie all values in the integer range 10 to 12. The representation
of the range operator by .. is problematic since dots are allowed but discouraged
in variable names and and are also used in floating point constants. BCPL no
longer allows consective dot in variable names and consecutive dots were never
valid in floating point constants, so s add..s div and 10..12 are valid range terms.
A P3 term has the following syntax:
T3 -> T4 — T3
| T4
Thus P3 is list of T4 terms separated by vertical bars. A P3 term matches a
location if any of its terms match the location.
A P2 term has the following syntax:
P2 -> a non manifest name
| BREAK
| LOOP
| ENDCASE
| NEXT
| EXIT
| ?
| relop P5
| relop ( E )
— [ Plist ]
| T3
where relop is one of =, ~=, <, >, <=, >=, #=, #~=, #<, #>, #<=, #>= and E is an
expression.
A non manifest name always matches its location successfully. Its purpose is
to declare a pattern variable associated with the location being matched. This
variable can be used anywhere in the pattern, expression or command of the
current match item.
The terms BREAK, LOOP, ENDCASE, NEXT and EXIT have the same effect in patterns
as when executed as a command, as described is section 2.3.8.
The pattern ? is always successful.
The pattern relop E if X relop E evaluates to TRUE where X is the value contained
in the current location.
38 CHAPTER 2. THE BCPL LANGUAGE

As an example of the use of pattern variables and relation terms, the pat-
tern [<x, x] will be successful if the current location contains a pointer to two
consective integers where the first is less than the second.
The term of the form [ em Plist ] assumes the current location holds a pointer
to consecutive locations . The match is only successful if every P1 term in the
Plist successfully matches its location. This construct can only be nested up to
a depth of 4.
If a the position the current location relative to P, the base of the current stack
frame, any pattern variables will associated with one of the following location.
P!a
P!a!r
P!a!r!s
P!a!r!s!t
P!a!r!s!t!u
where r, s, t and u are all offsets in the range 0 to 255, identfying the path from
an argument to the location associated with the declared name.
A P1 term has the following syntax
P1 -> P2 P1
| P2
P1 is thus just a list of consecutive T2 terms. It is is applied to a single location
and is only satisfied if at the T2 terms applied to this location are successful.
The syntax of Plist is as follows:
Plist -> P1 , Plist
— P1
A Plist is applied to the consecutive locations of the argument lists of function
or routine calls or the arguments of MATCH and EVERY constructs. It can also be
applied to the consecutive locations pointed to by a term of the form [ Plist ].
The consecutive P1 terms are applied to the consective locations. The Plist is
only successful if every P1 term successfully matches it argument location.

2.6 Separate Compilation

Large BCPL programs can be split up into sections that can be compiled sepa-
rately. When loaded into memory they can communicate with each other using
a special area of store called the Global Vector. This mechanism is simple and
machine independent and was put into the language since linkage editors at the
time were so primitive and machine dependent.
Variables residing in the global vector are declared by GLOBAL declarations
(see section 2.4.3). Such variables can be shared between separately compiled
sections. This mechanism is similar to the used of BLANK COMMON in Fortran,
2.6. SEPARATE COMPILATION 39

however there is an additional simple rule to permit access to functions and

routines declared in different sections.
If the definition of a function or routine occurs within the scope of a global
declaration for the same name, it provides the initial value for the corresponding
global variable. Initialization of such global variables takes place at load time.
The three files shown in Table 2.1 form a simple example of how separate
compilation can be organised.

File demohdr File demolib.b File demomain.b

GET "libhdr" GET "demohdr" GET "demohdr"

GLOBAL { f:200 } LET f(...) = VALOF LET start() BE

{ ... { ...
} f(...)
}

Table 2.1 - Separate compilation example

When these sections are loaded, global 200 is initialized to the entry point of
function f defined in demolib.b and so can be called from the function start
defined in demomain.b.
The header file, libhdr, contains the global declarations of all the resident
library functions and routines making all these accessible to any section that
started with: GET "libhdr". The library is described in the next chapter. Global
variable 1 is called start and is, by convention, the first function to be called
when a program is run.
Automatic global initialisation also occurs if a label declared by colon (:)
occurs in the scope of a global of the same name.
Although the global vector mechanism has disadvantages, particularly in the
organisation of library packages, there are some compensating benefits arising
from its extreme simplicity. One is that the output of the compiler is available
directly for execution without the need for a link editing step. Sections may
also be loaded and unloaded dynamically during the execution of a program
using the library functions loadseg and unloadseg, and so arbitrary overlaying
schemes can be organised easily. An example of where this is used is in the
implementation of the Command Language Interpreter described in Chapter 4.
The global vector also allows for a simple but effective interactive debugging
system without the need for compiler constructed symbol tables. Again, this was
devised when machines were small, disc space was very limited and modern day
linkage editors had not been invented; however, some of its advantages are still
relevant today.
40 CHAPTER 2. THE BCPL LANGUAGE

2.7 The FLT Feature

BCPL was originally designed for the implemention of compilers and other system
software such as text editors, pagination programs and operating systems. These
applications typically did not require floating point and so the language did not
include any floating point features. Indeed, many early machines on which BCPL
ran had word lengths of 16 or 24 bits which were of insufficient length for useful
floating point numbers. Even on 32-bit machines the precision of floating point
numbers is limited to about 6 decimal digits which is insuffient for serious scientific
calculations. For 50 years I resisted putting floating point into BCPL but have
recently given in. This is mainly due to the need to use 32-bit floating point in
the BCPL interface with the OpenGL graphics library described bcplraspi.pdf.
The same document also describes a flight simulator where the inaccuracy of 32-
bit floating point can be put down to random turbulence of the air through which
the aircraft is flying.
The FLT feature was added to BCPL in March 2018 to make computations
involving floating point numbers more convenient. This feature may look as if
data types have been added to BCPL, but the language still retains its typeless
nature in the sense that all expression values are the same size and the compiler
generates no error messages relating to data types. The sole effect of this feature
is to cause some integer operators such as + and - to be replaced automatically by
their floating point versions #+ and #- and to automatically replace some integer
constants by floating point ones. These conversions are specified by the follow-
ing simple rules, but before applying them it is assumed that all simultaneous
assignments have been replaced by sequences of non simultaneous assignments
before this rule is applied.
(1) Global, static and manifest, local variable and formal parameter names can
be prefixed by FLT giving them the FLT tag. All other names, namely FOR loop
control variables, functions, vectors and program labels, may not be given the
FLT tag.
(2) An expression has the FLT tag if it is a name declared with the FLT tag, a
floating point constant or it has one of the following leading operators FLOAT,
#ABS, #*, #/, #MOD, #+, #- and #->.
(3) If one of the operators ABS, *, /, MOD, +, -, =, ~=, <, <=, >, >=, :=, ABS:=, *:=,
/:=, MOD:=, += or -= has an operand with the FLT tag, the operator is replaced
by the corresponding floating point version. The operator -> is replaced by #->
if its second or third operand has the FLT tag..
(4) Expressions are evaluated in either FLT or non-FLT mode. Expressions are
evaluated in FLT mode if they are operands of FIX, #ABS, #*, #/, #MOD, #+, #-, #=,
#~=, #<, #<=, #> or #>=. The second and third operands of #-> and the second
operand of #:=, #ABS:=, #*:=, #/:=, #MOD:=, #+:= and #-:= are evaluated in
2.8. THE OBJLINE1 FEATURE 41

FLT mode. Expressions giving the values of static, manifest and local variable
names with the FLT tag are also evaluated in FLT mode. All other expressions
are evaluate in non-FLT mode.
(5) If the leading operator of an expression evaluated in FLT mode is one of ABS,
*, /, MOD, +, - or -> it is replaced by the floating point version. If an integer
constant is evaluated in FLT mode it is replaced by the corresponding floating
point constant.
These rules are applied repeatedly until there is no further change. As an
example, the following function:

LET f(a, FLT b, c) = VALOF

{ a, b, c := 1/2, 1/2, 2 * (a + b)
RESULTIS a + 2.0 * c + (0|#x3840000)
}

is automatically converted to:

LET f(a, FLT b, c) = VALOF

{ { a := 1/2; b #:= 1.0#/2.0; c #:= 2.0 #* (a #+ b) }
RESULTIS a #+ 2.0 #* c #+ (0|#x3840000)
}

Notice that the user almost never needs to use # to specify floating point opera-
tions. Note also that the expression (0|#x3840000|) is a way to protect an ex-
pression (#x38400000) from being evaluated in FLT mode. See com/xcmpltest.b
and bcpl4raspi.pdf from my home page for examples of the use of the FLT fea-
ture.
To see why the name in a vector declarations may not be given the FLT tag,
consider LET v = VEC 5. This initialises v with a pointer to a vector with 6
elements and the expression v+1 would point to the element at subscript position
one. Had v been given the FLT tag, v+1 would have been automatically converted
to v#+1.0 which is clearly meaningless.
The FLT tag is not permitted to qualify FOR loop control variables since, due
to floating point truncation and rounding errors, the number of iterations of a
loop such as FOR FLT x = 0.1 TO 10.0 DO... is properly defined.

2.8 The objline1 Feature

If a file named objline1 is found in the current directory or the other directories
searched by GET directives, its first line is copied as the first line of the compiled
Cintcode module. This will typically put a line such as:

#!/usr/local/bin/cintsys -c
42 CHAPTER 2. THE BCPL LANGUAGE

as the first line of the compiled object module. This line is ignored by the CLI
but under Linux it allows Cintcode programs to be called directly from a Linux
shell. If objline1 cannot be found no such line is inserted at the start of the
object module.
Chapter 3

The Library

This manual describes three variants of the BCPL system. The simplest is in-
voked by the shell command cintsys and provides a single threaded command
language interpreter. The system invoked by cintpos provides a multi-threaded
system where the individual threads (called tasks) are run in parallel and are
pre-emptible. A third version is available for some architectures and provides a
single threaded version in which the BCPL source is compiled into native machine
code. Although this version is faster, it is more machine dependent, has fewer
debugging aids and will only run a single command.
The libraries of these three systems have much in common and so are all
described together. The description of all constants, variables and functions have
a right justified line such as the following
CIN:y, POS:y, NAT:n
where CIN:, POS: and NAT: denote the single threaded, multi-threaded and native
code versions, respectively, and the letters y and n stand for yes and no, showing
whether the corresponding constant, variable or function is available on that
version of the system.
The resident library functions, variables and manifest constants are declared
in the standard library header file g/libhdr.h. Most of the functions are defined
in BCPL in either sysb/blib.b or sysb/dlib.b, but three functions (sys, chgco
and muldiv) are in the hand written Cintcode file cin/syscin/syslib. Most
functions relating to the multi-threaded version are defined in klib.b.
The following three sections describe the manifest constants, variables and
functions (in alphabetical order) provided by the standard library.

3.1 Manifest constants

B2Wsh CIN:y, POS:y, NAT:y

This constant holds the shift required to convert a BCPL pointer into a byte address.

43
44 CHAPTER 3. THE LIBRARY

Most implementations pack 4 bytes into 32-bit words requiring B2Wsh=2, but on 64-bit
implementations, such as native code on the DEC Alpha or the 64-bit Cintcode version
of BCPL, its value is 3.

bootregs CIN:y, POS:y, NAT:n

This is the location in Cintcode memory used in Cintpos to hold Cintcode registers
during system startup.

bytesperword CIN:y, POS:y, NAT:y

Its value is 1<<B2Wsh being the number of bytes that can be packed into a BCPL
word. On 32-bit implementations it is 4, and on 64-bit versions it is 8.

bitsperbyte CIN:y, POS:y, NAT:y

This specifies the number of bits per byte. On most systems bitsperbyte is 8.

bitsperword CIN:y, POS:y, NAT:y

It value is bitsperbyte*bytesperword being the number of bits per BCPL word.
It is usually 32, but can be 64.

CloseObj CIN:y, POS:y, NAT:y

This identifies the position of the close method in objects using BCPL’s version
of object oriented programming. Typical use is as follows:

CloseObj#(obj)

For more details, see mkobj described on page 65.

co c, co fn, co list, co parent, co pptr, co size CIN:y, POS:y, NAT:y

These are the system fields as the base of coroutine stacks. If a coroutine is sus-
pended, its pptr field holds the stack frame pointer (P) at the time it became sus-
pended. The parent field points to the parent coroutine, if it has one, or is -1 for root
coroutines, and is zero otherwise. The list field holds the next coroutine in the list of
coroutines originating from global colist. The fn and size fields hold the coroutine’s
main function and stack size, and the c field is a system work location. For more
information about coroutines, see createco described on page 58.

deadcode CIN:y, POS:y, NAT:n

To aid debugging, the entire Cintcode memory is initialised to deadcode. Typically
deadcode=#xDEADC0DE.

endstreamch CIN:y, POS:y, NAT:y

This is the value returned by rdch when reading from a stream that is exhausted.
Its value is normally -1.

entryword CIN:y, POS:y, NAT:n

To aid debugging, every functions entry point is marked by entryword. This is
normally followed by a function name compressed into a string of 11 characters. If the
3.1. MANIFEST CONSTANTS 45

function name is too long its first and last five character are packed into the string
separated by a single quote ’. Typically entryword=#x0000DFDF.

fl . . . CIN:y, POS:y, NAT:n

Constants of the form fl ... are mnemonics for the floating point operations
performed by the call sys(Sys flt, op, ...) as described near page 76.

globword CIN:y, POS:y, NAT:n

This constant is used to assist the debugging of Cintcode programs. If the ith global
variable is not otherwise set, its value is globword+i. Typically globword=#x8F8F0000.

id inscb, id inoutscb, id outscb CIN:y, POS:y, NAT:n

These constants are mnemonics for the possible values of the id field of a stream
control block. See scb id below.

InitObj CIN:y, POS:y, NAT:y

This identifies the position of the init method in objects using BCPL’s version of
object oriented programming. Typical use is as follows:
InitObj#(obj, arg1, arg2)

For more details, see mkobj described on page 65.

isrregs CIN:n, POS:y, NAT:n

Under Cintpos this is the location in Cintcode memory used to hold the Cintcode
registers representing the state at the start of the interrupt service routine.

klibregs CIN:n, POS:y, NAT:n

Under Cintpos This is the location in Cintcode memory used to hold Cintcode
registers during system startup.

mcaddrinc CIN:y, POS:y, NAT:y

This is the difference between machine addresses of consecutive words in memory
and is usually 4 or 8. Very occasionally, BCPL implementions have negatively growing
stacks, in which case mcaddrinc will be negative.

maxint, minint CIN:y, POS:y, NAT:y

The constant minint is 1<<(bitsperword-1) and maxint is =minint-1. They hold
the most negative and largest positive numbers that can be represented by a BCPL
word. On 32-bit implementations, they are normally #x80000000 and #x7FFFFFFF.

pollingch CIN:n, POS:y, NAT:n

This is the value returned by rdch if a charcter is not immediately available from
the currently selected stream. Its value is normally -3. Currently only TCP streams
under Cintpos provide the polling mechanism.

rootnodeaddr CIN:y, POS:y, NAT:n

This manifest constant is used in Cintsys and Cintpos to hold the address of the
root node. Its value is otherwise zero.
46 CHAPTER 3. THE LIBRARY

rtn . . . CIN:y, POS:y, NAT:y

The root node is a vector accessible to all running programs to provide access to
all global information. It is available in all versions of BCPL but many of its fields are
only used in Cintpos. The global variable rootnode holds a pointer to the root node.
On some systems the address of the root node is also held in the manifest constant
rootnodeaddr. Manifest constants starting with rtn give the positions of the fields
within the root node.

rtn abortcode CIN:y, POS:y, NAT:n

This rootnode field holds the most recent return code from a command language
interpreter (CLI). It is used by commands such as dumpsys and dumpdebug when in-
specting Cintcode memory dumps.

rtn adjclock CIN:y, POS:y, NAT:n

This rootnode field holds a correction in minutes to be added to the time of day
supplied by the system. It is normally set to zero.

rtn blklist CIN:y, POS:y, NAT:y

All blocks of memory whether free or in used are chained together in increasing
address order. This rootnode field points to the first in the chain.

rtn blib CIN:y, POS:y, NAT:n

Under Cintsys and Cintpos this rootnode field holds the appropriate versions of the
modules BLIB, SYSLIB and DLIB chained together.

rtn boot CIN:y, POS:y, NAT:n

Under Cintsys and Cintpos this rootnode field holds the appropriate version of the
BOOT module.

rtn boottrace CIN:y, POS:y, NAT:n

Under Cintsys and Cintpos this rootnode field holds 0, 1, 2 or 3. The default value
is 0 but can be incremented using the -v option. Larger values of boottace generate
more tracing information.

rtn bptaddr, rtn bptinstr CIN:y, POS:y, NAT:n

These each hold vectors of 10 elements used by the standalone debugger to hold
breakpoint addresses and operation codes overwritten by BRK instructions. They are
in the rootnode to make them accessible to the debug task in Cintpos and to the
dumpdebug command.

rtn clkintson CIN:n, POS:y, NAT:n

Under Cintpos, this boolean field controls whether clock interrupts are enabled.
It is provided to make single step execution possible within the interactive debugger
without interference from clock interrupts. For more details see the chapter on the
debugger starting on page 167.
3.1. MANIFEST CONSTANTS 47

rtn clwkq CIN:n, POS:y, NAT:n

Under Cintpos, this field is used to holds the ordered list of packets waiting to be
released by the clock device.

rtn context CIN:y, POS:y, NAT:n

Under certain circumstances the entire Cintcode memory is dumped in a compacted
form to the file DUMP.mem for later inspection by commands such as dumpsys and
dumpdebug. This field is set at the time a dump file is written to specify why the dump
was requested. The possible values are as follows:
1: dump caused by second SIGINT
2: dump caused by SIGSEGV
3: fault in BOOT or standalone debug
4: dump by user calling sys(Sys_quit, -2)
5: dump caused by non zero user fault code
6: dump requested from standalone debug

rtn crntask CIN:y, POS:y, NAT:n

Under Cintpos, this rootnode field point to the TCB of the currently running task,
which is the highest priority task that can run.

rtn days CIN:y, POS:y, NAT:n

This field holds the number of days since 1 January 1970. It is updated by the
interpreter normally within a milli-second of the date changing.

rtn dbgvars CIN:y, POS:y, NAT:n

This rootnode field holds vectors of 10 elements used by the standalone debugger
to hold the debugger variables V0 to V9. It is in the rootnode to make it accesible to
the debugger and to programs that inspect Cintcode memory dumps.

rtn dcountv CIN:y, POS:y, NAT:n

This holds a pointer to the debug count vector. These counters can be incremented
by calls of the form sys(Sys incdcount, n) or by similar calls in C within the Cintcode
interpreter. The zeroth element of dcountv holds it upper bound which is typically
511.

rtn devtab CIN:y, POS:y, NAT:n

Under Cintpos, this holds the Cintpos device table. The zeroth entry is the ta-
ble’s upperbound and each other entries is either zero, or points to the device control
block (DCB) of the corresponding device. Some devices are handled by polling in the
interpreter thread based on the count of Cintcode instructions obeyed. Currently the
clock (device -1) and ttyout (device -3) are handled in this way. This improved the
performance of output to the screen and causes the clock to have a resolution of about
1 milli-second although the actual clock precision is usually limited by the underlying
operating system.
48 CHAPTER 3. THE LIBRARY

rtn dumpflag CIN:y, POS:y, NAT:n

If dumpflag is TRUE when Cintsys or Cintpos exits, the entire Cintcode memory is
dumped in a compacted form to the file DUMP.mem for later inspection by commands
such as dumpsys or dumpdebug.

rtn envlist CIN:y, POS:y, NAT:n

This rootnode field holds the list of logical name-value pairs used by the functions
setlogval and getlogval, and the CLI command setlogval. The environment vari-
able held in envlist are distinct from those such as BCPLROOT held by the underlying
operating system but have a similar purpose.

rtn hdrsvar CIN:y, POS:y, NAT:n

This field holds the name of the environment variable giving the directories holding
BCPL headers, typically ”BCPLHDRS” or ”POSHDRS”. See Section 3.6 for more
details.

rtn idletcb CIN:n, POS:y, NAT:n

This rootnode field holds the TCB of the IDLE task for used by the standalone
debugger and the commands dumpsys and dumpdebug. The task number of the IDLE
task is zero but it is not a proper task and does not have an entry in the task table.
The Cintpos scheduler gives it control when all other tasks are suspended.

rtn info CIN:y, POS:y, NAT:n

This rootnode field holds a vector of information that can be shared between all
tasks. It is typically a vector of 50 elements. The use of these elements are system
dependent.

rtn insadebug CIN:n, POS:y, NAT:n

This rootnode field is used by the keyboard input device of Cintpos to tell it whether
to place a newly received character in a request packet or just store it in the lastch
field.

rtn intflag CIN:y, POS:y, NAT:n

This flag is set to TRUE on receiving an interrupt from the user (typically a SIGINT
signal generated by ctrl-C) and is reset to FALSE whenever the standalone debugger is
entered. Cintsys or cintpos exits if a user interrupt is received when intflag is TRUE
or if control is within BOOT or sadebug.

rtn gvecsize CIN:y, POS:y, NAT:n

This field holds the size of global vectors created from now on. The default size is
now 2000 but it can be set using the -g argument when entering cintsys or cintpos.

rtn keyboard CIN:y, POS:y, NAT:n

This rootnode field holds the stream control block for the standard keyboard device.
3.1. MANIFEST CONSTANTS 49

rtn klib CIN:y, POS:y, NAT:n

Under Cintpos this rootnode filed holds the the KLIB module. It is otherwise zero.

rtn lastch CIN:n, POS:y, NAT:n

This rootnode field holds the most recent character received from the keyboard
device. The standalone debugger uses it for polling input. On reading this field the
standalone debugger resets it to pollingch=-3.

rtn lastg, rtn lastp, rtn lastst CIN:y, POS:y, NAT:n

These rootnode fields hold the most recent settings of the Cintcode P, G and ST
registers. They are used by the commands dumpsys and dumpdebug when inspecting a
Cintcode memory dump caused by faults such as memory violation (SIGSEGV) when
all other Cintcode dumped registers are invalid.

rtn mc0, rtn mc1, rtn mc2, rtn mc3 CIN:y, POS:y, NAT:n
These hold the machine address of the start of the Cintcode memory and other
values used by the MC package.

rtn membase, rtn memsize CIN:y, POS:y, NAT:n

These rootnode fields hold, respectively, the start of the memory block chain and
the upper bound in words of the Cintcode memory.

rtn msecs CIN:y, POS:y, NAT:n

This field holds the number of milli-seconds since midnight. It is repeatedly updated
by the interpreter and its value is normally correct to the nearest milli-second.

rtn pathvar CIN:y, POS:y, NAT:n

This field holds the name of the environment variable giving the directories searched
by loadseg, typically ”BCPLPATH” or ”POSPATH”. See Section 3.6 for more details.

rtn rootvar CIN:y, POS:y, NAT:n

This field holds the name of the environment variable holding the system root
directory, typically ”BCPLROOT” or ”POSROOT”. See Section 3.6 for more details.

rtn scriptsvar CIN:y, POS:y, NAT:n

This field holds the name of the environment variable giving the directories holding
CLI script files, typically ”BCPLSCRIPTS” or ”POSSCRIPTS”. See Section 3.6 for
more details.

rtn screen CIN:y, POS:y, NAT:n

This rootnode field holds the stream control block for the standard screen device.

rtn sys CIN:y, POS:y, NAT:n

Under Cintsys and Cintpos, this holds the entry point to the sys function.

rtn system CIN:y, POS:y, NAT:y

This rootnode field holds 1 when Cintsys is running or 2 when Cintpos is running.
It is otherwise zero.
50 CHAPTER 3. THE LIBRARY

rtn tallyv CIN:y, POS:y, NAT:n

This rootnode field points to a vector used to hold profile execution counts. When
tallying is enabled, the value of tallyv!i is the count of how often the Cintcode
instruction at location i has been executed. The upper bound of tallyv is held in
tallyv!0. For more information about the profile facility see the stats command
described on page 151.

rtn tasktab CIN:y, POS:y, NAT:n

Under Cintpos, this rootnode field holds the Cintpos task table. The zeroth entry
is the table’s upperbound and the other entries are either zero or points to the task
control block (TCB) of the corresponding task. Note that the IDLE task is not held in
this table since it is not a proper task. The IDLE task TCB is held in the rootnode’s
idletcb field.

rtn tcblist CIN:y, POS:y, NAT:n

Under Cintpos, all TCBs are chained together in decreasing priority order. This
rootnode field points to the first TCB in this chain and so refers to the highest priority
task. The last TCB on the chain has priority zero and represents the idle task. If not
in Cintpos this field holds zero. If not in Cintpos this field holds zero.

rtn upb CIN:y, POS:y, NAT:n

This is the upperbound of the rootnode. It value is typically 80.

rtn vecstatsv CIN:y, POS:y, NAT:n

This points to a vector holding counts of how many blocks of each requested size
have been allocated by getvec but not yet returned. It is used by the vecstats
command.

rtn vecstatsvupb CIN:y, POS:y, NAT:n

This field hold the upper bound of vecstatsv.

saveregs CIN:n, POS:y, NAT:n

This is the location in Cintcode memory used in Cintpos to hold the Cintcode
registers at the time of the most recent interrupt.

scb . . . CIN:y, POS:y, NAT:n

Each currently open stream has a stream control block (SCB) that holds all that
the system needs to know about the stream. Manifest constants beginning scb allow
convenient access to the SCB fields. These are described below.

scb blength CIN:y, POS:y, NAT:n

This SCB field holds the length of the buffer in bytes. It is typically 4096.

scb block CIN:y, POS:y, NAT:n

This SCB field holds the current block number of a disc file. The first block of a
file has number zero.
3.1. MANIFEST CONSTANTS 51

scb buf CIN:y, POS:y, NAT:n

This SCB field is either zero or points the buffer of bytes, allocated by getvec,
associated with the stream.

scb bufend CIN:y, POS:y, NAT:n

This SCB field holds the size of the buffer in bytes.

scb encoding CIN:y, POS:y, NAT:n

This SCB field controls how codewrch treats extended characters written to this
stream. If its value is GB2312, the extended character is translated into one or two
bytes in GB2312 format, otherwise the translation is to a sequence of bytes in UTF-8
format. This field is normally set using either codewrch(UTF8) or codewrch(GB2312).

scb end CIN:y, POS:y, NAT:n

This SCB field hold the number of valid bytes in the buffer or -1, if the stream is
exhausted.

scb endfn CIN:y, POS:y, NAT:n

This SCB field is either zero or the function to close the stream. It is given the
SCB as its argument and it returns TRUE if the call is successful. It otherwise returns
FALSE with an error code in result2.

scb fd scb fd1 CIN:y, POS:y, NAT:n

These SCB fields hold a machine dependent file or mailbox descriptor which is often
implemented as a native machine code address. On some machines machine addresses
are 64 bits long and so cannot be held in a variable of a 32 bit version of BCPL. So the
BCPL system allocates two consecutive words to hold such values. In order to allow the
same header files be used for 32 and 64 bit BCPL and for 32 and 64 bit machines, two
words are allocated even when one word would be sufficient. How a machine addresse is
packed in a pair of BCPL words is implementation dependent except that null pointers
cause both words to be set to zero. This mechanism is used whenever native machine
addresses are held in BCPL variables.

scb id CIN:y, POS:y, NAT:n

This SCB field holds one of the values id inscb, id outscb or id inoutscb, indi-
cating whether the stream is for input, output or both.

scb lblock CIN:y, POS:y, NAT:n

This SCB field holds the number of last block. The first block of a stream is
numbered zero.

scb ldata CIN:y, POS:y, NAT:n

This SCB field holds the number of bytes in the last block of a stream.

scb pos CIN:y, POS:y, NAT:n

This SCB field points to the position within the buffer of the next character to be
52 CHAPTER 3. THE LIBRARY

transferred. This field is updated every time a character is transferred to or from a

stream.

scb rdfn CIN:y, POS:y, NAT:n

This SCB field is zero if the stream cannot perform input, otherwise it is the
function to refill (or replenish) the buffer with more characters. It is given the SCB as
its argument and returns TRUE if it successfully replenishes the buffer with at least one
character. It otherwise returns FALSE setting result2 to -1 if the end of file has been
encountered, -2 if there was a timeout before any character were read, -3 no character
was available in polling mode. Any other value in result2 is an error code.

scb reclen CIN:y, POS:y, NAT:n

A file is normally regarded as a potentially huge sequence of bytes, but can also be
treated as a sequence of fixed length records. The reclen SCB field holds the length
in bytes of such records. The first record of a file has number zero. Unless the length
of a file is a multiple of the record length, the length of last record of a file will be too
short.

scb size CIN:y, POS:y, NAT:n

This constant is equal to the number of words in a stream control block.

scb timeout CIN:y, POS:y, NAT:n

This SCB field holds the stream timeout value for TCP streams. If it is zero no
timeout is applied. If it is negative, data is only tranferred if it is immediately available.
If it is strictly positive it represents a timeout value in milli-seconds.

scb timeoutact CIN:y, POS:y, NAT:n

This SCB field controls the effect of a time out on this stream while reading using
rdch. A value of 0 causes the time out to be ignored, a value of -1 caused the rdch to
return with the value endstreamch, and a value of -2 causes rdch to return with the
value timeoutch.

scb type CIN:y, POS:y, NAT:n

This SCB field holds the type of the stream which will be one of the following:
scbt net, scbt file, scbt ram, scbt console or scbt mbx, scbt tcp. The last three
have strictly positive values causing output to be triggered by end-of-line characters,
while the first three are negative and only trigger output when the IO buffer is full.
TCP streams have type net or tcp, streams to and from disk file have type file, stream
to or from a vector in main memory have type ram, mbx specifies mailbox streams, and
console indicates that the stream is either to standard output or from standard input
which are normally the screen and keyboard, respectively.

scb task CIN:y, POS:y, NAT:n

Under Cintpos, this SCB field holds either zero or the number of the handler task
associated with the stream, if it has one.

scb upb CIN:y, POS:y, NAT:n

This constant is the upperbound of a stream control block. its value is scb size-1.
3.1. MANIFEST CONSTANTS 53

scb wrfn CIN:y, POS:y, NAT:n

This SCB field is zero if the stream cannot perform output, otherwise it is the
function to output (or deplete) the buffer. It is given the SCB as its argument and
returns TRUE if it successfully outputs the contents of the buffer. It otherwise returns
FALSE with an error code in result2.

scb write CIN:y, POS:y, NAT:n

This SCB field is TRUE if the buffer has been updated by functions such as wrch
since it was last written out (depleted).

scbt net, scbt file, scbt ram, scbt console, scbt mbx, scbt tcp
CIN:y, POS:y, NAT:n
These constants are mnemonics for the possible values of the type field of a stream
control block. See scb type above.

sectword CIN:y, POS:y, NAT:n

This word occurs near the start of a section of code just before a compiled string of
11 character representing the section name if a section name is specified in the source
code. If the name is less than 11 characters long it is padded with spaces at the end.
If the name has more than 11 characters, the string consists of the first and last five
separated by a prime (’). Typically sectword=#x0000FDDF.

stackword CIN:y, POS:y, NAT:n

As an aid to debugging, all words in runtime stacks are initialised to stackword.
Typically stackword=#xABCD1234.

Sys . . . CIN:y, POS:y, NAT:y

Manifest constants of the form Sys ... provide mnemonics for the operations
invoked by the sys function. The use of these manifest constants is described in pages
following Section 3.3 starting on page 73.

t bhunk, t bhunk64, t end, t end64, t hunk, t hunk64, t reloc, t reloc64

CIN:y, POS:y, NAT:n
These are constants identifying components of Cintcode object modules. Cintcode
modules hold the relocatable byte stream interpretive code used by all BCPL interpre-
tive systems. Constants with names ending with 64 are used in the 64-bit version of
Cintcode. For more details, see the description of loadseg on page 80.

tickspersecond CIN:y, POS:y, NAT:n

This constant no longer exists since time is now measured in milli-seconds (and dates
in days). In both Cintsys and Cintpos, delays measured in milli-seconds can be achieved
using delay(msecs) and delays until a specified absolute time can be done using
delayuntil(days, msecs). Under Cintpos, the clock device now takes packets that
specify absolute times (in days since 1 January 1970 and milli-second since midnight)
for their release. For example, sendpkt(notinuse, -1, 0, 0, 0, days, msecs) will
resume execution when the time specified by days and msecs is reached. The second
argument (-1) specifies the clock device.
54 CHAPTER 3. THE LIBRARY

timeoutch CIN:n, POS:y, NAT:n

This is the value returned by rdch when a timeout occurs while trying to read from
a stream. Its value is normally -2. Currently only TCP streams under Cintpos provide
the timeout mechanism.

ug CIN:y, POS:y, NAT:y

This constant specified the first Global variable available to user programs. Cur-
rently ug=200 so globals below this value are reserved for system use and the standard
library. Since ug may change it would be wise to use it.

3.2 Global Variables

This section describes the global variables declared in libhdr.h.

cis, cos CIN:y, POS:y, NAT:y

These are, respectively, the currently selected input and output streams. Zero
indicates that no stream is selected.

colist CIN:n, POS:y, NAT:n

This holds the list of currently existing coroutines.

consoletask CIN:n, POS:y, NAT:n

This is a variable used by command language interpreters.

currco CIN:n, POS:y, NAT:n

This points to the currently executing coroutine.

currentdir CIN:n, POS:y, NAT:n

This is a string holding the name of the current working directory.

globsize CIN:y, POS:y, NAT:y

This variable is in global zero and holds the size of the global vector. Its value is
normally 1000.

mainco busy CIN:n, POS:y, NAT:n

This is a variable used in the implementation of gomultievent under Cintpos.

multi count CIN:n, POS:y, NAT:n

This is a variable used in the implementation of gomultievent under Cintpos.

pktlist CIN:n, POS:y, NAT:n

Under Cintpos when running in multi-event mode, pktlist contains mapping from
packets to their corresponding coroutines.

randseed CIN:y, POS:y, NAT:y

This is the seed used by the random number generator randno.
3.3. GLOBAL FUNCTIONS 55

result2 CIN:y, POS:y, NAT:y

This global variable is used by some functions to return a second result.

returncode CIN:y, POS:y, NAT:n

This holds the return code of the command most recently executed by the command
language interpreter.

rootnode CIN:y, POS:y, NAT:n

This points to the rootnode.

start CIN:y, POS:y, NAT:y

This is global 1 and is, by convention, the main function of a program. It is the
first user function to be called when a program is run by the Command Language
Interpreter.

taskid CIN:n, POS:y, NAT:n

Under Cintpos this is the identifier of the currently executing task. It in not avail-
able under Cintsys.

tcb CIN:n, POS:y, NAT:n

Under Cintpos this is a pointer to the currently executing task.

userenv CIN:y, POS:y, NAT:y

This variable is available to the user to hold information that is preserved from
one CLI command to the next. The standard command language interpreter resets all
global variable from ug to the end of the global vector between commands. userenv is
not in this region of the global vector and so is preserved. Normally userenv is either
zero or points to a user defined structure holding environmental data.

3.3 Global Functions

One of the main purposes of the global vector is hold entry points of functions defined
in one module and used in a different module. This section describes the function
defined in the standard resident library. Most of these are defined in BCPL in the files:
sysb/klib.b, sysb/blib.b and sysb/dlib.b, one library (cin/syscin/syslib) is in
hand written Cintcode since it contains instructions that cannot be generated by the
BCPL compiler. The functions defined in syslib are sys, changeco and muldiv.
The standard library functions are described in alphabetical order.

abort(code) CIN:y, POS:y, NAT:n

This causes an exit from the current invocation of the interpreter, returning code
as the error code. If code is zero execution exits from the Cintcode system. If code
is -1 execution resumes using the faster version of the interpreter (fasterp). If code
is -2 the entire Cintcode memory is written to file DUMP.mem is a compacted form for
processing by CLI commands such as dumpsys or dumpdebug. If code is positive, under
normal conditions, the interactive debugger is entered.
56 CHAPTER 3. THE LIBRARY

res := appendstream(scb) CIN:y, POS:y, NAT:y

This function sets the position of stream scb to the end so that anything written
to the stream will be appended. The result is FALSE if scb is not an inout stream or
cannot be positioned for other reasons. It returns TRUE otherwise.

ch := binrdch() CIN:y, POS:y, NAT:y

This call behaves like rdch() but does not skip over carriage return (’*c’) char-
acters.

ch := binwrch(ch) CIN:y, POS:y, NAT:y

This call behaves like wrch(ch) but does treat ch as a special character and so
does not call deplete at the end of lines and does not insert carriage return (’*c’)
characters.

res := callco(cptr, arg) CIN:y, POS:y, NAT:y

This call suspends the current coroutine and transfers control to the corou-
tine pointed to by cptr. It does this by resuming execution of the function that
caused its suspension, which then immediately returns yielding arg as result. When
callco(cptr,arg) next receives control it yields the result it is given. The definition of
callco is in blib.b and is as follows.

LET callco(cptr, a) = VALOF

{ IF cptr!co_parent DO abort(110)
cptr!co_parent := currco
RESULTIS changeco(a, cptr)
}

callco always leaves the global currco is set to point to the target coroutine. This is
done by the Cintcode instruction CHGCO invoked by changeco.

res := callseg(name, a1, a2, a3, a4) CIN:y, POS:y, NAT:y

This function loads the compiled program from the file name, initialises its global
variables and calls start with the four arguments a1,...,a4. It returns the result of
this call, after unloading the program.

ch := capitalch(ch) CIN:y, POS:y, NAT:y

This function converts lowercase letters to uppercase, leaving other characters un-
changed.

res := changeco(val, cptr) CIN:y, POS:y, NAT:y

This function is used in the functions that implement the coroutine mechanism. In
callco, resumeco and cowait, it causes the current coroutine to become suspended
and store cptr in the global currco before giveing control to the specified coroutine.
Strangely, execution continues just after the call of changeco but with the P pointer
pointing to the stack frame of the function that caused the target coroutine to become
suspended. The call of changeco in each of callco, cowait and resumeco is imme-
diately followed by a RETURN statement which causes the corresponding function to
3.3. GLOBAL FUNCTIONS 57

return with result val. The only other use of changeco is in createco. This is more
subtle but can be understood by looking at the description of createco on page 58.

res := changepri(taskid, pri) CIN:n, POS:y, NAT:n

This Cintpos function attempts to change the priority of the specified task to pri.
It moves the specified task control block to its new position in the priority chain. If
the specified task is runnable and of higher priority than the current task, it is given
control leaving the current task suspended in RUN state. The result is non zero if
successful, otherwise it is zero with result2 set to 101 if taskid is invalid or to 102 if
the change would cause two tasks to have the same priority.

res := clihook(arg) CIN:y, POS:y, NAT:y

This function simply calls start(arg) and returns its result. Its purpose is to
assist debugging by providing a place to set a breakpoint in the command language
interpreter (CLI) just before a command in entered. Occassionally, a user may find it
useful to override the standard definition of clihook with a private version.

codewrch(code) CIN:y, POS:y, NAT:y

This routine uses wrch to write the Unicode character code as a sequence of bytes
in either UTF8 or GB2312 format. If the encoding field of the current output stream
is UTF8, the output is in UTF8 format as described in the following table.

Code range Binary value UTF8 bytes

0-7F zzzzzzz 0zzzzzzz
80-7FF yyyyyzzzzzz 110yyyyy 10zzzzzz
800-FFFF xxxxyyyyyyzzzzzz 1110xxxx 10yyyyyy 10zzzzzz
1000-1FFFFF wwwxxxxxxyyyyyyzzzzzz 11110www 10xxxxxx 10yyyyyy 10zzzzzz
etc etc etc

If the encoding field of the current output stream is GB2312, the output is in GB2312
format as described in the following table.

Decimal range GB2312 bytes

0 < dd < 127 <dd>
128 < xxyy < 9494 <xx+160> <yy+160>

res := compch(ch1, ch2) CIN:y, POS:y, NAT:y

This function compares two characters ignoring case. It yields -1 (+1) if ch1 is
earlier (later) in the collating sequence than ch2, and 0 if they are equal.

res := compstring(s1, s2) CIN:y, POS:y, NAT:y

This function compares two strings ignoring case. It yields -1 (+1) if s1 is earlier
(later) in the collating sequence than s2, and 0 if the strings are equal.
58 CHAPTER 3. THE LIBRARY

res := cowait(arg) CIN:y, POS:y, NAT:y

This call suspends the current coroutine and returns control to its parent by re-
suming execution of the function that caused its suspension, yielding arg as result.
When cowait(arg) next receives control it yields the result it is given. The definition
of cowait is in blib.b and is as follows.

LET cowait(a) = VALOF

{ LET parent = currco!co_parent
currco!co_parent := 0
RESULTIS changeco(a, parent)
}

cowait always leaves the global currco is set to point to the resumed coroutine. This
is done by the Cintcode instruction CHGCO invoked by changeco.

cptr := createco(fn, size) CIN:y, POS:y, NAT:y

BCPL uses a stack to hold function arguments, local variables and anonymous
results, and it uses the global vector and static variables to hold non-local quanitities.
It is sometimes convenient to have separate runtime stacks so that different parts of
the program can run in pseudo parallelism. The coroutine mechanism provides this
facility.
Coroutines have distinct stacks but share the same global vector, and it is natural
to represent them by pointers to their stacks. At the base of each stack there are six
words of system information as shown in figure 3.1.

resumption point

fn sz c P1 L1

coroutine chain suspended

stack frame
parent link
cptr

Figure 3.1: A coroutine stack

The resumption point is P pointer belonging to the function that caused the sus-
pension of the coroutine. It becomes the value of the P pointer when the coroutine
next resumes execution. The parent link points to the coroutine that called this one,
or is zero if the coroutine not active. The outermost coroutine (or root coroutine) is
marked by the special value -1 in its parent link. As a debugging aid, all coroutines
are chained together in a list held in the global colist. The values fn and sz hold the
main function of the coroutine and its stack size, and c is a private variable used by
the coroutine mechanism.
At any time just one coroutine (the current coroutine) has control, and all the
others are said to be suspended. The current coroutine is held in the global variable
currco, and the Cintcode P register points to a stack frame within its stack. Passing
3.3. GLOBAL FUNCTIONS 59

currco changeco
stack frame
resumption point

P1 L1 a

cptr
P CHGCO
resumption point

PC

P1 L1 a

PC currco P

Figure 3.2: The effect of changeco(a, cptr)

control from one coroutine to another involves saving the resumption point in the
current coroutine, and setting new values for the program counter (PC), the P pointer
and currco. This is done by changeco(a,cptr) as shown in figure 3.2. The function
changeco is defined by hand in syslib used by cintsys and cintpos and its body
consists of the single Cintcode instruction CHGCO. As can be seen its effect is somewhat
subtle. The only uses of changeco are in the definitions of createco, callco, cowait
and resumeco, and these are the only functions that cause coroutine suspension. In
the native code version of BCPL changeco is defined in mlib.s
The definition of createco is in blib.b and is as follows.

LET createco(fn, size) = VALOF

{ LET c = getvec(size+6)
UNLESS c RESULTIS 0
FOR i = 6 TO size+6 DO c!i := stackword
c!0 := c<<B2Wsh // resumption point
c!1 := currco // parent link
c!2 := colist // colist chain
c!3 := fn // the main function
c!4 := size // the coroutine size
c!5 := c // the new coroutine pointer
colist := c // insert into the list of coroutines
changeco(0, c)
c := fn(cowait(c)) REPEAT
}

The function createco creates a new coroutine by allocating its stack by the call
gevec(size+6). The variable c holds a pointer to the new coroutine stack and, as can
been seen, its first six words are initialised to hold system information, as follows.
60 CHAPTER 3. THE LIBRARY

c!0 resumption point

c!1 parent link
c!2 colist chain
c!3 fn – the main function
c!4 size – the coroutine size
c!5 c – the new coroutine pointer
The coroutine list colist is also set to c.
The call changeco(0, c) causes the P pointer to be set to c!0 which has been
initialied to the machine address of the base of the new coroutine stack. Execution
continues just after the call, namely at the REPEAT loop in the body of createco,
but in the coroutine environment of the newly created coroutine. The compiled code
for this loop will assume fn, size and c reside in positions 3, 4 and 5 relative to P, ie
in memory locations c!3, c!4 and c!5 and so execution behave as (naively) expected.
The first time cowait(c) is called in this loop, execution returns from createco with
the result c which is the pointer to the newly created coroutine.
When control is next transferred to this new coroutine, the value passed becomes
the result of cowait and hence the argument of fn. If fn(..) returns normally, its
result is assigned to c which is returned to the parent coroutine by the repeated call
of cowait. Thus, if fn is simple, a call of the coroutine convert the value passed, val
say, into fn(val). However, in general, fn may contain calls of callco, cowait or
resumeco, and so the situation is not always quite so simple.
To help understand the subtle effect of the call of changeco(0,c), look at figure 3.3
showing the state just after the call.

createco changeco
stack frame stack frame

P2 L2 fn sz c P1 L1 0 c

PC LP5 {
K9G 24 cowait(c)
LP3
K6 fn( ... )
SP5 c := ...
coroutine chain J -7 } REPEAT

fn sz c

colist
currco
The new coroutine
P

Figure 3.3: The state just after changeco(0,c) in createco

3.3. GLOBAL FUNCTIONS 61

devid := createdev(dcb) CIN:n, POS:y, NAT:n

This Cintpos function creates a device using the first available slot in devtab. The
device control block dcb must have already been initialised and linked to its device
driver. If successful it returns a negative value identifying the device. On failure it
returns zero with result2 set to 104 if the devtab is full, or to 106 if device initialisation
failed.

res := createtask(seglist, stsize, pri) CIN:n, POS:y, NAT:n

This Cintpos function creates a task using the first free slot in the task table. It
allocates space for the new task control block (TCB) and a copy of the specified segment
list, and initialises them both. It inserts the new TCB in priority chain of tasks and
returns the id of the newly created task if successful. It is left in DEAD state with no
stack or global vector and no packets on its work queue. If there is an error, it returns
zero with result2 set to 102 if there is already a task with priority pri, or to 103 if
there is insufficient memory or to 105 if the task table is full. A segment list is a small
vector whose zeroth element holds its upperbound and whose other elements hold lists
of sections of code typically loaded by loadseg.

datstamp(datv) CIN:y, POS:y, NAT:y

This sets datv!0 to the number of days since 1 January 1970, and datv!1 to the
number of milli-seconds since midnight, and for compatability with the older version
of datstamp datv!2=-1 indicating the new date and time format is being used.

dat to string(datv, v) CIN:y, POS:y, NAT:y

This call causes the time stamp in datv to be converted to three strings v, v+5 and
v+10. The string at v is set to the date in the form dd-mmm-yyyy. The string at v+5 is
set to the the current time in the form hh:mm:ss, and the string at v+10 is set to the
day of the week. The upper bound of v should be at least 14 to be safe. The time stamp
is typically obtained by a call of datstamp(datv) which sets datv!0 to the number of
days since 1 January 1970, datv!1 to the number of milli-seconds since midnight and
datv!2 to -1 indicting that the new date and time format is being used.
delay(msecs) CIN:y, POS:y, NAT:y
This call suspends execution for at least msecs milli-seconds. Under Cintpos, this is
achieved by sending a suitable packet to the clock device (using sendpkt) and waiting
for it to be returned.

delayuntil(days, msecs) CIN:y, POS:y, NAT:y

This call suspends execution until the specified date and time is reached. days
specifies the date as the number of days since 1 January 1970 and msecs is the number
of milli-seconds since midnight. Under Cintpos, the delay is achieved by sending a
suitable packet to the clock device (using sendpkt) and waiting for it to be returned.

deleteco(cptr) CIN:y, POS:y, NAT:y

This call takes a coroutine pointer as argument and, after checking that the corre-
sponding coroutine has no parent, deletes it by returning its stack to free store.
62 CHAPTER 3. THE LIBRARY

dcb := deletedev(devid) CIN:n, POS:y, NAT:n

This Cintpos function closes down the specified device and deallocates it device
identifier, but it does not return its device control block (DCB) to free store. It returns
any packets still on its work queue to the requesting tasks with both the pkt res1 and
pkt res2 fields set to -1. If successful, it returns the DCB of the deleted device. On
failure, it returns zero with result2 set to 101 indicating that devid was invalid. If any
of the released packets cause a higher priority task to become runnable, the control
passes to the highest priority one leaving the current task suspended in RUN state.
The clock device has identifier -1 and is permanently resident and cannot be deleted.

flag := deletefile(name) CIN:y, POS:y, NAT:y

This call deletes the named file, returning TRUE if successful, and FALSE otherwise.

res := deleteself(pkt, seg) CIN:n, POS:y, NAT:n

This Cintpos function first calls qpkt to return the packet if pkt is non zero, then
calls unloadseg(seg) if seg is non zero, before deleting the current task. This function
is defined in klib since it would be unsafe for it to be in a segment that may be
unloaded while it is being executes. It returns a non zero value if successful but, of
course, this value will never be seen! On failure, it return zero with result2 set to 108
indicating that the current task is not deletable.

res := deletetask(taskid) CIN:n, POS:y, NAT:n

This Cintpos function attempts to delete the specified task which must have an
empty work queue and be either the current task or in DEAD state. Its task control
block (TCB) is unlinked from the priority chain and removed from tasktab. Finally
its segment list and the TCB itself returned to free store. It returns a non zero value
if successful. On failure, it returns zero with result2 set to 101 if taskid is invalid, or
to 108 if the task is not deletable.

res := dqpkt(id, pkt) CIN:n, POS:y, NAT:n

This Cintpos function attempts to dequeue the given packet from the task or device
specified by id. If not found there, it may have already been returned to the current
task so its work queue is searched. The result is the id of the task or device whose
work queue contained the packet. If there is an error, the result is zero with result2
set to 101 for invalid id or 109 if the packet was not found. The id field of the packet
is set to the id of the task or device whose work queue contained the packet provided
that this is not the id of the current task.

endread() CIN:y, POS:y, NAT:y

This routine closes the currently selected input stream by calling endstream(cis).

endstream(scb) CIN:y, POS:y, NAT:y

This routine closes the stream whose control block is scb.

endwrite() CIN:y, POS:y, NAT:y

This routine closes the currently selected output stream by calling endstream(cos).
3.3. GLOBAL FUNCTIONS 63

scb := findappend(name) CIN:y, POS:y, NAT:y

This function opens an output stream specified by the file name name in append
mode causing all output to be appended onto the end of the file. If the file name is
relative and the prefix string is set, it is prepended to the name before attempting to
open the stream. If the file does not exist a zero length file of the given name is created.
If there is an error the result is zero.

n := findarg(keys, item) CIN:y, POS:y, NAT:y

The function findarg was primarily designed for use by rdargs but since it is
sometimes useful on its own, it is publicly available. Its first argument, keys, is a string
of keys of the form used by rdargs and item is a string. If the result is positive, it
is the argument number of the keyword that matches item, otherwise the result is -1.
During matching all letters are converted to uppercase, but this convention may change
in future.

scb := findinput(name) CIN:y, POS:y, NAT:y

This function opens an input stream. If name is the string "*" then it opens the
standard input stream which is normally from the keyboard, otherwise name is taken
to be a device or file name. If the file name is relative and the prefix string is set, it
is prepended to the name before attempting to open the stream. If the stream cannot
be opened the result is zero. See Section 3.3.2 for information about the treatment of
filenames.

scb := findinoutput(name) CIN:y, POS:y, NAT:y

This function opens a stream specified by the device or file name name that can
be used for both input and output. If name is the string "*" then output is normally
to the screen and input comes from the keyboard. If the file name is relative and the
prefix string is set, it is prepended to the name before attempting to open the stream.
If the stream cannot be opened, the result is zero. See Section 3.3.2 for information
about the treatment of filenames.

scb := findoutput(name) CIN:y, POS:y, NAT:y

This function opens an output stream specified by the device or file name name. If
name is the string "*" then it opens the standard output stream which is normally to
the screen. If the file name is relative and the prefix string is set, it is prepended to
the name before attempting to open the stream. If the stream cannot be opened, the
result is zero. See Section 3.3.2 for information about the treatment of filenames.

res := get record(v, recno, scb) CIN:y, POS:y, NAT:y

This attempts to read the record numbered recno from the file whose stream control
block is scb into the vector v. The record length must have been set already by a call
of setrecordlength. If get record is successful it returns TRUE, otherwise it returns
FALSE possibly because the end of file was reached before the whole record had been
read.

v := getlogname(logname) CIN:y, POS:y, NAT:y

This function searches the list of logical variables held in the root node and returns
its value if found, otherwise it returns zero.
64 CHAPTER 3. THE LIBRARY

v := getvec(upb) CIN:y, POS:y, NAT:y

This function allocates space using a first fit algorithm based on a list of blocks
chained together in memory order. Word zero of each block in the chain contains a flag
in its least significant bit indicating whether the block is allocated or free. The rest of
the word is an even number giving the size of the block in words. A pointer to the first
block in the chain is held in the rootnode.
getvec allocates a vector with upper bound upb from the first large enough free
block on the block list. If no such block exists it returns zero. A vector previously
allocated by getvec can be freed by the above call of freevec. Coalescing of adjacent
free blocks is performed by getvec.
An extra word is allocated just before the start of each block to hold its size, and
four or five words are added to the end of each block and filled with special data that
is checked when the block is returned to free store. This catches many common space
allocation errors.

res := globin(segl) CIN:y, POS:y, NAT:y

This function initialises the global variables defined in the list of program modules
given by its argument segl. It returns zero if the global vector was too small, otherwise
it returns segl.

res := hold(taskid) CIN:n, POS:y, NAT:n

This Cintpos function sets the HOLD bit in the task control block of the specified
task. It returns a non zero value if successful. If there is an error, it returns zero with
result2 set to 101 if taskid was invalid, and 110 if the specified task was already in
HOLD state. If the task holds itself control is given to next lower priority runnable
task.

cptr := initco(fn, size,a,b,c,d,e,f,g,h,i,j,k) CIN:y, POS:y, NAT:y

This function provides a convenient method of creating and initialising coroutines.
It definition is as follows:

LET initco(fn, size, a, b, c, d, e, f, g, h, i, j, k) = VALOF

{ LET cptr = createco(fn, size)
result2 := 0
IF cptr DO result2 := callco(cptr, @a)
RESULTIS cptr
}

A coroutine with main function fn and given size is created and, if successful, it is
initialised by callco(cptr, @a). Thus, fn should expect a vector containing up to 11
elements. Once the newly created coroutine has initialised itself, it returns control to
initco by means a call of cowait. The result of initco is the newly created coroutine
pointer, or zero on failure. The second result (in result2) is the value returned by the
first call of cowait in the newly created coroutine.

scb := input() CIN:y, POS:y, NAT:y

This function returns cis, the SCB of the currently selected input stream.
3.3. GLOBAL FUNCTIONS 65

count := instrcount(fn,a,b,c,d,e,f,g,h,i,j,k) CIN:y, POS:y, NAT:n

This function returns the number of Cintcode instructions executed when evaluating
the call: fn(a,b,c,d,e,f,g,h,i,j,k).
Counting starts from the first instruction of the body of fn and ends when its
final RTN instruction is executed. Thus when f was defined by LET f(x) = 2*x+1,
the call instrcount(f, 10) returns 4 since its body executes the four instructions:
L2; MUL; A1; RTN. The value returned by fn(a,b,c,d,e,f,g,h,i,j,k) is saved by
instrcount in the global variable result2.

flag := intflag() CIN:y, POS:y, NAT:n

This function provides a machine dependent test to determine whether the user is
asking to interrupt the normal execution of a program.

p := level() CIN:y, POS:y, NAT:y

This call returns the current stack frame pointer for use in a later call of longjump.

segl := loadseg(name) CIN:y, POS:y, NAT:n

This function calls sys(Sys loadseg, name) to loads the specified compiled pro-
gram into memory. See Sys loadseg on page 80 for details.

longjump(P , L) CIN:y, POS:y, NAT:y

This call causes execution to resume at label L in the body of a function or routine
that owns the stack frame given by P that must have been obtained by a previous call of
level. Jumps may only be used to points within the current coroutine. Jumps to labels
within the current function or routine can be performed using the GOTO command, so
level and longjump are only needed for non local jumps.

res := memoryfree() CIN:y, POS:y, NAT:n

This function checks that the free store chain is valid, outputting a error message
and calling abort(999) if not. If the chain is valid, it returns the current number of
unused words, and sets result2 to the memory size. This function can assist debugging
and helps with the detection of space leaks.

obj := mkobj(upb,fns,a,b,c,d,e,f,g,h,i,j,k) CIN:y, POS:y, NAT:y

This function creates and initialises an object. It definition is as follows:

LET mkobj(upb, fns, a, b, c, d, e, f, g, h, i, j, k) = VALOF

{ LET obj = getvec(upb)
IF obj DO
{ !obj := fns
InitObj#(obj, @a) // Send the init message to the object
}
RESULTIS obj
}

As can be seen, it allocates a vector for the fields of the object, initialises its
zeroth element to point to the methods vector and calls the initialisation method that
is expected to be in element InitObj of fns. The result is a pointer to the initialised
66 CHAPTER 3. THE LIBRARY

fields vector. If it fails, it returns zero. As can be seen the initialisation method receives
a vector of up to 11 initialisation arguments.

res := muldiv(a, b, c) CIN:y, POS:y, NAT:y

The result is the value obtained by dividing c into the double length product of a
and b, the remainder of this division is left in the global variable result2. The result
is undefined if it is too large to fit into a single length word or if c is zero. The result
is also undefined if any of a, b or c is the largest negative integer.
This version of muldiv is defined in the hand written Cintcode library syslib and
invokes the MDIV Cintcode instruction which is implemented efficiently. The older ver-
sion is invoked by sys(Sys muldiv,a,b,c) and uses binary long division implemented
in C. Both versions are believed to produce identical results except possibly when c=0.
As an example, the function defined below calculates the cosine of the angle between
two unit vectors in three dimensions using scaled integers to represent numbers with 6
digits after the decimal point.

MANIFEST { Unit=1000000 } // Scaling factor for numbers of the

// form ddd.dddddd
FUN inprod(v, w) = muldiv(v!0, w!0, Unit) +
muldiv(v!1, w!1, Unit) +
muldiv(v!2, w!2, Unit)

Remember that scaled fixed point values can be output conveniently using writef
as in:

writef("%10.6d*n", 123_456789)

which will output the following:

123.456789

newline() CIN:y, POS:y, NAT:y

This simply outputs the newline character (’*n’) to the currently selected output
stream.

newpage() CIN:y, POS:y, NAT:y

This simply outputs the newline character (’*p’) to the currently selected output
stream.

res := note(scb, posv) CIN:y, POS:y, NAT:y

If scr is a file stream, this function sets posv!0 and posv!1 to the current block
number and position within that block. For RAM streams, posv!0 and posv!1 are set
to zero and the position within the stream buffer. The result is TRUE if scb is a file or
RAM stream, and FALSE otherwise.

scb := output() CIN:y, POS:y, NAT:y

This function returns cos, the SCB of the currently selected output stream.
3.3. GLOBAL FUNCTIONS 67

scb := pathfindinput(name, pathname) CIN:y, POS:y, NAT:y

This function opens an input stream. If name is the string "*" then input comes
from standard input which is normally the keyboard, otherwise name is taken to be
a filename. If name is a relative file name and pathname is non zero, the directories
specified by the shell variable pathname are searched. The directories specified by the
shell variable are separated by either semicolons or colons, although under Windows
only semicolons are allowed. If the prefix string is non null and the filename, possibly
prefixed by a directory name, is relative then the prefix string is prepended before the
file is opened. If the file cannot be opened pathfindinput returns zero.

res := point(scb, posv) CIN:y, POS:y, NAT:y

This function sets the position of stream scb to that specified in posv whose elements
are scb!0 the block number and scb!1 the byte position within the block. If the new
position is in a different block the contents of the buffer buffer may have to be written
out and data from the new block read in. point may therefore fail if the stream was
not opened using findinput or findinoutput. It returns TRUE if successful, even if
positioned just after the last block of the file, ie block=lblock+1 and pos=end=0. It
returns FALSE, otherwise, possibly because the stream is not pointable or the posv is
out of range. It is advisable to test the result of point every time it is used.
For RAM streams posv!0 should be zero and posv!1 should be a position in the
buffer (which is entirely held in RAM).

res := put record(v, recno, scb) CIN:y, POS:y, NAT:y

This attempts to write a record numbered recno to the file whose stream control
block is scb taking data from the vector v. The record length must have been set already
by a call of setrecordlength. If put record is successful it returns TRUE, otherwise it
returns FALSE. If the last record of a file has number n, it is permissible to extend the
file by writing record n+1, but not one with a larger record number.

res := qpkt(pkt) CIN:n, POS:y, NAT:n

This Cintpos function queues the given packet on the end of the work queue of the
destination task or device (specified by pkt id!pkt). If this field is positive it refers to
a task, if it is -1 it refers to the clock device and other negative values refer to other
devices. If the packet is queued successfully this field is updated to hold the current
task’s identifier and the result is non zero, otherwise the result is zero with result2
set to 101 if the destination id is invalid, and to 111 if pkt link was not equal to
notinuse (=-1). If the destination was a runnable task of higher priority than the
current one, then the current task immediately becomes suspended in RUN state and
control is given to the destination, otherwise the current task continues to run normally.
Interaction with the resident Cintpos devices is described in Chapter 6.

n := randno(upb) CIN:y, POS:y, NAT:y

This function returns a random integer in the range 1 to upb. It uses a seed held
68 CHAPTER 3. THE LIBRARY

in global variable randseed which can be set using setseed described below. Its
implementation is as follows:

LET randno(upb) = VALOF

{ randseed := randseed*2147001325 + 715136305
RETURN ABS(randseed/3) MOD upb + 1
}

res := rdargs(keys, argv, upb) CIN:y, POS:y, NAT:y

This implementation of BCPL incorporates a command language interpreter which
is described in Chapter 4. Most commands require arguments and these are easily read
using rdargs.
The first argument (keys) specifies the argument format. The second and third
arguments provide a vector (argv) with a given upper bound (upb) into which the
decoded arguments will be placed. If rdargs is successful, it returns the number of
words used in argv to represent the decoded command arguments, but on failure, it
returns zero.
The string keys holds the list of argument keywords separated by commas (,).
Alternative keywords for a given argument are separated by equal signs (=). The
expected number of arguments is one more than the number of commas in the key
string. If rdargs returns successfully, this number of elements at the start of argv will
hold the decoded arguments.
Arguments can have qualifiers of the form /A, /K, /N, /S and /P. The qualifier
letters can be in either upper or lower case. The qualifier /A means that the argument
must be given. /K means that, if the argument is given, it must include its keyword.
/N specifies that the argument must be a number. /S indicates that the argument
is a switch parameter set to TRUE by its keyword. /P indicates that a prompt will
be given for the argument if it has not already been set. Prompting only happens if
the currently selected input and output streams are both connected to an interactive
terminal. If the prompt is for a switch argument (/S) it expects a yes/no response.
Typing yes or y is treated as yes, any other response is treated as no. If rdargs
returns successfully argv!0, argv!1 etc will hold the arguments settings. A setting of
zero means the argument was not given. A setting of -1 means the argument was a
switch set the TRUE. Otherwise, if /N was specified the setting will point to a word in
argv where the decoded integer is stored. If a /N was not specified, the setting will be
a BCPL string with its characters packed into argv. Note that an argument should not
have both /N and /S specified.
Command arguments are read from the currently selected input stream using a
decoding mechanism that permits both positional and keyed arguments to be freely
mixed. A typical use of rdargs occurs in the source of the input command as follows:

UNLESS rdargs("FROM/A,TO=AS/K,DATA/N/P,N/S", argv, 50) DO

{ writef("Bad arguments for: FROM/A,TO=AS/K,DATA/N/P,N/S*n")
...
}
3.3. GLOBAL FUNCTIONS 69

In this example, there are four possible arguments and their values will be placed in
the first four elements of argv. The first argument has keyword FROM and must receive
a value because of the qualifier /A. The second has alternative keywords TO and AS with
qualifier /K that insists the argument is introduced by one of its keywords. The third
argument has the qualifiers /N and /P indicating that it expects a number and that it
will be prompted for if not already given, and the last argument has the qualifier /S
indicating that it is a switch that can be set by the presence of its keyword.
Table 3.4 shows the values in placed in argv and the result when the call:
rdargs("FROM/A,TO=AS/K,DATA/N/P,N/S", argv, 50)
is given various argument strings. This example illustrates that keyword synonyms can
be defined using = within the key string. Positional arguments are those not introduced
by keywords. When one is encountered, it becomes the value of the lowest numbered
unset non-switch argument.

Arguments argv!0 argv!1 argv!2 argv!4 Result

abc TO xyz "abc" "xyz" 0 0 ~=0
to xyz from abc "abc" "xyz" 0 0 ~=0
as xyz abc n "abc" "xyz" 0 -1 ~=0
abc xyz - - - - =0
"from" to "to" "from" "to" 0 0 ~=0
abc data 123 to "to" "abc" "to" ->123 0 ~=0
data 123 to junk - - - - =0

Figure 3.4: rdargs("FROM/A,TO=AS/K,DATA/N/P,N/S", argv, 50)

To consolidate your understanding of rdargs, try compiling and running the pro-
gram: bcplprogs/tests/tstrdargs.b.

res := rdargs2(keys1, keys2, argv, upb) CIN:y, POS:y, NAT:y

This function behaves just like rdargs, specified above, except it uses key data that
is the concatenation of strings keys1 and keys2 thus allowing the key data to have up
to than 510 characters.

ch := rdch() CIN:y, POS:y, NAT:y

This call reads the next character from the currently selected input stream. If the
stream is exhausted, it returns the special value endstreamch. Input from the keyboard
is buffered until the ENTER (or RETURN) key is pressed to allow simple line editing
in which the backspace key may be used to delete the most recent character typed. See
Section 3.3.1 for more detailed information.

kind := rditem(v, upb) CIN:y, POS:y, NAT:y

This function is usually called from rdargs to read an item from the currently
selected input stream. After ignoring leading spaces and tabs, it packs the item into
70 CHAPTER 3. THE LIBRARY

the vector v whose upper bound is upb and returns an integer describing the kind of
item read. Table 3.5 gives the kinds of item that can be read and corresponding item
codes.

Example items Kind of item Item code

= 5
; 4
carriage return 3
"from"
"*ntwo words*n" Quoted string 2
abc
123-45*6 Unquoted string 1
end-of-stream Terminator 0
An error -1

Figure 3.5: rditem results

Within quoted strings *n represents the newline character, *s represents a space,

** represents an asterisk and *" represents a double quote character.

n := readflt() CIN:y, POS:y, NAT:y

This reads an optionally signed floating point number from the currently selected
input stream. Leading spaces, tabs and newlines are ignored. If the number is syn-
tactically correct, it returns its value with result2 set to zero, otherwise it returns
zero with result2 set to -1. In either case, it uses unrdch to replace the terminating
character.

n := readn() CIN:y, POS:y, NAT:y

This reads an optionally signed decimal integer from the currently selected input
stream. Leading spaces, tabs and newlines are ignored. If the number is syntactically
correct, it returns its value with result2 set to zero, otherwise it returns zero with
result2 set to -1. In either case, it uses unrdch to replace the terminating character.

res := recordnote(scb) CIN:y, POS:y, NAT:y

This call returns the number of the record containing the character pointed to by
the file position pointer of stream scb. The record length must have already been set
by a call of setrecordlength. The result is -1 if the stream is not suitable.

res := recordpoint(scb, recno) CIN:y, POS:y, NAT:y

This call sets the file position pointer of stream scb to point to the first byte of the
record whose number is recno. The record length must have already been set by a call
of setrecordlength. It returns TRUE if successful and FALSE otherwise.

res := release(taskid) CIN:n, POS:y, NAT:n

This Cintpos function will clear the HOLD bit in the specified task thus making it
potentially runnable. It returns a non zero value if successful. If the specified task does
3.3. GLOBAL FUNCTIONS 71

not exist it returns zero with 101 in result2. If the released task has higher priority
and is runnable it gains control leaving the current task suspended in RUN state. This
function is also called unhold.

flag := renamefile(oldname, newname) CIN:y, POS:y, NAT:y

The call renames the file oldname as file newname, deleting newname if necessary,
returning TRUE if the renaming was successful, and FALSE otherwise. Both oldname
and newname are strings.

res := resumeco(cptr, arg) CIN:y, POS:y, NAT:y

The effect of resumeco is almost identical to that of callco, differing only in the
treatment of the parent. With resumeco the parent of the calling coroutine becomes
the parent of the called coroutine, leaving the calling coroutine suspended and without
a parent. Systematic use of resumeco reduces the number of coroutines having parents
and hence allows greater freedom in organising the flow of control between coroutines.
The definition of resumeco is in blib.b and is as follows.

LET resumeco(cptr, a) = VALOF

{ LET parent = currco!co_parent
currco!co_parent := 0
IF cptr!co_parent DO abort(111)
cptr!co_parent := parent
RESULTIS changeco(a, cptr)
}

resumeco always leaves the global currco is set to point to the resumed coroutine.
This is done by the Cintcode instruction CHGCO invoked by changeco.

res := rewindstream(scb) CIN:y, POS:y, NAT:y

This function set the position of stream scb to its start, returning TRUE if successful,
and FALSE otherwise.

ch := sardch() CIN:y, POS:y, NAT:y

This function calls sys(Sys sardch) to read the next character from the keyboard
as soon as it is available, echoing the character to the screen.

sawrch(ch) CIN:y, POS:y, NAT:y

This function calls sys(Sys sawrch, ch) to write the specified character to the
screen.

sawritef(format,a,b,c,d,e,f ,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z)
CIN:y, POS:y, NAT:y
This function is similar to writef but performs its output using sawrch.

selectinput(scb) CIN:y, POS:y, NAT:y

This call executes cis := scb to select scb as the current input stream. It aborts
(with code 186) if scb is not an input stream.
72 CHAPTER 3. THE LIBRARY

selectoutput(scb) CIN:y, POS:y, NAT:y

This routine selects scb as the currently selected output stream. It aborts (with
code 187) if scb is not an output stream.

res := setbit(bitno, bitvec, state) CIN:y, POS:y, NAT:y

This function sets the specified bit in bitvec to 1 or 0 depending on whether state
is TRUE or FALSE, respectively. It returns a non-zero value if and only if the previous
setting of the bit was a one. See testbit below.

res := setflags(taskid, flags) CIN:n, POS:y, NAT:n

This Cintpos function sets the specified flags in the task control block of the specified
task. If successful it returns a non zero value with result2 set to the previous setting
of the flags field, otherwise it returns zero with result2 set to 101 indicating that
taskid was invalid. For more information about flags see testflags described below.

oldlogname := setlogname(logname, logvalue) CIN:y, POS:y, NAT:y

This sets the value of logical variable logname to the logvalue. By convention
logvalue should be a string. The list of logical name-value pairs is held in the root
node.

prevseed := setseed(newseed) CIN:y, POS:y, NAT:y

The current seed can be set to newseed by the call setseed(newseed). This function
returns the previous seed value.

sxpushval(sxv, val) CIN:y, POS:y, NAT:y

This pushes value val into the self expanding vector sxv. sxv points to the two word
control block of the self expanding vector. Initially both elements must be zero. When
non empty sxv!1 will be a vector, v say, containing the elements with v!0 will be the
subscript of the latest element in v. sxv!0 holds the upper bound of v. When the self
expanding vector needs more space, it allocates a new vector v using newvec freeing the
previous one after copying its elements into the new one. Clearly pointers to elements
of v may become invalid after any call of sxpushval. When the self expanding vector
is no longer needed, freevec(v) must be called.

srchwk(tcb) CIN:n, POS:y, NAT:n

This function is the Cintpos scheduler which is normally only called from within
one of the klib library functions or from the interrupt service routine. Its argument
points to the highest priority task control block that could possibly run. It searches
down the priority chain from this point until it finds the highest priority runnable task.
After setting the globals tcb and taskid appropriately, it gives this task control using
a call of sys(Sys rti,...).

res := stackfree(hwm) CIN:y, POS:y, NAT:n

If hwm=TRUE, this function returns the number of unused stack words above the high
water mark, otherwise it returns the number of words between the current stack frame
pointer and the end of stack. In either case it sets result2 to the stack size.
3.3. GLOBAL FUNCTIONS 73

code := start(a1, a2, a3, a4) CIN:y, POS:y, NAT:y

This function is, by convention, the main function of a program. If it is called from
the command language interpreter (see section 4), its first argument is zero and its
result should be the command completion code; however, if it is the main function of
a module run by callseg, defined below, then it can take up to 4 arguments and its
result is up to the user. By convention, a command completion code of zero indicates
successful completion and larger numbers indicate errors of ever greater severity

res := stepstream(scb, n) CIN:y, POS:y, NAT:y

This function advances the position of stream scb by n words, returning TRUE if
successful, and FALSE otherwise.

stop(code, reason) CIN:y, POS:y, NAT:y

This function is provided to stop the execution of the current command running
under control of the CLI. The arguments code and reason are placed in the CLI globals
cli returncode and cli result2 where they can be inspected by commands such as
if and why.

n := str2numb(str) CIN:y, POS:y, NAT:y

This function converts the string str into an integer. Characters other than 0 to
9 and - are ignored. The result is negative or zero if str%1=’-’. This function is no
longer recommended, string to number should be used instead.

n := string to number(str) CIN:y, POS:y, NAT:y

This attempts to set result2 to the integer represented by the string str. It returns
TRUE is successful and FALSE otherwise. The following are examples of acceptable
strings: "’A’", "123", "-99", "+63", "#377", "-#x7FF" and "+#b1011011".

res := sys(op,...) CIN:y, POS:y, NAT:y

The file sysc/cintsys.c contains the main program of the Cintsys system. It
also includes the definition of an important function dosys which provide access to
I/O operations and many other operating system primitives. The file sysc/cinterp.c
contains a C implementation of the Cintcode interpreter. With different compile time
settings this file can generate a faster version by reducing the number of debugging aids
present. Sometimes there is an even faster version of the interpreter implemented in
assembly language, see, for instance, sysasm/linux/cintasm.s. The BCPL function
sys provides an interface between BCPL and dosys.
The file sysc/cintpos.c contains the main program of the Cintpos system. It has
much is common with sysc/cintsys.c including the function dosys.
The sys function is defined by hand in cin/syscin/syslib and just invokes the
SYS Cintcode instruction. When SYS is encountered by the interpreter, it normally
just calls dosys passing the BCPL P and G pointers as arguments. But certain sys
operations such as sys(Sys quit,code) are processed directly by the interpreter.
As might be expected there are many sys operations concerned with interrupts
that are only available under Cintpos.
74 CHAPTER 3. THE LIBRARY

res := sys(Sys buttons) CIN:y, POS:y, NAT:y

On non standard machines such as the GP2X gaming machine there are buttons
that can be pressed. This call returns a bit pattern indicating which buttons are
currently pressed.

res := sys(Sys callc, fno, a1, a2 ...) CIN:y, POS:y, NAT:y

This makes the call cfuncs(args, g) where cfuncs is a C function defined in
sysc/cfuncs.c. The argument args points to memory locations holding fno, a1, a2,
etc., and g points to the base of the global vector.
The following table summarises the callc operations currently available (when
running under Linux).

res := sys(Sys callc, c name2ipaddr, a1) CIN:y, POS:y, NAT:y

The name or dotted decimals of a host is given in a1 and the result is its IP address
or -1 if there is an error.

res := sys(Sys callc, c name2port, a1) CIN:y, POS:y, NAT:y

The name or decimals of a port is given in a1 and the result is its IP address or -1
if there is an error.

res := sys(Sys callc, c newsocket) CIN:y, POS:y, NAT:y

The result is the file descriptor of a new socket or -1 if there is an error.

res := sys(Sys callc, c reuseaddr, a1, a2) CIN:y, POS:y, NAT:y

The file descriptor of a socket is given in a1. Id a2=1 the specified socket may be
reused. If there is an error the result is -1.

res := sys(Sys callc, c setsndbufsz, a1, a2) CIN:y, POS:y, NAT:y

This sets the send buffer size of socket a1 to a2 bytes. If there is an error the result
is -1.

res := sys(Sys callc, c setrcvbufsz, a1, a2) CIN:y, POS:y, NAT:y

This sets the receive buffer size of socket a1 to a2 bytes. If there is an error the
result is -1.

res := sys(Sys callc, c bind, a1, a2, a3) CIN:y, POS:y, NAT:y
This bind socket a1 to remote IP address a2 and remote port a3. If there is an
error the result is -1.

res := sys(Sys callc, c tcpconnect, a1, a2, a3) CIN:y, POS:y, NAT:y
This make a TCP/IP connection through socket a1 to remote IP address a2 and
remote port a3. If there is an error the result is -1.

res := sys(Sys callc, c tcplisten, a1, a2) CIN:y, POS:y, NAT:y

This causes socket a1 to wait for a TCP/IP connection to be requested by a remote
host. The maximum number of connections waiting to be accepted is given in a2. If
there is an error the result is -1.
3.3. GLOBAL FUNCTIONS 75

res := sys(Sys callc, c tcpaccept, a1) CIN:y, POS:y, NAT:y

This accepts a TCP/IP connection through socket a1. The result is the socket
number to be used for this connection or -1 if there is an error.

res := sys(Sys callc, c tcpclose, a1) CIN:y, POS:y, NAT:y

This closes socket a1. The result is -1 if there is an error.

res := sys(Sys callc, c fd zero, a1) CIN:y, POS:y, NAT:y

This clear every bit in the bit vector a1. The result is -1 if there is an error.

res := sys(Sys callc, c fd set, a1, a2) CIN:y, POS:y, NAT:y

This sets bit a1 in the bit vector a2. The result is -1 if there is an error.

res := sys(Sys callc, c fd isset, a1, a2) CIN:y, POS:y, NAT:y

This inspects bit a1 in the bit vector a2. The result is 1 if the bit was set and 0
otherwise.

res := sys(Sys callc, c fd select, a1, a2, a3, a4, a5) CIN:y, POS:y, NAT:y
This inspects bit a1 in the bit vector a2. The result is 1 if the bit was set and 0
otherwise.
The number of the bits to test is in a1. The bit vector identifying read sockets of
interest is in a2, The bit vector identifying write sockets of interest is in a3, The bit
vector identifying other sockets of interest is in a4. A pointer to two words holding
the timeout in seconds and microseconds is in a5. The result is the number of sockets
that can now be read or written to, or 0 if the timeout period has elapsed before any
sockets are ready. A result of -1 indicate an error.

res := sys(Sys callnative, f, a1, a2, a3) CIN:y, POS:y, NAT:y

This function is used to enter a subroutine in native machine code.

res := sys(Sys close, fp) CIN:y, POS:y, NAT:y

This closes the file whose file pointer is fp. It return 0 if successful.

res := sys(Sys cputime) CIN:y, POS:y, NAT:y

This returns the CPU time in milliseconds since the Cintcode system was entered.

res := sys(Sys datstamp, datv) CIN:y, POS:y, NAT:y

This sets datv!0 to the number of days since 1 January 1970, and datv!1 to the
number of milli-seconds since midnight, and for compatability with the older version
of datstamp datv!2=-1 indicating the new date and time format is being used.

res := sys(Sys delay, msecs) CIN:y, POS:y, NAT:y

In both Cintsys and Cintpos this call suspends Cintcode execution until the time
period has elapsed. It is normally better to use the library functions delay(msecs) or
delayuntil(days, msecs).
76 CHAPTER 3. THE LIBRARY

res := sys(Sys deletefile, name) CIN:y, POS:y, NAT:y

This deletes the file whose name is given by name. See page 92 for information
about the treatment of file names.

res := sys(Sys devcom, com, arg) CIN:n, POS:y, NAT:n

This is used in Cintpos to send commands from the interpreter thread to Cintpos
device threads.

res := sys(Sys dumpmem, context) CIN:y, POS:y, NAT:y

This call will dump the whole of Cintcode memory to the file DUMP.mem in a com-
pacted form that is typically inspected by either the commands dumpsys or dumpdebug.
By convention, context = 1 if SIGINT has been received, context = 2 if SIGSEGV
has been received, context = 3 if the dump was caused by BOOT detecting a fault,
context = 4 if the dump by the user call sys(Sys quit, -2), context = 5 if the
dump by a non zero return code from the interpreter, context = 6 if the dump by the
D command in the interactive debugger.

res := sys(Sys filemodtime, name, datv) CIN:y, POS:y, NAT:y

This sets the elements of the time stamp vector datv to represent the date and
time of the last modification of the file given by name returning TRUE if successful.
The first element datv!0 holds the number of days since 1 January 1970, datv!1 is the
number of milli-seconds since midnight and datv!2=-1 indicating that the new date
format is being used. If the file does not exist the call returns FALSE and setting the
three elements of datv to 0, 0 and -1, respectively.

res := sys(Sys filesize, fd) CIN:y, POS:y, NAT:y

This call return the size in bytes of the currently opened disk file whose file descrip-
tor is fd. The file descriptor is typically obtained by the expression scb!scb fd.

res := sys(Sys flt, op ,... ) CIN:y, POS:y, NAT:y

This call provides all the floating point operations available to BCPL. The required
operation is specified by op normally using a manifest constant (declared in libhdr)
such as fl mk, fl add or fl sin. All such operations are described below. BCPL
floating point numbers must fit in BCPL words and so are typically only 32 bits long
causing their precision and range to be somewhat limited. On 64-bit implementations
of BCPL, floating point numbers are much more precise.

res := sys(Sys flt, fl avail) CIN:y, POS:y, NAT:y

This call attempts returns -1 if all the Sys flt operations are available. It otherwise
return zero.

res := sys(Sys flt, fl mk, a, e) CIN:y, POS:y, NAT:y

This call attempts to return a floating point approximimation to the number a×10e
where a and e are signed integers.

res := sys(Sys flt, fl unmk, a) CIN:y, POS:y, NAT:y

This call decomposes the floating point number a returning the signed integer
3.3. GLOBAL FUNCTIONS 77

mantissa and leaving the decimal exponent in result2. For example, sys(Sys flt,
fl unmk, 1234.5678) might return 12345678 leaving -4 in result2. However, the
result may vary depending on the BCPL word length and the floating point represen-
tation used.

res := sys(Sys flt, fl float, a)

res := sys(Sys flt, fl fix, a) CIN:y, POS:y, NAT:y
The first call returns a floating point approximation of the integer a, and the second
attempts to return the closest integer to the floating point number a.

res := sys(Sys flt, fl abs, a)

res := sys(Sys flt, fl pos, a)
res := sys(Sys flt, fl neg, a)
res := sys(Sys flt, fl mul, a, b)
res := sys(Sys flt, fl div, a, b)
res := sys(Sys flt, fl add, a, b)
res := sys(Sys flt, fl sub, a, b) CIN:y, POS:y, NAT:y
The first three calls return, respectively, the absolute value of a, the value of a and
the negated value of a where a is a floating point number. The last four calls perform
floating point multiplication, division, addition and subtraction on their arguments.

res := sys(Sys flt, fl eq, a, b)

res := sys(Sys flt, fl ne, a, b)
res := sys(Sys flt, fl ls, a, b)
res := sys(Sys flt, fl gr, a, b)
res := sys(Sys flt, fl le, a, b)
res := sys(Sys flt, fl ge, a, b) CIN:y, POS:y, NAT:y
These six calls return TRUE if the corresponding floating point comparisons are
satisfied. Otherwise the result is FALSE.

res := sys(Sys flt, fl acos, a)

res := sys(Sys flt, fl asin, a)
res := sys(Sys flt, fl atan, a) CIN:y, POS:y, NAT:y
These calls return floating point approximations to the arc cosine, arc sine and arc
tangent of em a. The argument a is in radians and for acos the result is between 0
and π. For asin and atan it is between −π/2 and π/2.

res := sys(Sys flt, fl atan2, y, x) CIN:y, POS:y, NAT:y

This call return the angle in radians between x-axis and the line from the origin to
the point with cartesian coordinates (x, y). The result lies between −π and π.

res := sys(Sys flt, fl cos, a)

res := sys(Sys flt, fl sin, a)
res := sys(Sys flt, fl tan, a) CIN:y, POS:y, NAT:y
These calls return the cosine, sine and tangent of a.
78 CHAPTER 3. THE LIBRARY

res := sys(Sys flt, fl cosh, a)

res := sys(Sys flt, fl sinh, a)
res := sys(Sys flt, fl tanh, a) CIN:y, POS:y, NAT:y
These calls return the hyperbolic cosine, sine and tangent of a.

res := sys(Sys flt, fl exp, a)

res := sys(Sys flt, fl log, a)
res := sys(Sys flt, fl log10, a) CIN:y, POS:y, NAT:y
The first call returns an approximation to ea where e is the base of natural loga-
rithms. The second call return the natural logarithm of a, and the third call returns
log to the base 10 of a.

res := sys(Sys flt, fl frexp, a)

res := sys(Sys flt, fl ldexp, f, n) CIN:y, POS:y, NAT:y
The first call splits a floating-point number (a) into a fraction (f ) and exponent (n)
such that a is approximately equal to f × 2n . If possible the absolute value of f will be
between 0.5(inclusive) and 1.0(exclusive). The call returns f and stores n in result2
as an integer.. The second call is the inverse of frexp returning an approximation to
f × 2n .

res := sys(Sys flt, fl modf, a)

res := sys(Sys flt, fl mod, x, y) CIN:y, POS:y, NAT:y
The first call returns the fractional part (f ) of a storing the integer part (i) as a
floating-point number in result2. The sign of both f and i is the same as the sign of
a and a will equal i + f .
The second call returns f such that f has the same sign as x, the absolute value of
f is less than the absolute value of y, and there exists and integer k such that k × y + f
equals x.

res := sys(Sys flt, fl pow, a, b)

res := sys(Sys flt, fl sqrt, a) CIN:y, POS:y, NAT:y
The first call returns an approximation to ab , and the second call attempts to return
the non negative square root of a.

res := sys(Sys flt, fl ceil, a)

res := sys(Sys flt, fl floor, a) CIN:y, POS:y, NAT:y
The first call returns the smallest floating-point number not less than a whose value
is an exact integer and the second call returns the largest floating-point number not
greater than a whose value is an exact integer.

res := sys(Sys flt, fl F2N, s, x) CIN:y, POS:y, NAT:y

This returns the integer part of s × x. This is the scaled fixed point representation
of x when s is the scaled value representing 1.0. For example:

sys(Sys_flt, fl_F2N, 1_000, -1.234) => -1234

3.3. GLOBAL FUNCTIONS 79

res := sys(Sys flt, fl N2F, s, n) CIN:y, POS:y, NAT:y

This returns the floating point value corresponding to n/s. This is the floating
point number representing the fixed point scaled value n when the scaled number s
represents 1.0. For example:

sys(Sys_flt, fl_N2F, 1_000, 1_234) = 1.234

res := sys(Sys flt, fl radius2, a, b)

res := sys(Sys flt, fl radius3, a, b, c) CIN:y, POS:y, NAT:y
The first call returns the square root of a2 + b2 and the second returns the square
root of a2 + b2 + c2 . For example:

sys(Sys_flt, fl_radius2, 3.0, 4.0) => 5.000

sys(Sys_flt, fl_radius3, 1.0, 2.0, 2.0) => 3.000

sys(Sys freevec, ptr) CIN:y, POS:y, NAT:y

If ptr is zero it does nothing, otherwise it returns to free store the space pointed to
by ptr which must have previously been allocated by sys(Sys getvec,...). It checks
that the block is not already free and attempt to check that it has not been corrupted.

res := sys(Sys getpid) CIN:y, POS:y, NAT:y

This function returns the process id of the currently executing process.

str := sys(Sys getprefix) CIN:y, POS:y, NAT:y

This returns a pointer to prefix string which is in space allocated when Cintsys or
Cintpos was started.. See sys(Sys setprefix,...) on page 84.

res := sys(Sys getsysval, addr) CIN:y, POS:y, NAT:y

This function return the contents of the machine memory location whose word
address is addr which may be outside the normal range of the Cintcode memory.

res := sys(Sys gettrval, count) CIN:y, POS:y, NAT:n

This returns a value from the low level trace buffer. See Sys trpush for more
details.

res := sys(Sys getvec, upb) CIN:y, POS:y, NAT:y

This allocates a vector whose lower bound is 0 and whose upper bound is upb. It
returns zero if the request cannot be satisfied. A word is allocated just before the start
of the vector to hold its size, and several (typically 4 or 5) words are allocated just past
the end of the vector and filled with redundant data that is checked when the space is
returned to free store.

res := sys(Sys globin, seg) CIN:y, POS:y, NAT:n

This initializes the global variables defined in the loaded module pointed to by seg.
It returns zero is there is an error.
80 CHAPTER 3. THE LIBRARY

res := sys(Sys graphics,...) CIN:y, POS:y, NAT:y

This is currently only useful on the Windows CE version of the BCPL Cintcode
system. It performs operations on the graphics window. The graphics window is a fixed
size array of 8-bit pixels which can be written to and whose visibility can be switched
on and off.

res := sys(Sys inc, addr, amount) CIN:y, POS:y, NAT:y

This function adds amount atomically to the specified memory location and returns
it new value.

res := sys(Sys incdcount, n) CIN:y, POS:y, NAT:y

This function increments the specified counter held in the vector pointed to by the
field rtn dcountv in the rootnode. This operation is also available to the interpreter
code written in C.

res := sys(Sys interpret, regs) CIN:y, POS:y, NAT:n

This function enters the Cintcode interpreter recursively with the Cintcode regis-
ters set to values specified in the vector regs. On return the result is a return code
indicating why the interpreter returned, and the elements of regs hold the final state
of the Cintcode registers. These registers are described in the chapter on the design
of Cintcode starting on page 189 and the correspondence between the elements of regs
and the Cintcode registers is given on page 83. The return codes are given on page 84.

res := sys(Sys intflag) CIN:y, POS:y, NAT:y

This returns TRUE if the user has pressed a particular combination of keys to in-
terrupt the program that is currently running. On many systems this mechanism not
implemented and just returns FALSE.

res := sys(Sys loadseg, name) CIN:y, POS:y, NAT:n

This attempts to load a Cintcode module from file name looking first in the current
directory. If a valid module is not found there and name is a relative file name, it
searches through the directories specified by the environment variable whose name
is in the rtn pathvar element of the rootnode. This name is normally BCPLPATH
under Cintsys and POSPATH under Cintpos. See Section 3.6 for more information about
environment variables.
If loading is successful, loadseg returns the list of loaded program sections, other-
wise it returns zero. Before the loaded code can be used, its globals must be initialised
using globin.
Cintcode modules generated by the BCPL compiler are typically text files con-
taining the compiled code encoded in hexadecimal. The compiled form of the logout
3.3. GLOBAL FUNCTIONS 81

command:

SECTION "logout"
GET "libhdr"
LET start() BE abort(0)

is

000003E8 0000000E
0000000E 0000FDDF 474F4C0B 2054554F 20202020
0000DFDF 6174730B 20207472 20202020 7B1C2310
00000000 00000001 00000024 0000001C

The first two words (000003E8 0000000E) indicate the presence of a “hunk” of code
of size 14(000000E) words which then follow. The first word of the hunk (000000E)
is again its length. The next four words (0000FDDF 474F4C0B 2054554F 20202020)
contain the SECTION name "logout". These are followed by the four words 0000DFDF
6174730B 20207472 20202020 which hold the name of the function "start". The
body of start is compiled into one word (7B1C2310) which correspond to the Cintcode
instructions:

L0 Load A with 0
K3G 28 Call the function in global 28, incrementing the stack by 3
RTN Return from start – never reached

The remaining 4 words contain global initialisation data that is read backwards during
global initialisation invoked by sys(Sys globin,...). 0000001C (=28) is the highest
global variable referenced by this section. The pair 00000001 00000024 specifies that
the entry point at position 36 is the initial value of global 1, and the next entry
(00000000) marks the end of the global initialisation data.
The manifest constants t hunk, t reloc, t end, t hunk64, t reloc64, t end64,
t bhunk, and t bhunk64 are declared in libhdr for the convenience of programs that
generate or read Cintsys and Cintpos object modules. The example above shows t hunk
loading n 32-bit words encoded in hex bytes. Although the BCPL compiler used in both
Cintsys and Cintpos generates position independent code and has no need to modify
the loaded words of a hunk, other languages may need to perform relocation. This can
be done using t reloc which is followed by a 32-bit word n encoded in hex followed by a
further n words which each give the position of a word in the most recently loaded hunk
that needs to be modified by the addition of the base address of the hunk. The code
t bhunk is similar to t hunk only the data words (not the length field) are provided in
binary rather than hex characters. Such hunks are thus about half the size of character
based ones. The code t end marks the end of an object module, but end-of-file has
the same effect. Those codes containing the characters 64 provide equivalent facilities
for 64-bit versions of BCPL. Neither t reloc nor t reloc64 are currently available in
Cintsys or Cintpos.

sys(Sys lockirq) CIN:y, POS:y, NAT:y

Under cintpos, this call disables interrupts.
82 CHAPTER 3. THE LIBRARY

res := sys(Sys muldiv, a, b, c) CIN:y, POS:y, NAT:y

This invoke the C implementation of muldiv. It returns the result of dividing c into
the double length product of a and b. It sets result2 to the remainder. This function
is little used since a more efficient muldiv function is now defined in syslib invoking
the Cintcode instruction MDIV, see section 3.3.

fp := sys(Sys openappend, name) CIN:y, POS:y, NAT:y

This function opens an output stream specified by the file name name in append
mode causing all output to be appended onto the end of the file. If the file does not
exist a zero length file of the given name is created. If successful it returns the file
pointer to the given file, otherwise it returns zero.

fp := sys(Sys openread, name, envname) CIN:y, POS:y, NAT:y

This opens for reading the file whose name is given by the string name. It returns
0 if the file cannot be opened, otherwise it returns the file pointer for the opened file.
See page 92 for information about the treatment of file names. If name is a relative
filename, the file is first searched for in the current directory, otherwise, if envname is
non null, the directories specified by the environment variable envname are searched.

res := sys(Sys openreadwrite, name) CIN:y, POS:y, NAT:y

This opens for reading and writing the file whose name is given by the string name.
It returns 0 if the file cannot be opened, otherwise it returns the file pointer for the
opened file. See Section 3.3.2 for information about the treatment of file names and
Section 3.4 for information about random access files.

fp := sys(Sys openwrite, name) CIN:y, POS:y, NAT:y

This opens for writing the file whose name is given by the string name. It returns
0 if the file cannot be opened, otherwise it returns the file pointer for the opened file.
See page 92 for information about the treatment of file names.

res := sys(Sys platform) CIN:y, POS:y, NAT:n

This returns a machine dependent value indicating under which architecture Cintsys
or Cintpos is running.

res := sys(Sys pollsardch) CIN:y, POS:y, NAT:y

This returns the next character from standard input if it is immediately available,
otherwise it returns pollingch (=-3). If the input stream is exhausted it returns
endstreamch (=-1). Unlike sardch, the character is not echoed to the standard output
stream.

res := sys(Sys putsysval, addr, val) CIN:y, POS:y, NAT:n

This function sets atomically the contents of the machine memory location whose
word address is addr to val returning its previous setting. The address may point to
system work space outside the normal Cintcode memory.

sys(Sys quit, code) CIN:y, POS:y, NAT:n

This saves the Cintcode registers in the vector of registers given to the interpreter
3.3. GLOBAL FUNCTIONS 83

when it was invoked and returns with the result code to the (C) program that called
this invocation of the interpreter. This is normally used to exit from the Cintcode
system, but can also be used to return from recursive invocations of the interpreter
(see sys(Sys interpret,regs) above). A code of zero denotes successful completion
and, if invoked at the outermost level, causes the BCPL Cintcode System to terminate.

n := sys(Sys read, fp, buf, len) CIN:y, POS:y, NAT:y

This reads upto len bytes from the file specified by the file pointer fp into
the byte buffer buf. The file pointer fp must have been created by a call of
sys(Sys openread,...) or sys(Sys openreadwrite,...). The number of bytes ac-
tually read is returned as the result.

res := sys(Sys renamefile, old, new) CIN:y, POS:y, NAT:y

This renames file old to new. It return 0 if successful.

sys(Sys rti, regs) CIN:n, POS:y, NAT:n

Under Cintpos, this returns from an interrupt by setting the Cintcode registers to
the values specified by regs.

ch := sys(Sys sardch) CIN:y, POS:y, NAT:y

This returns the next character from standard input (normally the keyboard). The
character is echoed to standard output (normally the screen). If the -c or -- command
options are given when cintsys or cintpos is invoked, standard input is prefixed with
text from the command line. For details, see Section 12.2 on page 243.

sys(Sys saveregs, regs) CIN:n, POS:y, NAT:n

Under Cintpos, this saves the current Cintcode registers in regs.

sys(Sys sawrch, ch) CIN:y, POS:y, NAT:y

This sends character repesented by the least significant 8 bit of ch to the standard
output (normally the screen). If ch=10, the characters carriage return followed by
linefeed are transmitted.

res := sys(Sys seek, fd, pos) CIN:y, POS:y, NAT:y

This will set the file position pointer of the opened file whose descriptor is fd to
pos. The file descriptor is normally in the scb fd field of the stream control block for
that file. The value of pos must be between zero and the current number of bytes in
the file. See Section 3.4 for more information about random access files.

oldcount := sys(Sys setcount, newcount) CIN:y, POS:y, NAT:n

One of the Cintcode registers is called count which is inspected just before the
interpreter processes the next instruction. If count>0 it is decremented and the in-
struction processed. If count=0 the interpreter returns to the calling (C) program with
error code 3.
The Cintcode System normally has two resident interpreters. One is called cinterp
implemented in C and the other is called fasterp which is sometimes implemented in
assembly language. fasterp is faster than cinterp since it provides fewer debugging
84 CHAPTER 3. THE LIBRARY

aids, does not count instruction executions and does not implement the profiling feature.
Setting count to a negative value causes this faster interpreter to be invoked and setting
count to a positive value causes the slower interpreter to be used. Normally the CLI
command interpreter is used to make this switch, see Section 4.3.
With some debugging versions of fasterp, setting count to -2 causes it to exe-
cute just one instruction before returning with error code 10. This feature assists the
debugging of a new versions of fasterp and is particularly useful when fasterp is
implemented in assembly language.

regs!0 A register – work register

regs!1 B register – work register
regs!2 C register – work register
regs!3 P register – the stack frame pointer
regs!4 G register – the base of the global vector
regs!5 ST register – the status register (unused)
regs!6 PC register – the program counter
regs!7 Count register – see below
regs!8 MW register – Used only on 64-bit systems, see below

Both interpreter cinterp and fasterp returns when a fault such as division by zero
occurs or when a call of sys(Sys_quit,...) is made. Before returning, the interpreter
save the Cintcode registers in regs. The returned result is either the second argument
of sys(Sys_quit,...) or one of the builtin return codes in the following table:

-1 Re-enter the interpreter with a new value in the the count

register
0 Normal successful completion (by convention)
1 Non existent Cintcode instruction
2 BRK instruction encountered
3 Count has reached zero
4 PC set to a negative value
5 Division by zero
10 Single step interrupt from the fast interpreter (debugging)
11 The value of the watched location in the Cincode memory
has changed in the course of executing the previous instruc-
tion
12 Indirect address out of range
13 SIGINT received

res := sys(Sys setprefix, prefix) CIN:y, POS:y, NAT:y

This is primarily a function for the Windows CE version of the BCPL Cintcode
System for which there is no current working directory mechanism. The prefix string
is held in space that was allocated when the system started. It sets the prefix that
is prepended to all future relative file names. See Section 3.3.2 and the CLI prefix
command described on page 145.
3.3. GLOBAL FUNCTIONS 85

res := sys(Sys setraster, n, arg) CIN:y, POS:y, NAT:n

There is a variant of cintsys called rastsys that provides a way to generate data
for time-memory images, and cintpos has a similar variant called rastpos. These
systems can also generate bit streams that can be converted in sound related to the
execution of programs. The setraster operation controls the rastering feature as
follows. If n=3, it returns 0 if rastering is available and -1 otherwise. If n=2, the
memory granularity is set to arg bytes per pixel, the default being 12. If n=1, the
number of Cintcode instructions executed per raster line is set to arg, the default being
1000. If n is zero and arg is non-zero, rastering is activated sending its output to the
file with name arg (the rastering data file). Raster information is normally collected for
the duration of the next CLI command. If n and arg are both zero, the rastering data
file is closed. If n = 4 and arg=1, the system generated a bit stream file based on the
fifth bit of the address of every access to the Cintcode memory. This file can later be
converted to sound using the command rast2wav. The generated sound is somewhat
similar to that generated by the Edsac 2 computer in Cambridge in the early 1960s.
When not representing bit stream data, the raster file contains text using run length
encoding to represent raster lines. Typical output is as follows:

K1000 S12 1000 instruction per raster line, 12 bytes per pixel
W10B3W1345B1N 10 white, 3 black, 1345 white, 1 black, newline
W13B3W12B2N etc
...

See the CLI commands raster and rast2ps on page 148 for more information on
how to use the rastering facility. See also the command bits2ps.

res := sys(Sys settrcount, count) CIN:y, POS:y, NAT:n

This sets the private variable trcount used by the low level tracing mechanism to
the specified value returning it previous setting. Setting it to a negative value disables
the tracing mechanism. See Sys trpush for more details.

res := sys(Sys sound, fno, a1, a2 ...) CIN:y, POS:y, NAT:y

This calls sound(args, g) where sound is a C function defined in sysc/sound.c.
The argument args points to memory locations holding fno, a1, a2, etc., and g points
to the base of the global vector. Note that it may be necessary to run alsamixer to
enable the sound device and adjust its volume setting. The available sound functions
have mnemonic names declared in g/sound.h and are described below.

res := sys(Sys sound, snd test) CIN:y, POS:y, NAT:y

This returns TRUE is the Sys sound functions are available on the current system.

res := sys(Sys sound, snd waveInOpen, a1, a2 , a3 , a4 ) CIN:y, POS:y,

NAT:y
This opens a sound wave device for input. a1 is typically "/dev/dsp", "/dev/dsp1"
or a small integer, a2 is the sample format, eg 16 for S16 LE, 8 for U8. a3 is the number
of channels, typically 1 or 2 and a4 is the number of samples per second, typically 44100.
The result is the file (or device) descriptor of the opened device or -1 if error.
86 CHAPTER 3. THE LIBRARY

res := sys(Sys sound, snd waveInPause, a1) CIN:y, POS:y, NAT:y

This will pause sound wave sampling from device a1. Recently read samples can
still be read (to flush the buffered data).

res := sys(Sys sound, snd waveInRestart, a1) CIN:y, POS:y, NAT:y

Restart sound wave sampling.

res := sys(Sys sound, snd waveInRead, a1, a2 , a3 ) CIN:y, POS:y, NAT:y

Read samples from a sound wave input device a1, returning immediately. a2 is the
buffer in which to receive the samples and a3 is the number of bytes to read. The result
is the number of bytes actually transferred into the buffer.

res := sys(Sys sound, snd waveInClose, a1) CIN:y, POS:y, NAT:y

This closes sound wave input device a1.

res := sys(Sys sound, snd waveOutOpen, a1, a2 , a3 ) CIN:y, POS:y, NAT:y

This opens a sound wave device for output. a1 is typically "/dev/dsp",
"/dev/dsp1" or a small integer, a2 is the sample format, eg 16 for S16 LE, 8 for
U8. a3 is the number of channels, typically 1 or 2 and a4 is the number of samples
per second, typically 44100. The result is the file (or device) descriptor of the opened
device or -1 if error.

res := sys(Sys sound, snd waveOutWrite, a1, a2 , a3 ) CIN:y, POS:y, NAT:y

Write samples from a sound wave output device a1. a2 is the buffer holding the
samples and a3 is the number of bytes to be written. The result is the number of bytes
actually transferred from the buffer.

res := sys(Sys sound, snd waveOutClose, a1) CIN:y, POS:y, NAT:y

This closes sound wave output device a1.

res := sys(Sys sound, snd midiInOpen, a1) CIN:y, POS:y, NAT:y

This opens MIDI device for input specified by a1 which is typically "/dev/midi",
"/dev/dmmidi1" or a small integer. The result is the file (or device) descriptor of the
opened device or -1 if error.

res := sys(Sys sound, snd midiInRead, a1, a2 , a3 ) CIN:y, POS:y, NAT:y

This reads bytes from MIDI input device a1 into buffer a2. a3 is the number of
MIDI bytes to read. The result is the actual number of bytes transferred or -1 if there
was an error.

res := sys(Sys sound, snd midiInClose, a1) CIN:y, POS:y, NAT:y

This close MIDI input device a1.

res := sys(Sys sound, snd midiOutOpen, a1) CIN:y, POS:y, NAT:y

This opens a MIDI device for output. a1 is typically "/dev/midi", "/dev/dmmidi1"
or a small integer. The result is the file (or device) descriptor of the opened device or
-1 if error.
3.3. GLOBAL FUNCTIONS 87

res := sys(Sys sound, snd midiOutWrite1, a1, a2) CIN:y, POS:y, NAT:y
This writes a one byte MIDI message (a2) to MIDI device a1.

res := sys(Sys sound, snd midiOutWrite2, a1, a2 , a3 ) CIN:y, POS:y, NAT:y

This writes a two byte MIDI message (a2 a3) to MIDI device a1.

res := sys(Sys sound, snd midiOutWrite3, a1, a2 , a3 , a4 ) CIN:y, POS:y,

NAT:y
This writes a three byte MIDI message (a2 a3 a3) to MIDI device a1.

res := sys(Sys sound, snd midiOutWrite, a1, a2 ...) CIN:y, POS:y, NAT:y
This write a3 MIDI bytes from buffer a2 to MIDI output device a1. The result is
the number of bytes actually sent.

res := sys(Sys sound, snd midiOutClose, a1) CIN:y, POS:y, NAT:y

This closes MIDI output device a1.

sys(Sys setst, val) CIN:n, POS:y, NAT:n

Under Cintpos, this sets the Cintcode ST register to val. Interrupts are enabled only
when ST is zero. By convention, ST=1 while execution is within klib, ST=2 when
executing within the interrupt routine, and ST=3 during the initial bootstrapping
process.

res := sys(Sys shellcom, comstr) CIN:y, POS:y, NAT:y

This causes the command comstr to be executed by the command language shell
of the operating system under which Cintsys or Cintpos is running.

sys(Sys tally, val) CIN:y, POS:y, NAT:n

This call provides a profiling facility that uses a globally accessible tally vector to
hold frequency counts of Cintcode instructions executed. When val is TRUE the tally
vector is cleared and tallying is enabled. When val is FALSE tallying is disabled. When
tallying is active, the ith element of the tally vector is incremented every time the
instruction at location i of the Cintcode memory is executed. The size of the tally
vector can be specified by the -t command line argument (see Section 12.2) when the
interpreter is entered. The default size being typically 80000 words. The tally vector
is held in rootnode!rtn tallyv with the upper bound stored in its zeroth element. It
can thus be inspected by any program.
Statistics of program execution is normally gathered and analysed using the CLI
command stats (see Section 4.3).

pos := sys(Sys tell, fd) CIN:y, POS:y, NAT:y

This returns the current file position pointer of the opened file whose descriptor is
fd. The file descriptor is normally in the scb fd field of the stream control block for
that file. See Section 3.4 for more information about random access files.

sys(Sys tracing, val) CIN:y, POS:y, NAT:n

This sets the Cintcode tracing mode to val. When the tracing mode is TRUE, the
Cintcode interpreter outputs a one line trace of every Cintcode instruction executed.
88 CHAPTER 3. THE LIBRARY

sys(Sys trpush, val) CIN:y, POS:y, NAT:n

There is a low level circular trace buffer that can hold 4096 values, and a pri-
vate variable trcount that holds the number of values currently pushed into this
buffer. If trcount<0, low level tracing is disabled, but otherwise trpush pushes
val into the buffer at position trcount MOD 4096 and increments trcount. The call
sys(Sys settrcount, count) sets trcount to the specified value (possibly disabling
tracing) and returns its previous setting. The call sys(Sys gettrval, count) gets
the value in the trace buffer at position trcount MOD 4096. Normally this function is
only called when tracing is disabled. Under both Cintsys and Cintpos, trpush can also
be called from the parts of the system implemented in C.
This tracing mechanism is available both to the BCPL user and parts of the system
such as cintpos.c, cinterp.c and devices.c. Under Cintpos these low level tracing
functions use a mutex to control access to trcount and the circular buffer. It is thus
thread safe and so can be used to help debug subtle timing problems in the system
software. For an example of the use of this tracing mechanism see the command
com/testtr.b.

res := sys(Sys unloadseg, seg) CIN:y, POS:y, NAT:y

This unloads the the loaded module given by seg. If seg is zero it does nothing.
Unloading a module just returns the space it occupied to freestore.

sys(Sys unlockirq) CIN:n, POS:y, NAT:n

Under cintpos, this call enables interrupts.

res := sys(Sys usleep, usecs) CIN:y, POS:y, NAT:y

Under cintsys, this call causes the system to sleep for usecs micro-seconds. Under
cintpos, it causes the current task to sleep for usecs micro-seconds.

sys(Sys waitirq, msecs) CIN:n, POS:y, NAT:n

This call is typically only made from the body of the Cintpos Idle task. It suspends
the interpreter until either some Cintpos device issues an interrupt request or the
specified timeout occurs. It is typically implemented by waiting with a timeout on a
host operating system condition variable. When a device thread wishes to interrupt
the interpreter it send a signal via the appropriate condition variable. Unfortunately
some operating systems may take hundreds of milliseconds to reschedule the interpreter
thread. A possible but selfish solution is for the Idle task to execute a busy loop instead
of calling waitirq.

sys(Sys watch, addr) CIN:y, POS:y, NAT:n

This sets the address of a location of Cintcode memory to be inspected every time
the interpreter executes and instruction. When the watched value changes it returns
with result 12. The watch feature is disabled if addr is zero or if fasterp is being used.

n := sys(Sys write, fp, buf, len) CIN:y, POS:y, NAT:y

This writes len bytes to the file specified by the file pointer fp from the byte buffer
buf. The file pointer must have been created by a call of sys(Sys openwrite,...) or
3.3. GLOBAL FUNCTIONS 89

sys(Sys openreadwrite,...). The result is the number of bytes transferred, or zero

if there was an error.

pkt := taskwait() CIN:n, POS:y, NAT:n

If there is a packet in the task’s queue it is dequeued and returned as the result.
If there was no packet on the work queue this task is suspended in WAIT state and
control given to a lower priority task.

res := testbit(bitno, bitvec) CIN:y, POS:y, NAT:y

This function returns a non zero value if and only if the specified bit in bitvec
is a one. The bits are numbered from zero starting at the least significant bit of
bitvec!0. bitvec!0 holds bits 0 to bitsperword-1, bitvec!1 holds bits bitsperword
to 2*bitsperword-1, etc.

res := testflags(flags) CIN:n, POS:y, NAT:n

This Cintpos function tests and clears specified flags in the task control block of the
current task. Flags are bits in the tcb flags field of the task control block, and they
are normally called A, B, etc corresponding to consecutive bits from the least significant
end of the field. A flag is set if the corresponding bit is a one. The argument flags is a
bit pattern identifying which flags are being inspected. The result is FALSE if none of
the specified flags were set, and TRUE if at least one was, in which case result2 is set
to a bit pattern representing the flags that were set and have now been cleared.

unloadseg(segl) CIN:y, POS:y, NAT:y

This routine unloads the list of loaded program modules given by segl.

res := unrdch() CIN:y, POS:y, NAT:y

This attempts to step the current input stream back by one character position. It
returns TRUE if successful, and FALSE otherwise. A call of unrdch will always succeeds
the first time after a call of rdch. It is useful in functions such as readn where single
character lookahead is necessary. See Section 3.3.1 for more detailed information.

wrch(ch) CIN:y, POS:y, NAT:y

This routine writes the character ch to the currently selected output stream. If
output is to the screen, ch is transmitted immediately. It aborts (with code 189) if
there is a write failure.

writed(n, d) CIN:y, POS:y, NAT:y

writeu(n, d) CIN:y, POS:y, NAT:y
writen(n) CIN:y, POS:y, NAT:y
These routines output the integer n in decimal to the currently selected output
stream. For writed and writeu, the output is padded with leading spaces to fill a field
width of d characters. If writen is used or if d is too small, the number is written
without padding. If writeu is used, n is regarded as an unsigned integer.

writehex(n, d) CIN:y, POS:y, NAT:y

writeoct(n, d) CIN:y, POS:y, NAT:y
90 CHAPTER 3. THE LIBRARY

writebin(n, d) CIN:y, POS:y, NAT:y

These routines output, repectively, the least significant d hexadecimal, octal or
binary digits of the integer n to the currently selected output stream.

writes(str) CIN:y, POS:y, NAT:y

writet(str, d) CIN:y, POS:y, NAT:y
These routines output the string str to the currently selected output stream. If
writet is used, trailing spaces are added to fill a field width of d characters.

writef(format,a,b,c,d,e,f ,g,h,i,j,k,l,m,n,o,p,q,r,s,t,u,v,w,x,y,z)
CIN:y, POS:y, NAT:y
The first argument (format) is a string that is copied character by character to
the currently selected output stream until a substitution item such as %s or %i5 is
encountered when a value (usually the next argument) is output in the specified format.
The substitution items are given in table 3.6.
When a field width (denoted by n in the table) is required, it is specified by a
single character, with 0 to 9 being represented by the corresponding digit and 10 to
35 represented by the letters A to Z. Format characters are case insensitive but field
width characters are not. A recent entension allows the field width to be specified as a
decimal integer immediately following the percent, as in %12i meaning %iB.
Some examples of the %n.md substitution item are given below.
writef("%9.2d", 1234567) writes 12345.67
writef("%9.2d", -1234567) writes -12345.67
writef("%9.0d", 1234567) writes 1234567
writef("%9d", 1234567) writes 1234567
As an example of how the %p substitution item can be used, the following code:

FOR count = 0 TO 2 DO
writef("There %p\ is\are\ %-%n thing%-%ps.*n", count)

outputs:

There are 0 things.

There is 1 thing.
There are 2 things.

The implementation of writef (in sysb/blib.b) is a good example of how a vari-

adic function can be defined.

writeflt(x, w, p) CIN:y, POS:y, NAT:y

This routine outputs the floating point number x to the currently selected output
stream in a field of width w with p digits after the decimal point.

writee(x, w, p) CIN:y, POS:y, NAT:y

This routine outputs the floating point number x to the currently selected output
stream in exponential form in a field of width w with p digits after the decimal point.
3.3. GLOBAL FUNCTIONS 91

Item Substitution
%s Write the next argument as a string using writes.
%nt %tn Write the next argument as a left justified string in a field width of
n characters using writet.
%c Write the next argument as a character using wrch.
%# Write the next argument as a in UTF-8 or GB2312 character using
codewrch.
%nb %bn Write the next argument as a binary number in a field width of n
characters using writebin.
%no %on Write the next argument as an octal number in a field width of n
characters using writeoct.
%nx %xn Write the next argument as a hexadecimal number in a field width
of n characters using writehex.
%ni %in Write the next argument as a decimal number in a field width of n
characters using writed.
%n Write the next argument as a decimal number in its natural field
width using writen.
%nu %un Write the next argument as an unsigned decimal number in a field
width of n characters using writeu.
%n.md Write the next argument as a scaled decimal number in a field with
of n with m digits after the decimal point.
%+ Skip over the next argument.
%- Step back to the previous argument.
%% Write the character %.
%pc Plural formation. Write character c if the next argument is not 1.
%p\a\b\ Plural formation. Write text a if the next argument is 1, otherwise
write text b.
%f Take the next argument as a writef format string and call writef
recursively to process it passing it the remaining arguments. The
argument pointer is advanced by the appropriate amount.
%n.mf Write the next argument as a floating point number in a field with
of n with m digits after the decimal point. The output is generated
using writeflt.
%n.me Write the next argument as a floating point number in exponential
form in a field with of n with m digits after the decimal point. The
output is generated using writee.
%m The next argument is taken as a message number and processes
as for %f above using the message format string obtained by the
call get text(messno, str, upb) where str is a vector local to
writef to hold the message string. This provides an easy way
to generate messages in different languages. get text is a global
function typically defined by the user. The default version always
yields the message string "<mess:%-%n>".

Figure 3.6: writef substitution items

92 CHAPTER 3. THE LIBRARY

3.3.1 Streams
BCPL uses streams as a convenient method of obtaining device independent input and
output. All the information needed to process a stream is held in a vector called a
stream control block (SCB) whose fields have already been summarized in Section 3.1.
The element buf is either zero or holds the stream’s byte buffer which must have
been allocated using getvec and must be freed using freevec when the stream is
closed. The elements pos and end hold positions within the byte buffer, file holds a
file pointer for file streams or -1 for streams connected to the console. The element
id indicates whether the stream is for input, output or both and work is private work
space for the action function rdfn, wrfn which are called, repectively, when the byte
buffer becomes empty on reading or full on output. The function endfn is called to
close the stream.
Input is read from the currently selected input stream whose SCB is held in the
global variable cis. For an input stream, pos holds the position of the next character to
be read, and end points to just past the last available character in the buffer. Characters
are read using rdch whose definition is given in figure 3.7. If a character is available in
the buffer it is returned after incrementing pos. Exceptionally, the character carriage
return (CR) is ignored since on some systems, such as Windows, lines are terminated
with carriage return and linefeed while on others, such as Linux, only linefeed is used.
If the buffer is exhausted, replenish is called to refill it, returning TRUE if one or
more character are transferred. If replenish fails it returns FALSE with the reason why
in result2. Possible reasons are: -1 indicating end of file, -2 indicating a timeout
has occurred and -3 meaning input is in polling mode and no character is currently
available. By setting the timeoutact field of the SCB to -1, a timeout is treated as
end of file.
Whenever possible, the buffer contains the previously read character. This is to
allow for a clean and simple implementation of unrdch whose purpose is to step input
back by one character position. Its definition is given in figure 3.8.
Output is sent to the currently selected output stream whose SCB is held in the
global variable cos. The SCB field pos of an output stream holds the position in the
buffer of the next character to be written, and end holds the position just past the end
of the buffer. Characters are written using the function wrch whose definition is given
in figure 3.9. The character ch is copied into the byte buffer and pos incremented. If
the buffer is full, it is emptied by calling the element wrfn. If writing fails it return
FALSE, causing wrch to abort.

3.3.2 The Filing System

BCPL uses the filing system of the host operating system and so some details such as
the maximum length of file names are machine dependent. Previously, BCPL used to
follow the syntax of target machine files names, but recently BCPL attempts to be more
machine independent by mainly adopting the Linux style of names and converting them
to target machine form at runtime. The target machine format is set by a configuration
parameter set when the system was installed. The formats currently available are for
Unix, Windows and VMS.
3.3. GLOBAL FUNCTIONS 93

AND rdch() = VALOF

{ LET pos = cis!scb_pos // Position of next byte, if any
UNLESS cis DO abort(186)
IF pos<cis!scb_end DO { LET ch = cis!scb_buf%pos
cis!scb_pos := pos+1
IF ch=’*c’ LOOP // Ignore CR
RESULTIS ch
}
// If replenish returns FALSE, it failed to read any characters
// and the reason why is placed in result2 as follows
// result2 = -1 end of file
// result2 = -2 timeout
// result2 = -3 polling mode with no characters available.
// result2 = code error code
UNTIL replenish(cis) DO
{ IF result2=-2 DO
{ LET act = cis!scb_timeoutact // Look at the timeout action
IF act=-2 RESULTIS timeoutch // Timed out
IF act=-1 RESULTIS endstreamch // End of file reached
LOOP // Try replenishing again
}
RESULTIS result2<0 -> result2, endstreamch
}
} REPEAT

Figure 3.7: The definition of rdch

LET unrdch() = VALOF
{ LET pos = cis!scb_pos
IF pos<=scb_bufstart RESULTIS FALSE // Cannot UNRDCH past origin.
cis!scb_pos := pos-1
RESULTIS TRUE
}

Figure 3.8: The definition of unrdch

Within BCPL file names slashs (/) and back slashes (\) are regarded as separators
between the components of file names. File names may start with a colon prefix con-
sisting of letters and digits followed by a colon, as in TCP:shep.cl.cam.ac.uk:9000
or G:test.b. Such prefixes allow access to special features such as URLs used in
TCP/IP communication or to other filing systems. These are often dependent on the
host operating system.
A file name starting ’/’ or ’\’ or containing a colon is treated as an abso-
lute name; all others are relative names and are interpreted relative to the cur-
rent directory. A file name consisting of a single asterisk (*) is special and rep-
resents standard input (normally the keyboard) or standard output (normally the
screen) depending on context. Within a file name, the components dot (.) and
double dot (..) represent the current and parent directories, respectively. As
an example, the file name ../bcplprogs/demos/queens.b is valid and automati-
94 CHAPTER 3. THE LIBRARY

AND wrch(ch) = VALOF

{ LET pos = cos!scb_pos
IF pos >= cos!scb_bufend DO
{ // The buffer is full
UNLESS deplete(cos) RESULTIS FALSE
UNLESS cos!scb_buf RESULTIS TRUE // Must be writing to NIL:
pos := cos!scb_pos
}
// Pack the character and advance pos.
cos!scb_buf%pos := ch
pos := pos+1
cos!scb_pos := pos
// Advance end of valid data pointer, if necessary
IF cos!scb_end < pos DO cos!scb_end := pos
cos!scb_write := TRUE // Set flag to indicate the buffer has changed.
UNLESS ch<’*s’ & cos!scb_type<0 RESULTIS TRUE // Normal return
// The stream is interactive and ch is a control character.
IF ch=’*n’ DO wrch(’*c’) // Fiddle for Cygwin
// Call deplete at the end of each interactive line.
IF ch=’*n’ | ch=’*p’ RESULTIS deplete(cos)
RESULTIS TRUE
}

Figure 3.9: The definition of wrch

cally converted when used to ..\bcplprogs\demos\queens.b under Windows or to

[-.bcplprogs.demos]queens.b under VMS.
Some operating systems such as Windows CE2.0 have no concept of a current
working directory. For such systems there is a feature that allows users to spec-
ify a character string to be automatically prepended to any relative (non absolute)
file name before it is used. The prefix string is stored in static Cintcode space allo-
cated when Cintsys or Cintpos starts up. It can be inspected and changed using the
calls: sys(Sys getprefix) and sys(Sys setprefix, prefix), or the CLI command
prefix described on page 145. The prefix string is only used with relative file names not
already prefixed with directories given by path variables such as BCPLPATH or POSPATH.

3.4 Random Access

Disk files can be regarded as potentially huge vectors of bytes with the first byte being
at position zero of the file. An opened stream to or from a file has a file position pointer
that holds the position relative to the start of where the next byte will be transferred.
For any such stream this position can be read using note(scb, posv) or updated using
point(scb, posv). For read-write streams it is possible to read or write data at any
3.5. RAM STREAMS 95

position in the file.

Disk files can also be regarded as potentially huge collections of fixed length records.
The user must specify the record size by calling setrecordlength. The records of a
file are given consecutive numbers starting with zero, and can be read or written using
get record and put record. The record number of the next record to be transferred
can be obtained by calling recordnote and can be set using recordpoint.

3.5 RAM streams

A special form of random access stream is a RAM stream which can be created by
the call findinoutput("RAM:"). RAM streams hold all the data in main memory in
the stream buffer. As data is written to a RAM stream, its buffer is automatically
enlarged as needed. The data can be read back by calling rewindstream followed by
calls of rdch. Alternatively it can be accessed from the buffer held in scb!scb buf.
The number of valid bytes in the buffer is scb!scb end. When a RAM stream is closed
its buffer and scb are returned to free store.

3.6 Environment Variables

Most operating systems allow the user to set environment variables whose names consist
of letters and digits and whose values are arbitrary character strings. Both Cintsys and
Cintpos use such variables to specify directories to be searched when looking up files
in certain contexts. These directories are separated by semicolons or colons, but when
running under Windows only semicolons are allowed.
In the standard Cintsys system the environment variable BCPLROOT holds the file
name of the root directory of the system. BCPLPATH holds a list of directories that are
searched when attempting to load the Cintcode compiled form of a BCPL program.
BCPLHDRS holds the directories to be searched when the BCPL compiler is processing
a GET directive and BCPLSCRIPTS specified the directories to be searched when the c
command is looking for a command-command script.
In the standard Cintpos system these variables are called POSROOT, POSPATH,
POSHDRS and POSSCRIPTS. It is sometimes convenient to use other names, for instance,
NBCPLROOT, NBCPLPATH, NBCPLHDRS and NBCPLSCRIPTS might be used when developing
a new version of Cintsys. To make this possible the system allocates static space to
hold the names and provides the command setroot described on page 150 to allow
the user to change them. These names may be up to 63 characters long are accessi-
ble to commands such as bcpl, c and setroot via the rootnode fields rtn rootvar,
rtn pathvar, rtn hdrsvar and rtn scriptsvar.
When Cintsys (or Cintpos) starts up it requires a valid setting of rtn pathvar in
order to locate Cintcode modules such as BOOT and BLIB. The default setting of this
field is BCPLPATH (or POSPATH) but can be changed using the -cin argument at startup
as in

cintsys -cin NBCPLPATH

96 CHAPTER 3. THE LIBRARY

After loading the resident system control is passed to BOOT which updates the variable
names appropriately for the system being run. It is unlikely that the user will want
change them using setroot although it might be useful to use setroot to see what
names are currently being used.
If the value of an environment variable represents a list of directories, they should be
given using Linux style slash (/) separators and the directories separated by semicolons
(rather than the Linux style colons). This allows colon prefixes such as G: to be used
in, for instance, Windows version of the system. For compatibility with older systems,
colons may be used as an alternative to semicolons when not running under Windows.
When Cintpos starts up the process is similar except the setting of rtn pathvar is
POSPATH unless explicitly changed using -cin.
When installing cintsys or cintpos for the first time it is common to fail to set
the environment variables correctly. To help repair such mistakes, use the -f option
when calling cintsys or cintpos. This will output a trace of every time any file is
looked up using an environment variable. Even more information is generated if the
-v argument is also given (or even -vv). Until the system is working correctly it is
recommended that it is started using

cintsys -f

or

cintpos -f -v

3.7 Coroutine examples

This section contains examples that use the coroutine mechanism.
3.7. COROUTINE EXAMPLES 97

3.7.1 A square wave generator

The following function is the main function of a coroutine that generates square wave
samples.

LET squarefn(args) = VALOF

{ LET freq, amplitude, rate = args!0, args!1, args!2
LET x = 0
cowait(@freq) // Return a pointer -> [freq, amplitude, rate]
{ // freq is a scaled fixed point value with
// three digits after the decimal point.
LET currfreq = freq // These only change at the
LET curramplitude = amplitude // start of a complete cycle.
LET q4 = rate*1000
LET q2 = q4/2
UNTIL x > q2 DO { cowait(+curramplitude) // First half cycle
x := x + currfreq
}
UNTIL x > q4 DO { cowait(-curramplitude) // Second half cycle
x := x + currfreq
}
x := x - q4
} REPEAT
}

The following call creates a coroutine that initially generates a square wave with fre-
quency 440Hz and amplitude 5000 at a rate of 44100 samples per second.

sqco := initco(squarefn, 300, 440_000, 5_000, 44_100)

sqparmv := result2 // sqparmv -> [freq, amplitude, rate]

One second’s worth of samples can now be obtained by:

FOR i = 1 TO 44100 DO
{ LET sample = callco(sqco)
...
}

The frequency and amplitude can be changed by assignments such as:

sqparmv!0 := newfrequency
sqparmv!1 := newamplitude

Note that the new frequency and amplitude take effect at the start of the next complete
cycle.
Other examples of the use of initco can be found below.
98 CHAPTER 3. THE LIBRARY

3.7.2 Hamming’s Problem

A following problem permits a neat solution involving coroutines.

Generate the sequence 1,2,3,4,5,6,8,9,10,12,... of all

numbers divisible by no primes other than 2, 3, or 5”.

This problem is attributed to R.W.Hamming. The solution given here shows how data
can flow round a network of coroutines. It is illustrated in figure 3.10 in which each
box represents a coroutine and the edges represent callco/cowait connections. The
end of a connection corresponding to callco is marked by c, and an end corresponding
to cowait is marked by w. The arrows on the connections show the direction in which
data moves. Notice that, in tee1, callco is sometimes used for input and sometimes
for output.
w w c c w w c c w w
BUF1 TEE1 BUF2 TEE2 BUF3
w w
c c c

X2 X3 X5
w w w
c c
c w c
MER1 MER2
w

c c
MAIN

Figure 3.10: Coroutine data flow

The coroutine BUF1 controls a queue of integers. Non-zero values can be in-
serted into the queue using callco(BUF1,val), and values can be extracted using
callco(BUF1,0). The coroutines BUF2 and BUF3 are similar. The coroutine TEE1 is
connected to BUF1 and BUF2 and is designed so that callco(TEE1) executed in corou-
tine X2 will yield a value that TEE1 extracted from BUF1, after sending a copy to BUF2.
TEE2 similarly takes values from BUF2 passing them to BUF3 and X3. Values passing
through X2, X3 and X5 are multiplied by 2, 3 and 5, respectively. MER1 merges two
monotonically increasing streams of numbers produced by X2 and X3. The resulting
monotonic stream is then merged by MER2 with the stream produced by X5. The stream
produced by MER2 is the required Hamming sequence, each value of which is printed
by MAIN and then inserted into BUF1.
3.7. COROUTINE EXAMPLES 99

The BCPL code for this solution is as follows:

GET "libhdr"
LET buf(args) BE // Body of BUF1, BUF2 and BUF3
{ LET p, q, val = 0, 0, 0
LET v = VEC 200
{ val := cowait(val)
TEST val=0 THEN { IF p=q DO writef("Buffer empty*n")
val := v!(q MOD 201)
q := q+1
}
ELSE { IF p=q+201 DO writef("Buffer full*n")
v!(p MOD 201) := val
p := p+1
}
} REPEAT
}

LET tee(args) BE // Body of TEE1 and TEE2

{ LET in, out = args!0, args!1
cowait() // End of initialisation.
{ LET val = callco(in, 0)
callco(out, val)
cowait(val)
} REPEAT
}

AND mul(args) BE // Body of X2, X3 and X5

{ LET k, in = args!0, args!1
cowait() // End of initialisation.
cowait(k * callco(in, 0)) REPEAT
}

LET merge(args) BE // Body of MER1 and MER2

{ LET inx, iny = args!0, args!1
LET x, y, min = 0, 0, 0
cowait() // End of initialisation
{ IF x=min DO x := callco(inx, 0)
IF y=min DO y := callco(iny, 0)
min := x<y -> x, y
cowait(min)
} REPEAT
}
100 CHAPTER 3. THE LIBRARY

LET start() = VALOF

{ LET BUF1 = initco(buf, 500)
LET BUF2 = initco(buf, 500)
LET BUF3 = initco(buf, 500)
LET TEE1 = initco(tee, 100, BUF1, BUF2)
LET TEE2 = initco(tee, 100, BUF2, BUF3)
LET X2 = initco(mul, 100, 2, TEE1)
LET X3 = initco(mul, 100, 3, TEE2)
LET X5 = initco(mul, 100, 5, BUF3)
LET MER1 = initco(merge, 100, X2, X3)
LET MER2 = initco(merge, 100, MER1, X5)
LET val = 1
FOR i = 1 TO 100 DO { writef(" %i6", val)
IF i MOD 10 = 0 DO newline()
callco(BUF1, val)
val := callco(MER2)
}
deleteco(BUF1); deleteco(BUF2); deleteco(BUF3)
deleteco(TEE1); deleteco(TEE2)
deleteco(X2); deleteco(X3); deleteco(X5)
deleteco(MER1); deleteco(MER2)
RESULTIS 0
}

3.7.3 A Discrete Event Simulator

This is a benchmark test for a discrete event simulator using coroutines. It simulates
a network of n nodes which each receive, queue, process and transmit messages to
other nodes. The nodes are uniformly spaced on a straight line and the network de-
lay is assumed to be proportional to the linear distance between the source and the
destination. When a message arrives at a node it is queued if the node was busy,
otherwise it is processed immediately. After processing the message for random time,
it is sent to another randomly chosen node. After dispatching the message, the node
dequeues its next message and processes it if there is one, otherwise the node becomes
suspended. Initially every node is processing a message and every queue is empty.
There are n coroutines to simulate the progress of each message and the discrete event
priority queue is implemented using the heapsort heap structure. The simulation stops
at a specified simulated time. The result is the number of messages that have been
processed. A machine independent random number generator is used so the resulting
value should be independent of implementation language and machine being used.
The program is given below. When it is run using the default settings, it executes
435,363,350 Cintcode instructions and has 2,510,520 coroutine changes.
3.7. COROUTINE EXAMPLES 101

SECTION "cosim"
GET "libhdr"
GLOBAL {
priq:ug // The vector holding the priority queue
priqupb // The upper bound
priqn // Number of items in the priority queue
wkqv // The vector of work queues
count // count of messages processed
nodes // The number of nodes
ptmax // The maximum processing time
stopco // The stop coroutine
cov // Vector of message coroutines
ranv // A vector used by the random number generator
rani; ranj // subscripts of ranv
simtime // Simulated time
stoptime // Time to stop the simulation
tracing
// Functions
rnd; initrnd; closernd; prq; insertevent; upheap
downheap; getevent; waitfor; prwaitq; qitem; dqitem
stopcofn; messcofn
}

// ################### Random number generator #######################

// The following random number generator is based on one given
// in Knuth: The art of programming, vol 2, p 26.
LET rnd(n) = VALOF
{ LET val = (ranv!rani + ranv!ranj) & #x_FFF_FFFF
ranv!rani := val
rani := (rani + 1) MOD 55
ranj := (ranj + 1) MOD 55
RESULTIS val MOD n
}

AND initrnd(seed) = VALOF

{ LET a, b = #x_234_5678+seed, #x_536_2781
ranv := getvec(54)
UNLESS ranv RESULTIS FALSE
FOR i = 0 TO 54 DO
{ LET t = (a+b) & #x_FFF_FFFF
a := b
b := t
ranv!i := t
}
rani, ranj := 55-55, 55-24 // ie: 0, 31
RESULTIS TRUE
}
AND closernd() BE IF ranv DO freevec(ranv)
102 CHAPTER 3. THE LIBRARY

// ################### Priority Queue functions ######################

AND prq() BE
{ FOR i = 1 TO priqn DO writef(" %i4", priq!i!0)
newline()
}
AND insertevent(event) BE
{ priqn := priqn+1 // Increment number of events
upheap(event, priqn)
}

AND upheap(event, i) BE
{ LET eventtime = event!0
//writef("upheap: eventtime=%n i=%n*n", eventtime, i)
{ LET p = i/2 // Parent of i
UNLESS p & eventtime < priq!p!0 DO
{ priq!i := event
RETURN
}
priq!i := priq!p // Demote the parent
i := p
} REPEAT
}

AND downheap(event, i) BE
{ LET j, min = 2*i, ? // j is left child, if present
IF j > priqn DO
{ upheap(event, i)
RETURN
}
min := priq!j!0
// Look at other child, if it exists
IF j<priqn & min>priq!(j+1)!0 DO j := j+1
// promote earlier child
priq!i := priq!j
i := j
} REPEAT

AND getevent() = VALOF

{ LET event = priq!1 // Get the earliest event
LET last = priq!priqn // Get the event at the end of the heap
UNLESS priqn>0 RESULTIS 0 // No events in the priority queue
priqn := priqn-1 // Decrement the heap size
downheap(last, 1) // Re-insert last event
RESULTIS event
}

AND waitfor(ticks) BE
{ // Make an event item into the priority queue
LET eventtime, co = simtime+ticks, currco
insertevent(@eventtime) // Insert into the priority queue
cowait() // Wait for the specified number of ticks
}
3.7. COROUTINE EXAMPLES 103

// ###################### Queueing functions #########################

AND prwaitq(node) BE
{ LET p = wkqv!node
IF -1 <= p <= 0 DO { writef("wkq for node %n: %n*n", node, p); RETURN }
writef("wkq for node %n:", node)
WHILE p DO
{ writef(" %n", p!1)
p := !p
}
newline()
}

AND qitem(node) BE
// The message has reached this node
// It currently not busy, mark it as busy and return to process
// the message, other append it to the end of the work queue
// for this node.
{ // Make a queue item
LET link, co = 0, currco
LET p = wkqv!node
UNLESS p DO
{ // The node was not busy
wkqv!node := -1 // Mark node as busy
IF tracing DO
writef("%i8: node %i4: node not busy*n", simtime, node)
RETURN
}
// Append item to the end of this queue
IF tracing DO
writef("%i8: node %i4: busy so appending message to end of work queue*n",
simtime, node)
TEST p=-1
THEN wkqv!node := @link // Form a unit list
ELSE { WHILE !p DO p := !p // Find the end of the wkq
!p := @link // Append to end of wkq
}
cowait() // Wait to be activated (by dqitem)
}

AND dqitem(node) BE
// A message has just been processed by this node and is ready to process
// the next, if any.
{ LET item = wkqv!node // Current item (~=0)
UNLESS item DO abort(999)
TEST item=-1
THEN wkqv!node := 0 // The node is no longer busy
ELSE { LET next = item!0
AND co = item!1
wkqv!node := next -> next, -1 // De-queue the item
callco(co) // Process the next message
}
}
104 CHAPTER 3. THE LIBRARY

// ######################## Coroutine Bodies ##########################

AND stopcofn(arg) = VALOF
{ waitfor(stoptime)
IF tracing DO
writef("%i8: Stop time reached*n", simtime)
RESULTIS 0
}
AND messcofn(node) = VALOF
{ qitem(node) // Put the message on the work queue for this node
{ // Start processing the first message
LET prtime = rnd(ptmax) // a random processing time
LET dest = rnd(nodes) + 1 // a random destination node
LET netdelay = ABS(node-dest) // the network delay
IF tracing DO
writef("%i8: node %i4: processing message until %n*n",
simtime, node, simtime+prtime)
waitfor(prtime)
count := count + 1 // One more message processed
IF tracing DO
writef("%i8: node %i4: message processed*n",
simtime, node, dest, simtime+netdelay)
dqitem(node) // De-queue current item and activate the next, if any
IF tracing DO
writef("%i8: node %i4: sending message to node %n to arrive at %n*n",
simtime, node, dest, simtime+netdelay)
waitfor(netdelay)
node := dest // The message has arrived at the destination node
IF tracing DO
writef("%i8: node %i4: message reached this node*n",
simtime, node)
qitem(node) // Queue the message if necessary
// The node can now process the first message on its work queue
} REPEAT
}
// ######################### Main Program ############################
LET start() = VALOF
{ LET seed = 0
LET argv = VEC 50
UNLESS rdargs("-n/n,-s/n,-p/n,-r/n,-t/s", argv, 50) DO
{ writef("Bad arguments for cosim*n")
RESULTIS 0
}
nodes, stoptime, ptmax := 500, 1_000_000, 1000
IF argv!0 DO nodes := !(argv!0) // -n/n
IF argv!1 DO stoptime := !(argv!1) // -s/n
IF argv!2 DO ptmax := !(argv!2) // -p/n
IF argv!3 DO seed := !(argv!3) // -r/n
tracing := argv!4 // -t/s
3.7. COROUTINE EXAMPLES 105

writef("*nCosim entered*n*n")
writef("Network nodes: %n*n", nodes)
writef("Stop time: %n*n", stoptime)
writef("Max processing time: %n*n", ptmax)
writef("Random number seed: %n*n", seed)
newline()
UNLESS initrnd(seed) DO
{ writef("Can’t initialise the random number generator*n")
RESULTIS 0
}
stopco := 0
wkqv, priq, cov := getvec(nodes), getvec(nodes+1), getvec(nodes)
UNLESS wkqv & priq & cov DO
{ writef("Can’t allocate space for the node work queues*n")
GOTO ret
}
FOR i = 1 TO nodes DO wkqv!i, cov!i := 0, 0
priqn := 0 // Number of events in the priority queue
count := 0 // Count of message processed
simtime := 0 // Simulated time
IF tracing DO writef("%i8: Starting simulation*n", simtime)
// Create and start the stop coroutine
stopco := createco(stopcofn, 200)
IF stopco DO callco(stopco)
// Create and start the message coroutines
FOR i = 1 TO nodes DO
{ LET co = createco(messcofn, 200)
IF co DO callco(co, i)
cov!i := co
}
// Run the event loop
{ LET event = getevent() // Get the earliest event
UNLESS event BREAK
simtime := event!0 // Set the simulated time
IF simtime > stoptime BREAK
callco(event!1)
} REPEAT
IF tracing DO writef("*nSimulation stopped*n*n")
writef("Messages processed: %n*n", count)
ret:
FOR i = nodes TO 1 BY -1 IF cov!i DO deleteco(cov!i)
IF cov DO freevec(cov)
IF wkqv DO freevec(wkqv)
IF priq DO freevec(priq)
IF stopco DO deleteco(stopco)
closernd()
RESULTIS 0
fail:
writef("Unable to initialise the simulator*n")
GOTO ret
}
106 CHAPTER 3. THE LIBRARY

3.8 The BMP Graphics Library

The graphics library provides facilities for drawing pictures and outputing them to
file. This library is designed to generate .bmp files representing potentially large im-
ages using 8-bit or 24-bit coloured pixels. It is designed to create files representing 2
dimensional rectangular images using the .bmp file format. It should not be confused
with the SDL and GL libraries (described later) used to generate images on the display
screen suitable for interactive graphics typically used in computer games.
This library is initialised by a call of opengraphics which specifies the size of the
image to be drawn and whether 8 or 24 bit pixels are to be used. It also sets up a palette
of 256 colours if 8 bit pixels are to be used. The graphics header file is in g/graphics.h.
It declares the constants mode8bit, mode8bitalt and mode24bit for use in the call of
opengraphics. It also declares several variables starting at position g grbase which is
declared with value 450 in libhdr.h but can be redefined, if necessary, before inserting
graphics.h. The graphics global variables are as follows.

xsize, ysize
These hold the the number of pixels in each row and column of the canvas.

bmpmode
This is set by opengraphics to mode8bit, mode8bitalt or mode24bit.

bpp
This holds the the number of bytes (1 or 3) per pixel in canvas.

rowlen
This holds bpp*xsize, the number of bytes in canvas to represent a row.

canvassize
This holds the number of bytes (rowlen*ysize) in canvas.

canvasupb
This holds the UPB of canvas in words.

canvas
This holds the vector, allocated by getvec(canvasupb), of pixel bytes to represent
the image. Each pixel is either an 8-bit byte identifying a colour in the palette or 3
bytes giving the blue, green and red components of the colour directly.

palettev
If 8-bit pixels are being used, this holds the palette vector of 256 colours. The
colours are specified by values of the form #Xrrggbb in the least significant 24 bits of
each element of palettev. palettev is set to zero when 24 bit pixels are being used.

col white, col majenta, col blue, col cyan, col green, col yellow, col red,
col black
These variables hold either various 8 or 24 bit colour values.
3.8. THE BMP GRAPHICS LIBRARY 107

currx, curry, currcolour

These variables hold the current pixel location and the current 8 or 24 bit colour.
This library uses the convention that the bottom leftmost pixel has coordinate (0,0).
The direction of the x axis is to the right and the direction of the y axis is up. The
primary use of currx and curry is their use in drawto, drawby and drawch to make
drawing sequences of lines and characters more convenient.

3.8.1 The Graphics Functions

The BMP global functions are defined in g/graphics.b. They are as follows. Except
for opengraphics, closegraphics and wrgraph they are the same as those in the SDL
library.

opengraphics(xsize, ysize, mode)

This function sets bmpmode to mode and allocates the vector canvas. If 8 bit pixels
are specified by mode it allocates palettev and fills it with one of two sets of palette
colours. It also initialised all the other graphics variables. The canvas is initially filled
with white pixels (like a blank sheet of paper).

closegraphics()
This function closes the graphics library returning canvas to freestore and if
palettev was allocated it is also returned.

drawpoint(x, y)
This function places a pixel with colour specified by currcolour at position (x, y)
on the canvas.

drawpoint33(x, y)
This function places a 3x3 square of pixels with the colour specified by currcolour
centred at position (x, y) on the canvas.

drawch(ch)
This function draws a 8x12 array of pixels representing the given character. Its
colour is specified by currcolour on a white background. The bottom leftmost pixel
of the character is at (currx,curry). If ch is ’*n’, currx is set to 10 and curry is
decremented by 14, otherwise currx is incremented by 9.

drawstr(x, y,str)
This function calls drawch for each character in the given string starting at position
(x, y),

moveto(x, y)
This function sets currx and curry to x and y, respectively.

moveby(dx, dy)
This function increments currx and curry by dx and dy, respectively.

drawto(x, y)
108 CHAPTER 3. THE LIBRARY

This function draws a straight line of colour currcolour from (currx, curry) to
(x, y). It leaves (currx and curry) to x and y.

drawby(dx, dy)
This function draws a straight line of colour currcolour from (currx, curry) to
(currx+dx, curry+dy). It then increments currx and curry by dx and dy, respec-
tively.

drawrect(x, y, w, h)
This function draws the outline of the rectangle of width w and height h with the
bottom left corner at (x, y) using currcolour.

drawrndrect(x, y, w, h, radius)
This function draws the outline of the rectangle of width w and height h with its
bottom left corner at (x, y) with rounded corners of given radius. Its colour is specified
by currcolour. If radius is less than or equal to zero the corners are square, and if
radius is greater than half the shorter side length it is reduced to this value. currx and
curry are set to x1 and y1, respectively.

fillrect(x, y, w, h)
This function draws a rectangle of width w and height h with its bottom left corner
at (x, y). It is filled with the colour specified by currcolour.

fillrndrect(x, y, w, h, radius)
This function draws the rectangle of width w and height h with rounded corners of
given radius. The bottom left corner is at (x, y). It colour is specified by currcolour.
If radius is less than or equal to zero the corners are square, and if radius is greater
than half the shorter side length it is reduced to this value.

drawcircle(x, y, radius)
This function draws a circle centred at (x, y) with given radius. Its colour is specified
by currcolour.

fillcircle(x, y, radius)
This function draws a filled circle centred at (x, y) with given radius. Its colour is
specified by currcolour.

wrgraph(filename)
This function writes the image held in canvas to the given file in .bmp format. The
image is (currently) scaled to 300 DPI which corresponds to 11811 pixels per metre.
At this scale the size of an A4 page is 2490x3510 pixels.
There are two programs to illustrate how this graphics library can be used. They
are bcplprogs/tests/grtst.b and bcplprogs/tests/grpalette.b. If you are using
BCPL under Linux you can compile and run them as follows.
3.8. THE BMP GRAPHICS LIBRARY 109

cd ~/distribution/BCPL/bcplprogs/test
cintsys
c b grtst
grtst
ctrl-c
gimp grtst.bmp
ctrl-c
cintsys
c b grpalette
grpalette b8
ctrl-c
gimp palette.bmp
ctrl-c
cintsys
grpalette b24
ctrl-c
gimp palette.bmp

The image displayed by the last call of gimp is shown in figure 3.11 illustrating
some of the colours available when using 24 bit pixels.

Figure 3.11: The image created by grpalette b24

110 CHAPTER 3. THE LIBRARY

3.9 The SDL Graphics Library

The SDL Graphics Library implemented in C is available for many platforms including
Linux, Windows and OSX and BCPL has an interface with this library allowing the
user to create a window on the screen and repeatedly draw simple images allowing
simple interactive games to be implemented. It also provides access to the keyboard,
the mouse and joysticks. In due course this interface will allow the generation of sound.
To include these features it is necessary to install the SDL libraries on you machine
and then build cintsys using a Makefile such as MakefileSDL or MakefileRaspiSDL.
The SDL operations are invoked by calls of the form sys(Sys sdl,...). There
is a header file (g/sdl.h) declaring the various constants and globals available, and
g/sdl.b contains the definitions of several functions providing the interface. The con-
stant g sdlbase is set in libhdr.h to be the first global used in the SDL library. It
can be overridden by re-defining g sdlbase before GETting sdl.h.
A program using the SDL library should start with the following lines.

GET "libhdr"
MANIFEST { g_sdlbase=nnn } // Only used if the default setting of 450 in
// libhdr is not suitable.
GET "sdl.h"
GET "sdl.b" // Insert the library source code
.
GET "libhdr"
MANIFEST { g_sdlbase=nnn } // Only used if the default setting of 450 in
// libhdr is not suitable.
GET "sdl.h"

There are several programs that use SDL described in Chapter 4 of bcpl4raspi.pdf
available from my home page.

3.9.1 sdl.h details

The BCPL SDL functions make use of functions provided by the SDL library imple-
mented in C. These sometimes use machine addresses pointing to structures such as
those representing windows and surfaces. Such addresses are stored in BCPL in a pair
of words as described in the description of scb fd on page 51.
The global variables screen, screen1, currsurf, currsurf1, format, format1,
joystick and joystick1 hold address pairs for The SDL window, the currently se-
lected surface, the format structure for that window and a joystick.
A call of mkscreen sets the variables screenxsize and screenysize to the number
of pixels in the width and height of the created window. It alse sets fscreenxsize
and fscreenysize to the floating point versions of these variables. The variable
fscreencentrex and fscreencentrey are the floating point coordinates of the center
of the screen.
3.9. THE SDL GRAPHICS LIBRARY 111

The width and height of the currently seclected surface is held in currxsize and
currysize.
The vectors leftxv, leftzv, rightxv and rightzv are used by the functions such
as drawtriangle and drawtriangle3d defined in g/sdl.b that draw filled objects.
The variables miny and maxy are also used by these functions.
The vector depthv and variables miny and maxy are used by functions in g/sdl.b
and should not be touched by the user. The 3D drawing functions use depthv
to implement the hiding of pixels that are further from the eye than other pixel
at the same position on the screen. The upperbound of depthv is depthvupb
(=screenxsize*screenysize-1). The elements of depthv are scaled integers with
zfac units of depth corresponding to a distance of one pixel. The variab;e zfac is
actually a floating point number since the 3D drawing functions typiclally take floating
point pixel coordinates. If zfac is too small the the line of intersection of two nearly
parallel plane can become inaccurate. The maximum allowable scaled depth is held in
maxdepth (=-1 000 000 000).
Whenever anything is drawn it is given the colour held in the variable currclour.
The variables currx and curry give the starting position of 2D lines and characters.
They are updated after each line or characte is drawn. This allows long sequences of
2D lines or charaters to be drawn conveniently.
There is a similar mechanism for drawing 3D lines using the variables currx3d,
curry3d and currsz3d. These hold the integer x and y pixel coordindates of the next
3D line to be drawn with currsx3d being its scaled integer depth.
The variables mousex and mousey hold the current pointer position 0n the screen,
and mousebuttons is a bitpattern indicating which mouse buttons are currently
pressed.
Some or all of the variables eventtype, eventa1, eventa2, eventa3, eventa4 and
eventa5 are set by the function pollevents as described below.

sdle\_active
sdle\_keydown
sdle\_keyup
sdle\_mousemotion
sdle\_mousebuttondown
sdle\_mousebuttonup
sdle\_joyaxismotion
sdle\_joyballmotion
sdle\_joyhatmotion
sdle\_joybuttondown
sdle\_joybuttonup
sdle\_quit
sdle\_syswmeven
sdle\_videoresize
sdle\_userevent

sdle\_arrowup
sdle\_arrowdown
112 CHAPTER 3. THE LIBRARY

sdle\_arrowright
sdle\_arrowleft

sdl\_init_everything

sdl\_SWSURFACE // Surface is in system memory

sdl\_HWSURFACE // Surface is in video memory

sdl\_ANYFORMAT // Allow any video depth/pixel-format

sdl\_HWPALETTE // Surface has exclusive palette
sdl\_DOUBLEBUF // Set up double-buffered video mode
sdl\_FULLSCREEN // Surface is a full screen display
sdl\_OPENGL // Create an OpenGL rendering context
sdl\_OPENGLBLIT // Create an OpenGL context for blitting
sdl\_RESIZABLE // This video mode may be resized
sdl\_NOFRAME // No window caption or edge frame

3.9.2 Functions defined in sdl.b

This section describes all the functions defined in g/sdl.b in alphabetical order.

alloc2dvecs()
this function is only used in drawtriangle. It allocates and initialises the vectors
leftxv and rightxv.

alloc3dvecs()
this function is only used in drawtriangle2d. It allocates and initialises the vectors
leftxv, lefttszv, rightxv and rightszv. .

blitsurf(srcptr, dstptr, x, y)
This copies the source surface into the specified position of the destination surface.

bltsurfrect(srcptr, srcrect, dstptr, x, y)

Copy the specified rectangle from the source surface to the specified position in the
destination surface.

rc := closesdl()
This closes down the SDL library returning all allocated space to freestore.

crossprod(v1, v2, v3)

This computes the cross product of v1 and v2 which are both vectors with three
floating point elements. The result is thus a vector orthogonal to v1 and v2 whose
length is the sine of the angle between the vectors multiplied by the product of their
lengths. The vectors v1, v1 and v1 are in right handed orientation.

drawby(dx, dy)
This is just calls drawto(currx+dx, curry+dy).
3.9. THE SDL GRAPHICS LIBRARY 113

drawby3d(FLT dx, FLT dy, FLT dz)

This is just calls drawto3d(currx+dx, curry+dy, currz+dz).

drawch(ch)
Draw a 12x8 character at the position specified b y and curry and increment curry
by 9. If ch was ’*n’ set currx to 10 and decrement curry by 11.

drawcircle(x0,y0, radius)
Not yet described

drawf(x, y, form, a, b, c,..., t)

Not yet described

drawfillcircle(x, y, radius)
Not yet described

drawfillrect(x0,y0, x1,y1)
Not yet described

drawfillrndrect(x0, y0, x1, y1, radius)

Not yet described

drawpoint(x, y)
This draws a point at location (x,y) on the currently selected surface. It colour is
te one set by the most recent call of setcolour.

drawpoint3d(FLT x, FLT y, FLT z)

This draws a point at location (x,y) on the currently selected surface. If the z
value at that position is greater than z the pixel is not updated.

drawpoint3di(x, y, sz)
Not yet described

drawquad(x1,y1, x2,y2, x3,y3, x4,y4)

Not yet described

drawquad3d(FLT x1, FLT y1, FLT z1, FLT x2, FLT y2, FLT z2, FLT x3, FLT
y3, FLT z3, FLT x4, FLT y4, FLT z4)
Not yet described

drawrect()
Not yet described

drawrndrect()
Not yet described

drawstr()
Not yet described
114 CHAPTER 3. THE LIBRARY

drawto()
Not yet described

drawto3d()
Not yet described

drawto3d()
Not yet described

drawto3di()
Not yet described

drawtriangle()
Not yet described

drawtriangle3d(FLT x1, FLT y1, FLT z1, FLT x2, FLT y2, FLT z2, FLT x3,
FLT y3, FLT z3)
Not yet described

drawwrch(ch)
Not yet described

fillsurf(col)
Not yet described

freesurface(surfptr)
Not yet described

getevent()
Not yet described

getmousestate()
Not yet described

hidecursor()
Not yet described

rc := initsdl()
This initialises the SDL library and sets the global variables used by sdl.b. It
must be called before any other SDL operations can be performed. It returns TRUE if
successful.

res := inprod(v1, v2)

This returns the inner product of v1 and v2 which are both vectors with three
floating point elements. The result is thus the cosine of the angle between the vectors
multiplied by the product of their lengths.
3.9. THE SDL GRAPHICS LIBRARY 115

colour := maprgb(r,g,b)
This return a value representing the colour specified by its r, r and r components.
It uses the colour represention chosen during the call of mkscreen.

rc := mkscreen(title, xsize, ysize)

This creates a window of specified size with the given title. It returns TRUE if
sucessful.

mksurface(w, h, surfptr)
Not yet described

moveby(dx, dy)
This is just calls moveto(currx+dx, curry+dy).

moveby3d(FLT dx, FLT dy, FLT dz)

This is just calls moveto3d(currx+dx, curry+dy, currz+dz).

moveto(x, y)
This selects position (x,y) in the currently selected surface used by subsequent
calls of drawch, drawto and drawby. Its depth coordinate is given the value zero.

moveto3d(FLT x, FLT y, FLT z)

This selects position (x,y,z) in the currently selected surface used by subsequent
calls of drawch, drawto and drawby. By convention smaller values of z are deeper into
the screen.

sdldelay(msecs) This causes a real time delay of the specified number of millisec-
onds.

sdlmsecs()
This return the number of real time milliseconds since the current command was
entered.

selectsurface(surfptr, xsize, ysize)

This selects a surface for use in subsequent drawing commands. The arguments
xsize and ysize specify the size of the surface in pixels. The pair of BCPL word used
to represent the machine address of the surface is pointed to by surfptr. See scb fd
on page ?? for more details.

rc := setcaption(title)
This resets the title of the window created by mkscreen. It returns TRUE if success-
ful.

setcolour(col)
This sets the colour to be used in subsequent drawing commands. The colour should
be one returned by a call of maprgb.
116 CHAPTER 3. THE LIBRARY

setcolourkey(col)
This sets the colour that has the special property that when an attempt is made
to to draw a pixel with this colour the pixel is left with its previous colour. This
mechanism is used, for example, when displaying the moving coloured circles by the
program bcplprogs/raspi/bucket.b.

setlims(x0,y0, x1,y2) This function is used by drawtriangle which draws a

filled 2D triangle. It updates entries in leftxv and rightxv for each value of y between
y0 and y1.

setlims3d(x0,y0,sz0, x1,y2,sz1) This function is used by drawtriangle3d

when drawing a filled 3D triangle. It updates entries in leftxv, leftszv, rightxv
and rightszv for each value of y between y0 and y1. The arguments are all integers
with the z components being scaled to cause zfac units to correspond to a distance of
one pixel.

showcursor()
This causes the cursor to be displayed.

standardize(v)
This divides the three floating point elements of the vector v by length of the vector
leaving the elements of v set to the direction cosines of the given vector.

updatescreen()
This causes the surface that is currently being drawn to be copied the display
screen.

write ch slice(x, y, ch, line)

This function is used by drawch to plot a row pixels of a 8x12 character.
The pixels to be drawn on the screen by the next call of updatescreen are held
in a vector pointed to by the global variable screen. In order to implement hidden

surface removal there is a second vector depthv holding the z coordinates of pixels in
screen. This vector is only used by the 3D drawing functions. The depth values were
held as scaled integers with zfac units corresponding to a distance of one pixel.
This eliminated some of the problems caused by using integers to represent depth
which was particularly noticable when two nearly parallel 3D plane interected. Un-
fortunately floating point numbers caused other problems mainly due to their lack of
precision so integers are again going to be used but 64 units of an element of depthv
will now represent a distance of one pixel. Even though these units are used in depthv,
all the 3D drawing functions will use the convention that one unit in x, x and x will
represent a distance of one pixel. The fractional values in depthv only occur when
drawing 3D lines and triangles. The vertices of lines and triangles always occur on
pixel boundaries.
3.9. THE SDL GRAPHICS LIBRARY 117

3.9.3 sys(Sys sdl,...) calls

rc := sys(Sys sdl, sdl avail

This return TRUE if the SDL facilities are available.

rc := sys(Sys sdl, sdl init)

This ...

rc := sys(Sys sdl, sdl setvideomode, width, height, bbp, flags)

rc := sys(Sys sdl, sdl quit)

rc := sys(Sys sdl, sdl locksurface, surfptr)

rc := sys(Sys sdl, sdl unlocksurface, surfptr)

rc := sys(Sys sdl, sdl getsurfaceinfo, surfptr, ptr)

rc := sys(Sys sdl, sdl getfmtinfo. fmtptr)

rc := sys(Sys sdl, sdl geterror, str)

rc := sys(Sys sdl, sdl updaterect, surfptr, left, top, right, bottom)

rc := sys(Sys sdl, sdl loadbmp, filename of a .bmp image)

rc := sys(Sys sdl, sdl blitsurface. src, srcrect, dest, destrect)

rc := sys(Sys sdl, sdl setcolourkey, surfptr, flags, colorkey)

rc := sys(Sys sdl, sdl freesurface, surfptr)

rc := sys(Sys sdl, sdl setalpha, surfptr, flags, alpha)

118 CHAPTER 3. THE LIBRARY

rc := sys(Sys sdl, sdl imgload, filename)

rc := sys(Sys sdl, sdl delay, msecs)

rc := sys(Sys sdl, sdl flip, surfptr)

rc := sys(Sys sdl, sdl displayformat, surfptr)

rc := sys(Sys sdl, sdl waitevent, pointer)

rc := sys(Sys sdl, sdl pollevent, pointer)

rc := sys(Sys sdl, sdl getmousestate, pointer)

rc := sys(Sys sdl, sdl loadwav, file, spec, buff, len)

rc := sys(Sys sdl, sdl freewav, buffer)

rc := sys(Sys sdl, sdl wm setcaption, string)

rc := sys(Sys sdl, sdl videoinfo, v)

rc := sys(Sys sdl, sdl maprgb. formatptr, r, g, b)

rc := sys(Sys sdl, sdl drawline,)

rc := sys(Sys sdl, sdl drawhline, )

rc := sys(Sys sdl, sdl drawvline,)

rc := sys(Sys sdl, sdl drawcircle, )

3.9. THE SDL GRAPHICS LIBRARY 119

rc := sys(Sys sdl, sdl drawrect, )

rc := sys(Sys sdl, sdl drawpixel, )

rc := sys(Sys sdl, sdl drawellipse, )

rc := sys(Sys sdl, sdl drawfillellipse, )

rc := sys(Sys sdl, sdl drawround. )

rc := sys(Sys sdl, sdl drawfillround, )

rc := sys(Sys sdl, sdl drawfillcircle, )

rc := sys(Sys sdl, sdl drawfillrect, )

rc := sys(Sys sdl, sdl fillrect, )

rc := sys(Sys sdl, sdl fillsurf, )

rc := sys(Sys sdl, sdl numjoysticks)

rc := sys(Sys sdl, sdl joystickopen, index, jpyptr)

rc := sys(Sys sdl, sdl joystickclose, index)

rc := sys(Sys sdl, sdl joystickname, index)

rc := sys(Sys sdl, sdl joysticknumaxes, joyptr)

rc := sys(Sys sdl, sdl joysticknumbuttons, joyptr)

120 CHAPTER 3. THE LIBRARY

rc := sys(Sys sdl, sdl joysticknumballs, joyptr)

rc := sys(Sys sdl, sdl joysticknumhats, joyptr)

rc := sys(Sys sdl, sdl joystickeventstate, aeg

rc := sys(Sys sdl, sdl getticks)

rc := sys(Sys sdl, sdl showcursor)

rc := sys(Sys sdl, sdl hidecursor)

rc := sys(Sys sdl, sdl mksurface)

rc := sys(Sys sdl, sdl setcolourkey)

rc := sys(Sys sdl, sdl joystickgetbutton)

rc := sys(Sys sdl, sdl joystickgetaxis)

rc := sys(Sys sdl, sdl joystickgetball)

rc := sys(Sys sdl, sdl joystickgethat)

3.10 The GL Graphics Library

This library is still under development
OpenGL is a sophisticated graphics library allowing 3D images to be drawn on
the screen efficiently using the full power of the graphics hardware available on most
machines. On most desktop and laptop machines the full OpenGL library is available,
but on handheld devices only a simplified version called OpenGL ES is available. The
BCPL interface is designed to work with whichever version of OpenGL is available.
This library essentially provides a subset of the OpenGL ES features. Note that the
GL interface on the Raspberry Pi uses OpenGL ES.
3.11. THE SOUND LIBRARY 121

To include these features in cintsys it is necessary to install the OpenGL li-

braries on you machine and then build cintsys using a Makefile such as MakefileGL,
MakefileRaspiGL or MakefileVCGL.
The GL library uses the sys(Sys gl,...) functions. There is a header file
(g/gl.h) declaring the various constants and globals available in the GL library, and
g/gl.b contains the definitions of several functions providing the interface to OpenGL.
The constant g glbase is set in libhdr to be the first global used in the GL library.
It can be overridden by re-defining g glbase after GETting libhdr.
A program wishing to use the OpenGL library should start with the following lines.

GET "libhdr"
MANIFEST { g_glbase=nnn } // Only used if the default setting of 450 in
// libhdr is not suitable.
GET "gl.h"
GET "gl.b" // Insert the library source code
.
GET "libhdr"
MANIFEST { g_glbase=nnn } // Only used if the default setting of 450 in
// libhdr is not suitable.
GET "gl.h"

This library will be described in Chapter 5 of bcpl4raspi.pdf available from my home

page.

3.11 The Sound Library

This library is under development
The sound library uses the sys(Sys sound,...) functions to provide facilities for
reading, writing and analysing sound data. There is a sound header file (g/sound.h)
declaring various constants and globals available in the sound library. The sound library
itself is in g/sound.b and can be inserted into a program by the following statements.

GET "libhdr"
MANIFEST { g_sndbase=nnn } // Only used if the default setting of 400 in
// libhdr is not suitable.
GET "sound.h"
GET "sound.b" // Insert the library source code

The manifest constant g sndbase specifies the position of the first global variable to
be used by the sound library.

3.11.1 The Sound Constants

The sound library is not yet available.
122 CHAPTER 3. THE LIBRARY

3.11.2 The Sound Global Variables

The sound library is not yet available.

3.11.3 The Sound Functions

The sound library is not yet available.

3.12 The EXT Library

This library is designed to allow users to construct their own extension library involving
code in C and assembly language. Its structure is similar to that of the SDL and GL
libraries.
It uses the sys(Sys ext,...) functions to interface with C code defined in
sysc/extfn.c, and has two header files ext.h and ext.b providing the BCPL in-
terface. Programs using the EXT library should start with the following statements.

GET "libhdr"
MANIFEST { g_extbase=nnn } // Only used if the default setting of 900 in
// libhdr is not suitable.
GET "ext.h"
GET "ext.b" // Insert the library source code
Chapter 4

The Command Language

The Command Language Interpreter (CLI) is a simple interactive interface between the
user and the system. It loads and executes previously compiled programs that are held
either in the current directory or one of the directories specified by the shell environ-
ment variable (typically BCPLPATH or POSPATH) whose name is in rootnode!rtn path.
These commands are described in Section 4.3 and their source code can be found in
the com directory. The command language is a combination of the features provided
by the CLI and the collection of commands that can be invoked. Under Cintpos, a
similar CLI program provides command language interpreters in several contexts such
as those created by the commands: run, newcli, tcpcli and mbxcli. Details of the
implementation of both CLIs are given at the end of this chapter from page 155.
Commands can set a return code in the global returncode with zero meaning
successful termination and other values indicating the severity of the fault. Commands
that set a non zero return code are expected to leave a reason code in result2. The
CLI copies the return code and reason code of the previous command into the CLI
variables cli returncode and cli result2, respectively. These can be inspected by
commands such as if and why and also used by the CLI to terminate a command-
command if the failure was severe enough. For details, see the command failat on
page 140 below.

4.1 Bootstrapping Cintsys

When Cintsys is started, control is passed to the interpreter which, after a few initial
checks, allocates vectors for the memory of the Cintcode abstract machine and the tally
vector available for statistics gathering. The Cintcode memory is initialised suitably
for sub-allocation by getvec, which is then used to allocate space for the root node,
the initial stack and the initial global vector. The initial state shown in figure 4.1 is
completed by loading the object modules SYSLIB, BLIB and BOOT, and initialising the
root node, the stack and global vector. Interpretation of Cintcode instructions now
begins with the Cintcode register PC, P and G set as shown in the figure, and Count set
to -1. The other registers are cleared. The first Cintcode instruction to be executed is
the first instruction of the body of the function start defined in sysb/boot.b. Since

123
124 CHAPTER 4. THE COMMAND LANGUAGE

no return link has been stored into the stack, this call of start must not attempt
to return in the normal way; however, its execution can still be terminated using
sys(Sys quit,0).
The global vector and stack shown in figure 4.1 are used by start and form the
running environment both during initialization and while running the debugger. The
CLI, on the other hand, is provided with a new stack and a separate global vector,
thus allowing the debugger to use its own globals freely without interfering with the
command language interpreter or running commands. The global vector of 1000 words
is allocated for the CLI and this is shared by the CLI program and its running com-
mands. The stack, on the other hand, is used exclusively by the command language
interpreter since it creates a coroutine for each command it runs.
Tally vector

blklist

stack globals

rootnode PC P G

Entry to start

SYSLIB BLIB BOOT 0

Figure 4.1: The initial state

Control is passed to the CLI by means of the call sys(Sys interpret,regs) which
recursively enters the intepreter from an initial Cintcode state specified by the vector
regs in which that P and G are set to point to the bases of a new stack and a new global
vector for CLI, respectively, PC is the location of the first instruction of startcli, and
count is set to -1. This call of sys(Sys interpret,regs) is embedded in the loop
shown below that occurs at the end of the body of start.
{ LET res = sys(Sys_interpret, regs) // Call the interpreter
IF res=0 DO sys(Sys_quit, 0)
debug res // Enter the debugger
} REPEAT
At the moment sys(Sys interpret,regs) is first called, only globsize, sys and
rootnode have been set in CLI’s global vector and so the body of startroot must
be coded with care to avoid calling global functions before their entry points have be
placed in the global vector. Thus, for instance, instead of calling globin to initialise
the globals defined in BLIB, SYSLIB and DLIB, the following code is used:
sys(Sys_globin, rootnode!rtn_blib)
If a fault occurs during the execution of CLI or a command that it is running,
the call of sys(Sys interpret,regs) will return with the fault code and regs will
4.2. BOOTSTRAPPING CINTPOS 125

hold the dumped Cintcode registers. A result of zero, signifying successful completion,
causes execution of Cintsys to terminate; however, if a non zero result is returned, the
debugger in entered by means of the call debug(res). Note that the Cintcode registers
are available to the debugger since regs is a global variable. When debug returns,
the REPEAT-loop ensures that the command language interpreter is re-entered. The
debugger is briefly described in the Chapter 7.
On entry to startroot, the coroutine environment is initialised by setting currco
and colist to point to the base of the current stack which is then setup as the root
coroutine. The remaining globals are the initialised and the standard input and output
streams opened before loading the CLI program by means of the following statement:
rootnode!rtn_cli := globin(loadseg("syscin/cli"))

The command language interpreter is now entered by the call start().

4.2 Bootstrapping Cintpos

Bootstrapping Cintpos is somewhat more complicated than bootstrapping Cintsys since
there are more resident modules of code, and the Cintpos system structures and resident
tasks must be set up. Bootstrapping starts when the cintpos program is entered. It
first decodes the command arguments, possibly changing the Cintcode memory or tally
vector sizes. It then allocates these vectors, initialising every word of the Cintcode
memory with the value #xDEADCODE. It also allocates a vector to hold counts of how
many blocks of each requested size have been allocated getvec but not yet freed. It
then allocates and initialises the stack and global vector to be used by BOOT. The
rootnode is then initialised, including the setting of the fields: rtn boot (holding the
module boot), rtn klib (holding the module klib), rtn blib (holding the modules
blib, syslib and dlib) and rtn sys (holding the entry point to the function sys).
The initial values of the Cintcode registers are now placed in the register set
bootregs. The Cintcode interpreter is entered to start execution from this initial
state. If the interpreter returns a non zero result, a message containing this value is
written to the standard output stream, and, if the rtn dumpflag field of the root node
is TRUE, the entire Cintcode memory is dumped to the file DUMP.mem in compacted form
suitable for inspection by commands such as dumpsys or dumpdebug.

4.2.1 The Cintpos BOOT module

The function start in boot is the very first BCPL compiled code to be entered when
Cintpos starts. On entry, the Cintcode registers A, B and C are zero, P and G point
to BOOT’s stack and global vector, and ST is set to 2, indicating that we are in boot
and that interrupts are disabled. The global vector has already been initialised to hold
all the entry points in boot, klib, blib, syslib and dlib, but the stack currently is
filled entirely with the value stackword=#xABCD1234 except for its zeroth word which
was set by cintpos to hold the stacksize. To improve the behaviour of the standalone
debugger, this stack is turned into a root coroutine stack of the specified size, initialising
the globals currco and colist appropriately.
126 CHAPTER 4. THE COMMAND LANGUAGE

All console input and output within BOOT and the standalone debugger is done us-
ing the standalone version of rdch and wrch, so these globals are updated appropriately.
BOOT next initialises the variables used by the standalone debugger. These include
the vectors bpt addr, bpt instr and bpt dbgvars which respectively hold breakpoint
addresses, breakpoint instructions that have been overwritten by the BRK instruction,
and the vector of the 10 standalone debugger variables V0 to V9. These three vectors
are placed in the rootnode to make them accessible both to the DEBUG task and to
dumpdebug when it is inspecting a system dump.
BOOT now creates and initialises a global vector and a stack to be used during the
further initialisation of the Cintpos system. The all elements of the global vector are
given values of the form globword(=#x8F8F0000)+n, except for the globals globsize,
sys, rootnode, currco and colist, the last two being set to zero. Every element of
the stack is set to stackword (=#xABCD1234). The register set klibregs is initialised,
giving zero to A, B and C, the stack and global vector pointers to P and G, the value
one to ST to indicate execution is in KLIB and interrupts are disabled, and the entry
point startroot in PC. This register set is then handed to a recursive call of the
interpreter. This inner call is the one than performs the rest of the initialisation and
enters the normal execution of Cintpos. In due course the interpreter will return with
a completion code which controls what BOOT should do next.
A completion code of zero signifies successfully completion and BOOT causes the
termination of cintpos. A return code of -1 is special, causing BOOT to re-enter
the interpreter immediately. Its purpose is to allow a running program to change
which interpreter is used. There are typically two interpreters: a slow one in which all
debugging aids are turned on, and a fast one in which most aids are turned off. The call
sys(Sys interpret, regs) selects the fast interpreter if the count register in regs
is -1, otherwise it selects the slow interpreter. The return code -2 allows a running
program to invoke the dumpmem mechanism to write the file DUMP.mem representing
the current state of the entire Cintcode memory. All other completion codes causes
BOOT to invoke the standalone debugger.
BOOT cunningly places a private version of the sys function in its global vector
so that, even if a breakpoint is set in the public version of sys, BOOT and in partic-
ular the standalone debugger can continue to work as normal. When BOOT invokes
the interpreter for the first time execution begins at the start of startroot which is
described in the next section.

4.2.2 startroot
This function creates the Cintpos running environment and loads all the resident system
tasks. Finally it enters the Cintpos scheduler which, in turn, gives control to the Idle
task which sends a packet to the root CLI task. After some initialisation, this issues the
first CLI prompt inviting the user to type in a command. Knowledge of the underlying
structures used by Cintpos is key to understanding how Cintpos works. They are
described in this section in the order in which startroot creates them.
startroot is entered by the recursive call of interpret from BOOT with a new
stack and a different global vector from that used by BOOT. If the interpreter sub-
4.2. BOOTSTRAPPING CINTPOS 127

sequently detects a fault it returns to BOOT’s running environment giving control to

the interactive debugger allowing the user to inspect the stack and global vector that
were current at the time the fault.
Althought startroot has three formal parameters fn, size and c, it was entered
in a non standard way and these have not been given values. However, the base of
startroot’s stack is at @fn-3. This points to the zeroth element holding the stack size
with all other elements are already set by BOOT to stackword (#xABCD1234). This
stack is turned into a coroutine stack by updating its bottom six elements appropriately.
Care is taken to ensure that the code that performs this initialisation is not itself using
the stack locations that it is updating. This is one of the reasons why startroot was
given three parameters.
The function rootcode is now called to create the Cintpos resident structures. At
this moment the base of the global vector is at @globsize (=Global 0), all its elements
are filled with words of the form globword+n (=#8F8F0000+n), except for globsize
which holds the upper bound of the global vector, sys which holds the entry point
of the sys function, rootnode which points to the rootnode, and currco and colist
which both point to the newly created coroutine stack. The other globals are now
initialised by two calls of sys(Sys_globin,...).
Cintpos has two vectors tasktab and devtab that provide access to all Cintpos
tasks and devices. These are allocated and cleared, and pointers to them are placed in
the rootnode.
The resident Cintpos devices are now created. These have device identifiers -1,
-2 and -3 corresponding to the clock, the keyboard and the screen. Most Cintsys
devices are implemented using separate threads of the underlying operating system.
Such devices have device control blocks (DCBs) held their entries in devtab. A DCB
has fields used for communication between its device thread and the interpreter. One
of these is the work queue of packets sent by client tasks but not yet processed by the
device. It has been found that interaction with some device threads is too slow to be
satisfactory and so have been replaced by an implementation based on polling by the
interpreter. This currently applies to the clock and screen devices. As far as the user is
concerned, these devices still have the same indentifiers and still work as before but are
faster. An entry in devtab points to a DCB. Devices not using the polling mechanism
use threads of the host operating system, other devices are handled entirely by the
interpreter thread. The only resident devices currently using a separate threads are
the keyboard and TCP devices. Device threads are created using the kernel function
createdev defined in sysb/klib.b, and the C code for the resident device threads can
be found in sysc/devices.c.
The Cintcode abstract machine can receive interrupts. The mechanism is as follows.
If a device wishes to interrupt the interpreter it sets the variable irq to TRUE, and just
before the interpreter starts to execute an instruction, if the Cintpos ST register is
zero (indicating that interrupts are enabled), it saves the current Cintpos registers and
enters the interrupt service routine using the register set in isrregs. The interrupt
service routine has its own stack but shares the same global vector a the Cintpos kernel.
It always starts execution at the start of the function irqrtn with Cintcode register
ST set to 3 to indicate that an interrupt is being serviced. The interrupt sevice routine
128 CHAPTER 4. THE COMMAND LANGUAGE

may return control to the interrupted task or it may enter the scheduler if another task
deserves to gain control.
Before creating the resident tasks, startroot initialises a few more rootnode fields.
These are rtn tcblist and rtn crntask both set to zero since there are currently no
Cintpos tasks, rtn blklist set to the start of the memory block list used by getvec
and freevec, rtn clkintson set to FALSE to globally disable interrupts, rtn clwkq
set to zero representing an empty list of packets for the clock device, and rtn info set
to a cleared table of 50 elements.
The resident tasks are now created using suitable calls of createtask. Each time
createtask is called it allocates a task control block (TCB) giving it the next available
task identifier and updating the appropriate entry in tasktab to point to it. Such tasks
are initially given a state of #b1100 indicating that they are DEAD, not HELD and
have no packets in the work queue. The first task to be created is a special one called
Idle whose body is in cin/syscin/idle and although createtask will have chosen
identifier one for it, this must be replaced by zero and it entry in tasktab removed.
It is given a startup packet and an initial state of #b1101 indicating it is DEAD, not
HELD but has a packet and so can be given control by the scheduler when it is run.
Six more resident tasks are now created, all have state #b1100. They are the root
command language interpreter that initially waits for commands from the keyboard,
and interactive debugging task, the console handler providing communication between
the keyboard and tasks, the file handler providing access to disk files, the mailbox
handler that provides a mechanism that lets tasks send and receive short messages via
named mailboxes and the TCP handler providing TCP/IP communication.
Just after Cintpos starts up the status command will output the following.

Task 1: Root_Cli running CLI Loaded command: status

Task 2: Debug_Task waiting DEBUG
Task 3: Console_Handler waiting COHAND
Task 4: File_Handler waiting FH0
Task 5: MBX_Handler waiting MBXHAND
Task 6: TCP_Handler waiting TCPHAND

Once the kernel structure and all the resident tasks have been set up, the system
can be started by entering the scheduler which is a function called srchwk defined in
sysb/klib.b. It take one argument which is a pointer to the highest priority TCB
that could possibly run. It searches through the chain of TCBs that are linked in
decreasing priority order looking at only the status field of each. This field is sufficient
to tell whether the corresponding task can run or not. It has 4 bits IWHP. The I bit is
a 1 if the task has been interrupted in which case its Cintcode registers will be packed
elsewhere in the TCB. The W bit is a 1 if the task is suspended in taskwait waiting
for a packet to arrive from another task or a device. The H bit is 1 if the task is in
HOLD state indicating that it cannot run even if it otherwise would be ready to do
so, and the P bit is a 1 if the tasks’s work queue is not empty. A task cannot be both
interrupted and waiting for a packet and the setting of both the I and W bits have
a special meaning, namely that the task is in DEAD state having no runtime stack
or global vector. There are thus 16 posible states a task can have of which only six
indicate that it is runnable, they are as follows.
4.3. COMMANDS 129

#b0000
This task is runnable but has no packet on its work queue. It is either the current
task or it gave up control voluntarily by for instance sending a packet to a higher
priority task. When it next gains control it will immediately return from the
function that caused it to give up control.

#b0001
This is just like the case above except there is a packet on its work queue.

#b0101
This indicates that the task is waiting for a packet and that one has arrived.
It is thus runnable and when given control the first packet on its work queue
will be dequeued and returned as the result of the taskwait call that caused its
suspension.

#b1000
This indicates the task is in interrupted state with an empty work queue. It is
thus runnable and when given control it will resume execution using the Cintcode
register values saved in the TCB when it was interrupted.

#b1001
This indicates the task is in interrupted state with a non empty work queue.
It is thus runnable and when given control it will resume execution using the
Cintcode register values save in the TCB when it was interrupted.

#b1101
This is a task in DEAD state (with no stack or global vector) but it now has a
startup packet on its work queue. It is thus runnable and when given control will
be initialised with a new stack and global vector and its main function start
in global variable 1 will be called with the startup packet as its first argument.
This packet will have been dequeued.

4.3 Commands
This section describes the Command Language Interpreter commands whose source
code can be found in either cintcode/com or cintpos/com. The rdargs argument
format string for each command is given.

abort NUMBER CIN:y, POS:y, NAT:y

The command: abort n calls the BLIB function abort with argument n. If n is
zero, this causes a successful return from the BCPL system. If n is non zero, the
interactive debugger is entered with fault code n. The default value for n is 99. The
interactive debugger is described in section 7.

adjclock OFFSET CIN:y, POS:y, NAT:y

The syntax of the OFFSET argument is [-][h][:m], that is: an optional minus sign,
followed by an optional number of hours, possibly followed by :m to specify a number
130 CHAPTER 4. THE COMMAND LANGUAGE

of minutes. The offset is converted into a signed integer representing the number of
minutes to be added to the time of day as supplied by the system. If adjclock is not
given an argument, it just outputs the current offset.

alarm AT/A,MESSAGE CIN:n, POS:y, NAT:n

This command is only available under Cintpos. Its first parameter has the format:
[+][[hours:]minutes:]seconds. If + is present the time is relative to now. The
command suspends itself until the specified time, then outputs the time followed by
the message. Typical usage is as follows:

run alarm +3:30 "Your time is up!"

After three and a half minute a message such as the following will appear.

*** Alarm: time is 15:13:14 - Your time is up!

append FROM,TO/K CIN:y, POS:y, NAT:y

This command appends the FROM file on to the end of the TO file. If the TO file does
not initially exist, an empty one is created.

bbcbcpl FROM/A,TO/K,REPORT/K,NONAMES/S,MAX/S,SECTLEN/S
CIN:y, POS:y, NAT:y
This invokes a reconstruction of the BCPL compiler for the BBC Microcomputer
marketed by my brother’s company RCP in the early 1970s. The FROM argument
specifies the BCPL source file. The TO argument specifies the desination file for the
compiled 16-bit Cintcode. The REPORT argument specifies a file to hold error messages.
The NONAMES argument causes the compiler not to embed function names in the com-
piled code. The MAX argument has no effect in this version of the compiler and SECTLEN
controls whether the length of a section of compiled code includes it length in its first
word.
This reconstruction was created to possibly help reconstruct the BBC Domes-
day Project that ran on the BBC Microcomputer using a Philips 12” laser disc.
Related commands are mapcode and prbbcocode. For more information, see
bcplprogs/bbcmicro/ in the standard BCPL distribution.

bbcbcpl32 FROM/A,TO/K,REPORT/K,NONAMES/S,MAX/S,SECTLEN/S
CIN:y, POS:y, NAT:y
This command is similar to bbcbcpl but generates 32-bit Cintcode suitable for the
current BCPL Cintcode system. The BBC BCPL programs may need minor changes
needed because the BCPL word length has increased from 16 to 32 bits.
This command is now obsolete since using bbc2bcpl is a more satisfactory way to
run BBC BCPL on the modern BCPL Cintcode system.
bbcbcpl32 will soon be deleted.

bbc2bcpl FROM/A,TO/K,HARD/S,EQCASES/S,T64/S,-h/S CIN:y, POS:y, NAT:y

4.3. COMMANDS 131

This command convert the BBC BCPL program given by FROM to a destination file
given by TO. The result can be compiled and run under the modern BCPL Cintcode
system. HARD causes the program to abort on every detected error, EQCASES toggles
whether the case of letters in reserved word and identifiers are ignored. By default the
case of letters is ignored. Several minor replacements are made, such as LOGAND, LOGOR,
LSHIFT, EQ and LE are replaved by &, |, <<. = and <=. Missinf close sections brackets are
automatically inserted with the correct indentation when multiple sections are closed
by a tagged closing bracket. All dots in identediters are replaced by underscores. The
cases of letters used in identifiers is modified to agree with the way each identifier was
written when first encountered. To do this the program reads GET files but leave the
GET directives unchanged.

bcpl FROM/A,TO/K,VER/K,SIZE/K/N,TREE/S,NONAMES/S,
D1/S,D2/S,OENDER/S,EQCASES/S,BIN/S,XREF/S,GDEFS/S,HDRS/K,
GB2312/S,UTF8/S,SAVESIZE/K/N,HARD/S,T32/S,T64/S,
OPT/K,TREE2/S,NOSELST/S CIN:y, POS:y, NAT:y
This invokes the BCPL compiler. The FROM argument specified the name of the file
to be compiled. If the TO argument is given, the compiler generates code to the specified
file. Without the TO argument the compiler will output the OCODE intermediate
form to the file ocode as a compiler debugging aid. This file can be converted to a
more readable form using the procode command, described below. The VER argument
redirects the standard output to a named file. The SIZE argument specified the size
of the compiler’s work space. The default is 100,000 words. The NONAMES switch
causes the compiler not include section and function names in the compiled code. The
switches D1 and D2 control compiler debugging output. D1 causes a readable form of the
compiled Cintcode to be output. D2 causes a detailed trace of the internal working of
the codegenerator to be output. D1 and D2 together causes a slightly more detailed trace
of the internal working of the codegenerator. OENDER causes code to be generated for a
machine with the opposite endianess of the machine on which the compiler is running.
EQCASES causes all identifiers to be converted to uppercase during compilation. This
allows very old BCPL programs to be compiled. BIN causes the target Cintcode to be
in binary rather than the ASCII encoded hexadecimal normally used. The XREF option
causes a line to be output by the compiler for each non local identifier occurring in the
program. A typical such line is as follows:

all G:201 LG queens.b[9] all&~(ld|col|rd)

It shows that the variable all was declared as global variable 201 and its was loaded
in the compilation of statements on line 9 of the program queens.b and the context of
its use was: all&~(ld|col|rd). These lines can be filtered and sorted to form a cross
reference listing of a program. See, for instance, the file BCPL/cintcode/xrefdata
or Cintpos/cintpos/xrefdata. If both VER and XREF are specified the xref data is
appended to the verification stream. This allows the xref data generated by several
separate compilations to be concatenated. The resulting file can be filtered and sorted
by the sortxref command. Typical usage is as follows:
132 CHAPTER 4. THE COMMAND LANGUAGE

delete -f rawxref
c compall "ver rawxref xref"
sort rawxref to xrefdata
delete rawxref

The GDEFS switch is a debugging aid to output the global numbers of any global
function defined in the program. For example:

bcpl gdefs com/bench100.b to junk

generates the following output:

BCPL (3 July 2007)

G 1 = start
G259 = trace
G260 = schedule
G261 = qpkt
G262 = wait
G263 = holdself
G264 = release
G270 = idlefn
G271 = workfn
G272 = handlerfn
G273 = devfn
Code size = 1436 bytes

The UTF8 and GB2312 options specify the default encoding for extended characters
in string and character constants. This default can be overridden in individual constants
using the *#u and *#g escape sequences, as described on page 18.
The SAVESIZE option allows the user to specify the number of words in the argument
stack used to hold function return information. The default value is three making room
for the old P pointer, the return address and the entry point of the current function.
When compiling into native code using the Sial mechanism, the save space size may
be different, since, for instance, some or all of this information may be stored in the
hardware (SP) stack.
The HARD options causes both syntax and translation phase errors to call
abort(100). This is useful in commands such as: c compall hard allowing each
error in a long sequence of compilations to be inspected separately.
The arguments T32 and T64 specify whether the target architecture is for 32 or 64
bit BCPL.
The argument OPT gives a list of conditional compilation option names consisting
of letters, digits, underline and dot, separated by plus signs or any other characters
not allowed in option names. These options are declared at the start of compilation of
every BCPL section.
The debugging options TREE and TREE2 cause the parse tree to be output before
and after the the conversions caused by the FLT feature.
The option NOSELST causes the compiler to avoid using the Ocode instructions
SELLD and SELST when compiling the OF operator. Less efficient code is compiler using
shifts and logical instructions. This option causes the bcpl2sial command not to use the
4.3. COMMANDS 133

new SIAL function codes selld, selst and xselst, enabling older Sial codegenerators
to continue to work.

bcpl2sial FROM/A,TO/K,VER/K,SIZE/K/N,TREE/S,NONAMES/S,
D1/S,D2/S,OENDER/S,EQCASES/S,BIN/S,XREF/S,GDEFS/S,HDRS/K,
GB2312/S,UTF8/S,SAVESIZE/K/N,HARD/S,T32/S,T64/S,
OPT/K,TREE2/S,NOSELST/S CIN:y, POS:y, NAT:y

This command compiles a BCPL program into the internal assembly language Sial
which is designed as a low level intermediate target code for BCPL and is described in
Section 10.1. The command sial-sasm, described below, can be used to convert Sial
into a human readable form and various commands, such as sial-386, sial-alpha and
sial-arm will convert Sial to assembly language for corresponding architectures. The
bcpl2sial command uses the same front end as bcpl and so takes the same command
arguments as the bcpl command.

bcplxref FROM/A,TO/K,PAT/K CIN:y, POS:y, NAT:y

This command outputs a cross reference listing of the program given by the FROM
argument. This consists of a list of all identifiers used in the program each having a list
of line numbers where the identifier was used and a letter indicating how the identifier
was declared. The letters have the following meanings:

V Local variable
P Function or Routine
L Label
G Global
M Manifest
S Static
F FOR loop variable

The TO argument can be used to redirect the output to a file, and the PAT argument
supplies a pattern to restrict which names are to be cross referenced. Within a pattern
an asterisk will match any sequence of characters, so the pattern a*b* will match
identifiers such as ab, axxbor axbyy. Upper and lower case letters are equated. This
command has largely been superceded by the xref option in the bcpl command and
the related sortxref command.

bench100 CIN:y, POS:y, NAT:y

This is a simple benchmark program used to test the efficiency of systems imple-
mentation languages.

bgpm FROM,TO/K,UPB/K CIN:y, POS:y, NAT:y

This is an implementation of Christopher Strachey’s GPM macrogenerator. It takes
input from the FROM file if specified, otherwise it reads from the standard input stream.
The TO argument specifies the file to receive the macrogenerated result, otherwise this
is sent to the standard output stream. The UPB argument specified the amount of
memory that bgpm may use.
134 CHAPTER 4. THE COMMAND LANGUAGE

A macro call is enclosed in square brackets ([ and ]) and contains arguments

separated by backslash characters (\). The arguments are macro expanded as they are
read in. To avoid macro expansion text can be enclosed within nested quotation marks
({ and }). On reaching the close square bracket at the end of a macro call, the zeroth
argument is looked up in the environment of defined macros and macrogeneration
continues from the beginning of its value. When the end of this value is reached the
expansion of the call is complete and macrogeneration continues from just after the
closing square bracket. While a macro call is being expanded, a parameter of the form
^n is replaced by a copy of the nth argument of the current call. The number n is given
as a sequence of decimal digits. The character ’‘’ introduces a comment consisting
of all remaining character of the current line followed by all white space characters
including newlines up to but not including the next non white space character. The
following macros are predefined.
[def\name\value]
This causes a macro with the given name and value to be declared.
[set\name\value]
This updates a named macro with a new value which may be truncated if necessary.
[eval\expression]
This evaluate the given integer expression consisting of numbers and the numeric
operators *, /, %, + and -. Parentheses may be used for grouping and spaces may
appear anywhere except within numbers.
[lquote]
[rquote]
These macros expand to the quotation marks { and } respectively.
[eof]
This macro generates the end of file symbol and can be used to terminate input
from the standard input stream.
A simple definition and call is the following.
[def\xxx\{arg0 is ^0, arg1 is ^1 and arg2 is ^2}]
[xxx\yyy\zzz]
This would generate:
arg0 is xxx, arg1 is yyy and arg2 is zzz
For an extremely obscure example see: BCPL/cintcode/perm.bgpm.

bin-hex FROM/A,TO/K CIN:y, POS:y, NAT:y

This outputs the bytes of the FROM in hex. For instance, if the file xxx was
ABCDEFGH
12345678
Then the command bin-hex xxx would generate
41 42 43 44 45 46 47 48 0A 31 32 33 34 35 36 37
38 0A
Unless TO is specified output is sent to the terminal..
4.3. COMMANDS 135

bin-x8 FROM/A,TO/K CIN:y, POS:y, NAT:y

This outputs the words of the FROM in hex. For instance, if the file xxx was

ABCDEFGH
12345678

Then the command bin-x8 xxx would generate

44434241 48474645 3332310A 37363534 00000A38

The default TO file name is JUNK.

bits2wav FROM,TO/K,B/N,S/N,A/N,D/N,T/S CIN:y, POS:y, NAT:y

This commands converts a bit stream file generated by the raster command to a
.wav sound file. FROM and TO specify the source and destination files. B specified the bit
rate in bits per second. S specifies the .wav sample rate which should be 11025, 22050
or 44100. A and D control the signal filtering and T turns on tracing as a debugging
aid. The default FROM file is RASTER.bits. The default TO file is RASTER.wav and the
default sample rate is 11025 samples per second.

bmake TARGET,FROM/K,TO/K,-m/S,-l/S,-p/S,-r/S,-s/S,-c/S,-d/S
CIN:y, POS:y, NAT:n
This command provides an approximation the make command found in other sys-
tems. It uses a makefile (normally bmakefile) to generate a CLI sequence of commands
to bring a specified target up to date. The makefile is expanded using the BGPM macro-
generator and parsed to form a set of pattern rules and explicit rules. Each rule has a
target, an optional set of items on which the target depends and a possibly empty CLI
command sequence to execute if the target need to be brought up to date.
Pattern rules generate explicit rules when needed. They contain parameters of the
form <tag>. Within a pattern all tags must be the same and must be declared in the
target of the rule.
The optional first argument (TARGET) is normally a file name and specifies the target
to make. If no target is specified, the target of the first rule is used. The optional FROM
argument specified the makefile name. The default makefile is bmakefile. The optional
TO argument specifies where the output is to be sent.
The -m argument causes bmake to output the makefile file after macrogeneration.
The -l argument outputs the makefile as a sequence of lexical tokens. The -p argument
outputs the set of rule patterns. The arguments -r and -s output the explicit rules
before and after the application of the rule patterns, respectively. The -c argument
outputs the sequence of commands required to bring the target up to date. The -d
argument generates a debugging trace of the execution of bmake.
The BGPM macrogenerator is described elsewhere, but the version use in bmake
uses the following special characters:
136 CHAPTER 4. THE COMMAND LANGUAGE

% Comment - skip all characters until a non white space char-

acter on a later input line.
[ Start of a new macro call.
! Argument separator in macro calls.
# Argument item prefix.
] End of macro argument list.
{ Open quote character.
} Close quote character.

A typical macro definition and call is as follows:

[def!xxx!{This output results from the call {[xxx!}#1{]}}]

[xxx!yyy]

This would generate:

This output results from the call [xxx!yyy]

The syntax of bmake rules is as follows:

target-item <= item ... item << command-sequence >>
Every rule must have a target item and a body consisting of a possibly empty com-
mand sequence enclosed in << and >> brackets. The command-sequence is an arbitrary
sequence of characters not containing >>. The item list may be empty and, if so, the
symbol <= may be omitted. White space including newlines are allowed anywhere
between items.
Pattern rules contain parameter of the form <tag> as in:

cin/<f> <= com/<f>.b g/hdr.h << c bc <f> >>

Such rules are only used when there is no explicit rule for a given target. When a rule
pattern is applied all occurrences of its parameter are replaced by the text that allowed
the target item to match the required target. So if cin/echo must be brought up to
date and has no explicit rule, the above pattern will automatically add the following
explicit rule to the set:

cin/echo <= com/echo.b g/hdr.h << c bc echo >>

A target is out of date if it does not exist or if any of the items it depends on are out
of date or have a modify dates later than that of the target. A target is brought up to
date by, first, bringing the items it depends on up to date and then executing the CLI
command sequence given by the body.
Items may consist of any sequence of characters not including %, [, !, ], {, }, =, or
white space, and < and > may only appear in parameters.
In normal use, bmake generates a command-command file to bring the target up
to date and then returns to the CLI to cause this file to be executed. The -c option
allows the command-command file to be inspected without execution.
4.3. COMMANDS 137

bounce CIN:n, POS:y, NAT:n

This command is part of the bounce demonstration that is only available under
Cintpos. It is normally invoked by the command: run bounce which creates a new
CLI task and then enters the bounce program whose main loop is:

qpkt(taskwait()) REPEAT

which repeatedly suspends the task until a packet is received then immediately returns
it to the sender. Packets are normally sent to the bounce task using the send command,
described below.

break TASK/A,A/S,B/S,C/S,D/S,E/S,ALL/S CIN:n, POS:y, NAT:n

This Cintpos command is used to break the normal execution of a specified task.
The first argument gives the task number and the remaining arguments specify which
flags to set. If no flags are specified flag B is set. If ALL is specified all the flags from
A to E are set.

c command-file arguments CIN:y, POS:y, NAT:y

The c command allows a file of commands to be executed as though they had just
been typed in. The argument command-file gives the name of the file containing the
command sequence. It first looks in the current directory then the directories specified
by the scripts environment variable whose name is in the rtn scriptsvar field of
the rootnode, and finally, if that fails, it looks in the directory specified by the root
environment variable whose name is in the rtn rootsvar field of the rootnode.
Unless explicitly changed, the characters ’=’, ’<’, ’>’, ’$’ and ’.’ have special mean-
ings within a command command. A dot ’.’ at the start of a line starts a directive
which can specify the command command’s argument format, or replace one of the
special character with an alternative. There are six possible directives as follows:

.KEY or .K str Argument format string.

.DEFAULT or .DEF key value Give key a default value, optionally, = is
allowed between the key and value.
.BRA ch Use ch instead of <
.KET ch Use ch instead of >
.DOLLAR ch Use ch instead of $
.DOT ch Use ch instead of .

All directives must occur at the start of the command file. The .KEY directive
specifies a format string of the form used by rdargs (see page 69) that describes
what arguments can follow the command file name. The .DEFAULT directive specifies
the default value that a specified key should have if the corresponding argument was
omitted. The remaining directives allow the special characters to be changed.
The command sequence occurs after all the directives and may contain items of the
form <key$value> or <key> where key is one of the keys in the format string and value
is a default value. Such items are textually replaced by its corresponding argument or
a default value. If $value is present, this overrides (for this item only) any default that
might have been given by a .DEFAULT directive.
138 CHAPTER 4. THE COMMAND LANGUAGE

casech FROM/A,TO/A,DICT/K,U/S,L/S,A/S CIN:y, POS:y, NAT:y

This command systematically converts all reserved words of a BCPL program to
upper case and changing all identifiers to upper case (U), lower case (L, or in the form
given by a specified dictionary (DICT). The A switch causes all letters including those
in strings to be converted to upper case.

changepri TASK/N/A,PRI=PRIORITY/N CIN:n, POS:y, NAT:n

This Cintpos command changes the priority of the specified task to a specified
value. If two arguments are given the first identifies the task and the second the new
priority. If only one argument is given it is treated as the new priority of the current
task. A Cintpos priority can be any positive integer but there is the restiction that no
two tasks can have the same priority.

checksum FROM/A,TO/K CIN:y, POS:y, NAT:y

This command calculates a check sum for the file specified by the FROM argument,
sending the result to the file specified by the TO argument.

cmpltest CIN:y, POS:y, NAT:y

This is a test program that checks for errors in the BCPL compiler and Cintcode
interpreter.

cobench CIN:y, POS:y, NAT:y

This is a benchmark program to test the efficiency of coroutines.

cobounce CIN:y, POS:y, NAT:y

This is a simple coroutine benchmark that bounces a message between two corou-
tines. On my 1.66Ghz Pentium laptop it outputs the following:

0.000> cobounce
Calling the bounce coroutine 10000000 times
About 7812500 coroutine changes per second
done
2.560>

This shows that transferring control between coroutines is about 12 time faster than
transferring control between Cintpos tasks as demonstrated by the send command
described below.

compare FILE1/A,FILE2/A,TO/K,OPT/K CIN:y, POS:y, NAT:y

This command compares two files outputting a description of how they differ to the
TO file if specified, or to standard output if not. The OPT string consists of items of the
form Wn, Mn and Rn, separated by spaces or commas. Each n is a number greater than
zero. Wn means truncate all input lines to no more than n characters. Mn search for up
to n mismatching lines. Rn means that n lines must match before synchronisation is
restored after a mismatch.
4.3. COMMANDS 139

cosim -n/n,-s/n,-p/n,-r/n,-t/s CIN:y, POS:y, NAT:y

This is a demonstration program showing how to write a discrete event simulator
using coroutines, and it is also be used as a benchmark. Its arguments can set the
variables n, s, p and r that configure the test, and the -t switch turns on run time
tracing to check that the simulator is behaving correctly. For a full description and
listing of this program see Section 3.7.3.

dat TO/K,MSECS/S CIN:y, POS:y, NAT:y

This commands output the current date and time to the TO file, if specified, other-
wise it is sent to the standard output stream. The MSECS options causes the time to
have higher precision. Typical output is as follows:

Monday 23-Apr-2010 14:04:12

Monday 23-Apr-2010 14:04:14.392

date TO/K CIN:y, POS:y, NAT:y

This commands output the current date to the TO file, if specified, otherwise it is
sent to the standard output stream. Typical output is as follows:

Monday 23-Apr-2010

delete ,,,,,,,,,-f/S CIN:y, POS:y, NAT:y

This command will delete up to ten given files. If the -f argument is given, no
error message is generated if any file to be deleted does not exist.

detab FROM/A,TO/K,SEP/K CIN:y, POS:y, NAT:y

This command copies the file give by the FROM argument to the file given by the TO
argument replacing all tab characters by spaces. The tabs are separated by a distance
specified by the SEP argument. The default is 8.

dumpmem ON/S,OFF/S CIN:y, POS:y, NAT:y

The ON switch causes Cintsys or Cintpos to set the dumpflag in the rootnode to
TRUE. OFF causes the dumpflag to be set to FALSE. If the dumpflag is TRUE when a
fault occurs or when a return from the interpreter occurs, the entire Cintcode memory
is output in a compacted form. Such memory dumps are sent to the file DUMP.mem
for later inspection by commands such as sysdebug, dumpsys, posdebug and dumppos.
Calling dumpmem without arguments causes an immediate memory.

dumppos FROM,TO/K CIN:y, POS:y, NAT:y

This outputs a readable form of a Cintpos memory dump specified by the FROM
argument. If FROM is not given it uses the file DUMP.mem. The output is sent to the TO
file if given, otherwise it goes to standard output.

dumpsys FROM,TO/K CIN:y, POS:y, NAT:y

This outputs a readable form of a Cintsys memory dump specified by the FROM
argument. If FROM is not given it uses the file DUMP.mem. The output is sent to the TO
file if given, otherwise it goes to standard output.
140 CHAPTER 4. THE COMMAND LANGUAGE

easter YEAR/N,CYCLE/S CIN:y, POS:y, NAT:y

This command outputs the date of Easter Sunday for 10 years from the year given
by the YEAR argument. If the YEAR argument is nott given the output starts from the
current year.
After many years the sequence of Easter dates repeats. Giving the CYCLE option
causes the program to discover the length of this cycle.

echo TEXT,TO/K,APPEND/S,N/S CIN:y, POS:y, NAT:y

This command outputs its first argument TEXT, if given. The text will be followed
by a newline unless the switch N is set. If the TO argument is given, text is sent to the
specified file othewise it goes to the standard output stream. The APPEND switch causes
the output to be appended to the TO stream, after creating an empty file if necessary.

edit FROM/A,TO,WITH/K,VER/K,OPT/K CIN:y, POS:y, NAT:y

This command is meant to provide a simple line editor. It used to run on the Tripos
Portable Operating System but has not yet been modified to run on this version of the
system.

endcli CIN:n, POS:y, NAT:n

This Cintpos command causes a CLI task to commit suicide.

enlarge /A,TO/K CIN:y, POS:y, NAT:y

This command output a large version of its first argument either to file or to stan-
dard output. For instance: enlarge Hello will generate the following:

## ## ######## ## ## ######
## ## ######## ## ## ########
## ## ## ## ## ## ##
######## ###### ## ## ## ##
## ## ## ## ## ## ##
## ## ## ## ## ## ##
## ## ######## ######## ######## ########
## ## ######## ######## ######## ######

fact CIN:y, POS:y, NAT:y

This is a simple example program used in the console session demonstration pre-
sented on page 8.

fail RC/N,REASON/N CIN:y, POS:y, NAT:y

This command returns to the CLI with the specified return code and second re-
sult. The default return code is 10 and the default second result is zero. Unlike the
quit command described below, it does not cause the current command-command to
terminate.

failat FAILLEVEL/N CIN:y, POS:y, NAT:y

This sets the CLI fail level to its argument if given, otherwise it outputs the current
setting. The CLI only issues a warning message if a command yields a return code
greater than or equal to the fail level value.
4.3. COMMANDS 141

fast CIN:y, POS:y, NAT:n

This is a program selects the fast interpreter.

getlogname NAME CIN:y, POS:y, NAT:y

This command outputs the value of a given logical variable name. If none is given
it lists the names and values of all logical variables. The list of logical name value pairs
is held in the root node element rtn envlist.

harness CIN:n, POS:y, NAT:n

This is Cintpos command whose purpose test a system by generating a sequences
of timed events specified by a script.

help ,,,,,,,,,,,,,,,#HELPDIR/K,#TO/K,#TRACE/S CIN:y, POS:y, NAT:y

This command is meant to provide a help facility but has not yet been transferred
to Cintsys or Cintpos.

hex-bin FROM/A,TO/K CIN:y, POS:y, NAT:y

This is the inverse of the bin-hex command. It reads pairs hex digit outputting
the corresponding 8-bit bytes.

hexdump FROM/A,N/N,P/N,RL/K/N,RLB/K/N,TO/K,
X1/S,X2/S,X4/S,LIT/S,BIG/S CIN:y, POS:y, NAT:y
This program dumps a file specified by FROM in a combination of hex and character
forms. If either RL or RLB is given the file is treated as a sequence of records. RL gives
the record length in BCPL words and RLB gives it in bytes. The P and N arguments
give the number of the first record to dump and N specifies how many to dump. If
neither RL nor RLB is given P gives the number of the first byte to dump and N gives the
number of bytes to dump. X1 causes the file to be dumped as a sequence of individual
bytes. X2 causes the file to be dumped as a sequence of 16-bit words, and X4 causes
the file to be dumped as a sequence of 32-bit words. LIT or BIG specify whether to
use little-ender or big-ender ordering when dumping words. It neither are specified the
enderness of the current computer is used. If the file bc is as follows:

#!/home/mr/distribution/BCPL/cintcode/cintsys -s
.k file/a,arg
echo "bcpl com/<file>.b to cin/<file> hdrs BCPLHDRS <arg>"
bcpl com/<file>.b to cin/<file> hdrs BCPLHDRS <arg>

then the command: hexdump bc 64 would generate the following:

Dump of bc from 0 to 63 little-ender mode
0/ 0: 682F2123 2F656D6F 642F726D 72747369 #!/h ome/ mr/d istr
16/ 4: 74756269 2F6E6F69 4C504342 6E69632F ibut ion/ BCPL /cin
32/ 8: 646F6374 69632F65 7973746E 732D2073 tcod e/ci ntsy s -s
48/ 12: 206B2E0A 656C6966 612C612F 650A6772 ..k file /a,a rg.e

hold TASK/N/A CIN:n, POS:y, NAT:n

This is only available under Cintpos. It causes the specified task to be put into
HOLD state to stop it being available to run. Its inverse is unhold described below.
142 CHAPTER 4. THE COMMAND LANGUAGE

idvec ADDRESS/A CIN:n, POS:y, NAT:n

This Cintpos command attempts to identify the vector at a given address. Two
example call are given below:

0.000 1> idvec 23522

Stack of task 4
0.000 1> idvec 15994
Code section of task 5: MBXHAND
0.000 1>

if ,NOT/S,WARN/S,ERROR/S,FAIL/S,EQ/K,VAREQ/K,EXISTS/K: CIN:y, POS:y, NAT:y

This command normally ends with a semicolon and the remainder of the line is
conditionally executed by the CLI depending on whether the if condition is satisfied.
The return code and second result of the previous CLI command are held in the globals
cli returncode and cli result2. If one of WARN, ERROR or FAIL was given, the
if command tests whether the previous command’s return code greater or equal to
warn(=5), error(=10) or fail(=20). If the EQ argument was given, it tests whether the
return code is the same as the first argument. If VAREQ is given, it specifies is a logical
variable name and the value of this variable is compared with the first argument. The
EXISTS argument is a file name whose existence is tested. The NOT switch complements
the condition.

input TO/A,TERM/K CIN:y, POS:y, NAT:y

This command will copy text from the current input sending it the the file specified
by the AS argument. The input is terminated by a line starting with /* or the value of
the TERM argument if given.

interpreter FAST/S,SLOW/S| CIN:y, POS:y, NAT:y

This command allows the user to select the fast (cintasm) or the slow (cinterp)
version of the interpreter. If no arguments are given the fast one is selected. It is
implemented using sys(Sys quit,-1) or sys(Sys quit,-2) as described on page 82.

join ,,,,,,,,,,,,,,,AS/A/K,CHARS/S CIN:y, POS:y, NAT:y

This command will concatenat several files sending the result to the file specified
by the AS argument. If the CHARS switch is given the files are treated as text files,
otherwise they are copied in binary.

lab LABEL/A CIN:y, POS:y, NAT:y

This command has no effect. Its sole purpos is be the destination of skip com-
mands.

library FROM,OVERRIDE/S,CANCEL/K,LIST/S,-g/S,TO/K CIN:n, POS:y, NAT:n

This rather dangerous command allows the user to add or delete sections of resident
system code. If the FROM argument is given the specified file is loaded and its sections
added to the end of the chain of BLIB sections pointed to by the root node field
rtn blib. If OVERRIDE is given the newly loaded sections are allowed to replace previous
ones with the same section names, otherwise all newly loaded sections must have names
4.3. COMMANDS 143

distinct from those already in the BLIB chain. The CANCEL argument specifies the name
of a section to remove from the BLIB chain. The LIST switch argument causes a list
of the section names in the BLIB chain to be output. The argument -g causes a list of
all the global functions defined in the BLIB chain to be output including the names of
the sections they are in. The TO argument specifies the name of a file where the output
is to be sent. It is often useful to sort this file using sortlines. Normally the library
command is only used during the initialisation of special purpose versions of Cintsys
or Cintpos, or when one wishes to see which functions are defined in BLIB.

logout CIN:y, POS:y, NAT:y

This command causes an exit from the BCPL Cintcode System, typical returning
to an operating system shell.

makeinit ,,,,,,,,,,,TO/A/K,STKSIZE/K,GLOBSIZE/K CIN:y, POS:y, NAT:y

This command is used by the native code version of BCPL to generate a C program
used to initialise a native code compilation of BCPL program. It takes a list of BCPL
source files and writes to the TO file a C program that will perform the necessary runtime
initialisation of them. This program also sets the runtime stack size and global vector
size to 50000 and 1000, respectively, unless overridden by the STKSIZE and GLOBSIZE
arguments. The resulting C program should compiled and linked with the native code
compilations of the BCPL files and various library modules. For more information look
in the directory BCPL/natbcpl of the standard BCPL distribution. An example of the
use of makeinit is given on page 218.

map BLOCKS/S,NAMES/S,CODE/S,MAPSTORE/S,TO/K,PIC/S CIN:y, POS:y, NAT:y

This command outputs the Cintcode memory in a form that depends on the ar-
guments given. The output goes to the screen unless a filename is given using the TO
keyword. BLOCKS outputs a list of all blocks whether allocated or free in the block
chain used by getvec. CODE outputs a list of all code sections currently in memory.
MAPSTORE output the code sections and function entry points currently in memory, and
PIC outputs a picture of what memory is currently allocated.

map BLOCKS/S,NAMES/S,CODE/S,MAPSTORE/S,TO/K,PIC/S CIN:y, POS:y, NAT:y

This command outputs the Cintcode memory in a form that depends on the ar-
guments given. The output goes to the screen unless a filename is given using the TO
keyword. BLOCKS outputs a list of all blocks whether allocated or free in the block
chain used by getvec. CODE outputs a list of all code sections currently in memory.
MAPSTORE output the code sections and function entry points currently in memory, and
PIC outputs a picture of what memory is currently allocated.

mapcode FILE/A,TO/K,-r/N,-t/S,-f/S CIN:y, POS:y, NAT:y

This command inspects both 16-bit Cintcode and 6502 machine code used on the
BBC Microcomputer displaying such data in a readable form. FROM specifies the data
file, TO specified the output file. The -r specifies the address of the first byte of
machine code to display. The -t argument turns on debugging tracing and -f causes
extra information to be displayed about each byte of data being inspected.
144 CHAPTER 4. THE COMMAND LANGUAGE

mbxcli MBXNAME CIN:n, POS:y, NAT:n

This command creates a new CLI task taking input from the specified mailbox,
typically MBX:name. If no argument is specified the default mailbox MBX:commands is
used. Any task can write command lines to a mailbox in a first come first served
manner and any CLI created by mbxcli can read and perform them, similarly in a first
come first served manner. If a mailbox CLI performs the endcli command it commits
suicide.

mbxrx -n/N,-d/N,-b/K CIN:n, POS:y, NAT:n

This command is designed to test the mailbox system under Cintpos. It will read
a number of mailbox lines specified by the -n argument. Each line read is written to
the standard output stream. It then delays for a number of milli-seconds specified by
the -d argument before reading the next mailbox line. The mailbox is specified by the
-b argument with the default being MBX:junk.

mbxtx -n/N,-d/N,-b/K CIN:n, POS:y, NAT:n

This command is designed to test the mailbox system under Cintpos. It will write
a number of lines specified by the -n argument to a mailbox. Each line sent is written
to the standard output stream. It then delays for a number of milli-seconds specified
by the -d argument before sending the next mailbox line. The mailbox is specified by
the -b argument with the default being MBX:junk.

mcpl CIN:y, POS:y, NAT:y

This command compiles an MCPL program into Mintcode. See the MCPL distri-
bution for more details.

mcpl2mial CIN:y, POS:y, NAT:y

This command compiles an MCPL program into MIAL.

mial-386.b CIN:y, POS:y, NAT:y

This translates the MIAL form of an MCPL program into Pentium assembly lan-
guage.

mial-masm CIN:y, POS:y, NAT:y

This translates the MIAL form of an MCPL program into a mnemonic form.

mkdata NAME,SIZE/N CIN:y, POS:y, NAT:y

This creates a file with given name and size. The default name is junk and the
default size is 4096*3+10 bytes. Byte i of the created file is i MOD 256 except every 64th
character is a newline and the first 6 characters of every line hold a decimal number
giving the position of the first character of that line.

mkjunk NAME,SIZE/N CIN:y, POS:y, NAT:y

This creates a file as described in the mkdata command and then tests random
access to this file by overwriting some of its bytes.
4.3. COMMANDS 145

newcli CIN:n, POS:y, NAT:n

This Cintpos command creates a new CLI task.

nlconv FILE,TOUNIX/S,TODOS/S,Q/S CIN:y, POS:y, NAT:y

This command replaces the specified file with one in which line endings have been re-
placed by those appropriate for the desination system which is specified by the switches
TOUNIX (the default) or Windows systems (TODOS). The Q argument quietens the com-
mand.

origbcpl CIN:y, POS:y, NAT:y

This is an old version of the BCPL compiler dated 13 August 2001 sometimes used
for benchmarking purposes.

playback FROM/A,WAIT/S,NOTIME/S CIN:y, POS:y, NAT:y

This plays back a console session recording made using the record command.

playfast FROM,TO/K CIN:y, POS:y, NAT:y

This copies a specified recording file (created by the record command) to the
specified output enclosing timing bytes in square brackets.

playtime FROM/A CIN:y, POS:y, NAT:y

This outputs how long a specified recording (created by the record command) will
take to playback.

posdebug FROM CIN:y, POS:y, NAT:y

This is an interactive debugger that allows the user to inspect a given Cintpos
memory dump file. The default file name is DUMP.mem. See dumpmem described above.

prbbcocode FROM,TO/K CIN:y, POS:y, NAT:y

This command converts a 16-bit OCODE file used by the BCPL compiler for the
BBC Microcoputer into a more readable form. FROM specifies the Ocode file. The TO
argument specifies the destination file. If it is missing it sends the result to the screen.

prefix PREFIX,UNSET/S CIN:y, POS:y, NAT:y

This command is primarily for systems that do not have the concept of a current
working directory. If the first argument is given, it becomes the current prefix string. If
UNSET is specified, the prefix string is unset, and if no argument is given the current
prefix is output. This command is implemented using sys(Sys setprefix,prefix) and
sys(Sys getprefix) described on page 84. See also Section 3.3.2.

preload ,,,,,,,,, CIN:y, POS:y, NAT:y

This command will preload up to 10 commands into the Cintcode memory. Without
arguments, it outputs the list of all preloaded commands and their sizes. Preloading
improves the efficiency of command execution and is also useful in conjunction with the
stats command, see below. Preloaded commands can be removed using the unpreload
command.
146 CHAPTER 4. THE COMMAND LANGUAGE

prmcode CIN:y, POS:y, NAT:y

This command converts an MCODE (intermediate code for MCPL) file specified
by FROM to a more readable form. If FROM is missing it reads from the file MCODE. If
the TO argument is missing it sends the result to the screen. The file MCODE is a
byproduct of the mcpl command, see mcpl above.

procode FROM,TO/K CIN:y, POS:y, NAT:y

This command converts an OCODE (intermediate code for BCPL) file specified by
FROM to a more readable form. If FROM is missing it reads from the file OCODE. If the TO
argument is missing it send the result to the screen.

prompt PROMPT,P0/S,P1/S,P3/S,P4/S,NO/S CIN:y, POS:y, NAT:y

If the NO switch is given prompts are disabled, otherwise they will be enabled.
Under Cintpos, disabling prompts is useful, for instance, if a CLI task is taking input
from a TCP/IP connection where the source of the commands is another program. The
PROMPT argument is optional, but if present will be the new prompt format string. The
switch parameters P0 to P4 select commonly used prompt formats. The CLI generates
prompts using a call of the following form.

writef(prompt, cpumsecs, taskno, hours, mins, secs, msecs)

where prompt is the prompt format string, cpumsecs is the time in milliseconds used
by the previous command, taskno is the current task number under Cintpos and zero
otherwise. The arguments hours, mins, secs and msecs represent the current time of
day. The default prompt format under Cintpos is: "%+%n> " and under the other
systems is: "%5.3d> ". An example of how it might be used is as follows.

0>
0> prompt "%+%+%z2:%z2:%z2 %-%-%-%-%-%5.3d> "
15:11:52 0.000>
15:11:55 0.000> bench100
bench mark starting, Count=1000000
starting
finished
qpkt count = 2326410 holdcount = 930563
these results are correct
end of run
15:12:14 10.690>

This shows that bench100 finished execution 14 seconds after 3:12pm after running for
10.690 seconds.

quit RC/N,REASON/N CIN:y, POS:y, NAT:y

This causes a CLI command-command to terminate returning a completion code of
zero unless overridden by the RC argument. If REASON is given it is placed in result2.
4.3. COMMANDS 147

This command differs from fail since it terminates the execution of a command-
command while fail allows a command-command to continue run.

rast2ps FROM,SCALE/N,TO/K,ML/N,MH/N,MG/N,FL/N,FH/N,FG/N,
DPI/K/N,INCL/K,A5/S,A4/S,A3/S,A2/S,A1/S,A0/S CIN:y, POS:y, NAT:y
This command converts a raster data file (written using the raster command
described below) into a postscript file suitable for printing.
The FROM parameter specifies the name of the raster data file. RASTER is the default.
SCALE specifies a magnification as a percentage. The default is 80. The TO parameter
specifies the name of the postscript file to be generated. RASTER.ps is the default.
The parameters ML and MH specify the low and high limits of the address space to be
processed. MG specifies the separation of the grid line on the memory axis. The default
values of MH and FH are given by the FROM file. The default values of ML and FL are
both zero. Unless MG and FG are given, suitable values are chosen automatically. The
units are in bytes. The parameters FL and FH specify the low and high limits of the
instruction count axis to be displayed. FG specifies the separation of the grid line on the
memory axis. DPI specified the approximate number of dots per inch used by the output
device. The default is 300. An specified the output page size. The default is A4. The
INCL parameter specifies the name of a file to be copied into the postscript file. This
file allows annotations to be made in the picture. The file cintcode/origbcplps.h
was used to annotate the memory time graph shown in Figure 4.2. This file contains
lines such as:

F2 setfont
(SYN) 1.1 35 2 PDL
(TRN) 8.1 30 1.7 PUL
(CG) 15.3 36 2.1 PUR
(GET Stream) 0.45 270 1.7 PUL
...
(OCODE Buffer) 13.9 245 2 PDR
% 8.5 150 MVT (HELLO WORLD) SC
F3 setfont
(Self Compilation of the Cintcode BCPL Compiler) TITLE

The postscript macros PDL, PUL, PUR and PDR draw arrows with specified labels, byte
address, instruction count and arrow lengths. The arrow directions are respectively:
down left, up left, up right and down right. The macro MVT moves to the specified
position in the graph and SC draws a string centered at that position. The TITLE
macro draws the graph title and F2 and F3 are fonts suitable for the labels and title.
The resulting postscript file can, of course, be further editied by hand.

rast2wav FROM,TO/K,n/N,s=secs/N,r/N,d/N,stereo/N,t/N CIN:y, POS:y, NAT:y

This command converts a raster data file (written using the raster command
described below) into .wav sound file based on the pattern of memory accesses during
the CLI command following the call of raster.
The FROM parameter specifies the name of the raster data file. RASTER is the default.
The TO parameter specifies the name of the .wav file to be generated. RASTER.wav is
148 CHAPTER 4. THE COMMAND LANGUAGE

the default. By default, the program only generated notes that are equal temperament
semitone (12 per octave), but the n argument allows the user to specify a different
number of notes per octave, susch as 24 or 41. The duration of the generated sound file
can be specified using the s or secs argument. The default .wav sample rate is 44100
per second, but 22050 or 11025 can be specified using the r argument. By default notes
are numbered upwards from 0 to 60 with 12 notes per octave the lowest note id C two
octaves below middle C. The d option is a debugging aid that causes all notes other
that a specified one to be silent. This allows the algorithm to choose when to sound a
note to be tested including how it volumes envelope changes. The t argument is not
yet implemented but willin due course generate trace output during the execution of
rast2wav. This command was written the sound generated by EDSAC 2’s loudspreaker
in the 1960s was a remarkably useful debugging aid.
As a demonstration, origbcpl.wav or origbcpl.mp3 is the sound of the early
version of the BCPL compiler compiling itself, and the following command sequence
from a bash prompt will generate an approximation to Bach’s Invention no 10, bwv
784. These demonstrations are not yet ready.
rastsys
c bc bwv784
raster
bwv784
rast2wav
ctrl-c
audacity RASTER.wav

raster COUNT/N,SCALE/N,TO/K,BITS/S,HELP/S CIN:y, POS:y, NAT:y

This command controls the generation of raster data but only works when the
BCPL Cintcode system is running under the rastering interpreter rasterp. The im-
plementation uses sys(Sys setraster,...) calls that are described on page 85. If
raster is called without the BITS options it activates the rastering mechanism for the
duration of the next CLI command. Without the BITS option the default TO file is
RASTER. The format of this file is outlined on page 85.
The COUNT argument specifies the number of Cintcode instructions to obey per
raster line. The default is 1000. The SCALE argument gives the number of byte addresses
per unit on the memory axis. The default being 8.
The raster data file can be processed and converted to Postscript using the rast2ps
command described above. Typical use of the raster command is following script,
starting from a linux bash prompt:
rastsys
raster
origbcpl com/origbcpl.b to junk
rast2ps incl origbcplps.h
ctrl-c
ps2pdf RASTER.ps
okular RASTER.pdf

This will create a .pdf file for an early version of the BCPL compiler compiling itself,
4.3. COMMANDS 149

similar to that shown in Figure 4.2. For a more detailed view of the parse tree while
SYN is being compiled, try:

rastsys
raster
origbcpl com/origbcpl.b to junk
rast2ps incl origbcplps.h ml 350000 mh 500000 fh 6000000
ctrl-c
ps2pdf RASTER.ps
okular RASTER.pdf

500K
OCODE Buffer

GET Stream
450K

400K
TRN Parse Tree
SYN Parse Tree CG Parse Tree Compiled Code Buffer
350K
Declaration Vector

300K

250K
Input Stream Code Output Stream

200K

150K
Stack CG
SYN TRN
100K

50K

0K
0.00M 2.00M 4.00M 6.00M 8.00M 10.00M 12.00M 14.00M 16.00M 18.00M 20.00M 22.00M

Self Compilation of the Cintcode BCPL Compiler

Figure 4.2: Self compilation memory-time graph

If raster is called with the BITS option, the next CLI command will generate a bit
stream file corresponding to the fifth bit of every Cintcode byte address accessed. The
default TO file name is RASTER.bits. This file contains one byte for every 8 memory
references so can become very large. It can be converted to a .wav sound file using the
bits2wav command.

record TO,OFF/S CIN:n, POS:y, NAT:n

This Cintpos command starts sending a recording data including timing information
of the current console sessions to the specified file. The recording is stopped by the
command record off. See the commands playback, playfast, and playtime.
150 CHAPTER 4. THE COMMAND LANGUAGE

rename FROM/A,TO=AS/A/K CIN:y, POS:y, NAT:y

This will rename the file given by FROM to that specified by the AS argument.

repeat CIN:y, POS:y, NAT:y

This attempt to reposition CLI input to the start of the current command line
thereby causing it to be executed again. For example:
wait 3; echo hello; repeat
will output hello to the screen every 3 seconds until interrupted by the D flag (set by
@d).

run command-line CIN:n, POS:y, NAT:n

This Cintpos command creates a new CLI task giving it command-line to execute.
On complete this new CLI task commits suicide.

send TASK/N,COUNT/N CIN:n, POS:y, NAT:n

This is part of the Cintpos bounce demonstration. It repeatedly sends a packet
to the specified task the specified number of times. The default task number is 7 and
the default count is 1000000. It can be used to measure the efficiency of inter-task
communication. On my 1.66Ghz Pentium laptop, send runs for 3.19secs corresponding
to about 630000 task changes per second.

setflags TASK,A/S,B/S,C/S,D/S,E/S,QUIET/S CIN:n, POS:y, NAT:n

This Cintpos command sets the specified flags in the task control block of the given
task. Unless QUIET is given it outputs the previous setting of the flags.

setlogname NAME,VALUE CIN:y, POS:y, NAT:y

This command sets or possible displays Cintsys or Cintpos logical variables. These
must not be confused with shell environments variables described in Section 3.6.
Cintsys and Cintpos logical variables are held in a linked list held in the rootnode
element rtn envlist. If both NAME and VALUE are given, the given logical variable
name is given the specified value, but if no value is given the specified variable is unset.
If setlogname is called without arguments, the names and values of all logical variables
are output. A running program can lookup and set logical variables using the functions
getlogname and setlogname.

setroot ROOT,PATH,HDRS,SCRIPTS CIN:y, POS:y, NAT:y

If no arguments are given it just outputs the current settings of the four environ-
ment variable names. Otherwise, the specified variables are given new names.

shellcom COMMAND/A CIN:y, POS:y, NAT:y

This command causes its argument to be processed by the command language
interpreter shell of the underlying operating system (typically Linux or Windows). It
does not return until the shell has completed processing the command.

sial-arm FROM,TO/K CIN:y, POS:y, NAT:y

This command converts the Sial intermediate code generated by bcpl2sial to the
equivalent assembly language for machines using the ARM processor.
4.3. COMMANDS 151

sial-386 FROM,TO/K CIN:y, POS:y, NAT:y

This command converts the Sial intermediate code generated by bcpl2sial to the
equivalent assembly language for i386 machines such as Pentiums.

sial-alpha CIN:y, POS:y, NAT:y

This command converts the Sial intermediate code generated by bcpl2sial to the
equivalent assembly language for DEC Alpha machines.

sial-sasm CIN:y, POS:y, NAT:y

This command converts the Sial intermediate code generated by bcpl2sial into a
human readable form.

sial-vax CIN:y, POS:y, NAT:y

This command converts the Sial intermediate code generated by bcpl2sial to the
equivalent assembly language for VAX machines.

skip LABEL CIN:y, POS:y, NAT:y

The command skip label skips through the command stream until a line starting
with lab label is encountered. It then skips until the end of that line before resuming
normal command execution from there. The skip command is only allowed within
command-commands.

slow CIN:y, POS:y, NAT:n

This is a program selects the slow interpreter.

sortlines FROM/A,TO/K CIN:y, POS:y, NAT:y

This command sorts the lines specified by the FROM file sending the result to
the TO file, removing duplicate lines. Output is sent to the screen if the TO param-
eter is not given. This can be used to sort the data generated by the command:
library -g to junk.

sortxref FROM/A,TO/K,FNS/S CIN:y, POS:y, NAT:y

This command sorts the lines specified by the FROM file sending the result to the TO
file, removing duplicate lines. Output is sent to the screen if the TO parameter is not
given. Only lines lines containing G:, M:, F: or S: are included, and if FNS is specified,
only lines also containing FN or RT are included. This is useful when processing cross
reference data generated by the BCPL compiler when the XREF parameter is specified.
A typical cross reference listing can be found in cintcode/xrefdata.

stack SIZE CIN:y, POS:y, NAT:y

The command stack n causes the size of the coroutine stack allocated for subse-
quent commands to be n words long. Without an argument it outputs the current
setting.

stats TO/K,PROFILE/S,ANALYSIS/S CIN:y, POS:y, NAT:y

This command controls the tallying facility which counts the execution of individual
Cintcode instructions. If no arguments are given, stats turns on tallying by clearing
152 CHAPTER 4. THE COMMAND LANGUAGE

the tally vector and causing tallying to be enabled for the next command to be executed.
Subsequent commands are not tallied, making it possible to process the tally vector
while it is in a static state. Typical usage of the stats command is illustrated below:

preload queens Preload the program to study

stats on Enable stats gathering on next command
queens Execute the command to study
interpreter Select the fast interpreter (cintasm)
stats automatically selects the slow one
stats to STATS Send instruction frequencies to file
or
stats profile to PROFILE Send detailed profile info to file
or
stats analysis to ANALYSIS Generate statistical analysis to file

status TASK,FULL/S,TCB/S,SEGS/S,CLI=ALL/S CIN:n, POS:y, NAT:n

This Cintpos command outputs information about all currently existing Cintpos
tasks.

syncdemo CIN:n, POS:y, NAT:n

This is a program to demonstrate various synchronisation mechanisms implemented
using coroutines and multi-event tasks.

sysdebug FROM CIN:y, POS:y, NAT:y

This is an interactive debugger that allows the user to inspect a given Cintsys
memory dump file. The default file name is DUMP.mem. See dumpmem described above.

sysinfo CIN:y, POS:y, NAT:y

This outputs some information about the current BCPL system and the host ma-
chine on which it is running. Typical output is as follows:

This version of BCPL is running on a little ender machine

The BCPL word length is 32 bits

The host address size = 64 bits

system CIN:y, POS:y, NAT:y

This command outputs a message indicating whether the current system is Cintsys,
Cintpos or Unknown. It determines which by inspecting the rootnode field rtn system.

taskid FORMAT CIN:n, POS:y, NAT:n

This command calls writef with the given format and the current task number as
the second argument. The default format is "Taskid=%n*n".

tcpaddr HOST,PORT CIN:n, POS:y, NAT:n

This attempts to output the IP address and port number given the names of the
host and port.
4.3. COMMANDS 153

tcpbench -n/K,-k/K,-s/K,-h/K,-t/S,master/s,slave/s CIN:n, POS:y, NAT:n

This is a benchmark program to test the efficiency of TCP/IP communication. For
information about what it does and how to use it, see the comments at the start of the
source code.

tcpcli PORT,NOPROMPT/S CIN:n, POS:y, NAT:n

This command creates a new CLI task communicating through the given port. The
default port number is 8000. If NOPROMPT is specified the newly created CLI will not
issue prompts.

tcpdump CIN:n, POS:y, NAT:n

This outputs the list of Cintpos TCP/IP devices that currently exist. The list
includes information about sockets, states and associated hosts and port numbers.

tcprx HOST,PORT CIN:n, POS:y, NAT:n

This is a TCP/IP demonstration program to be used in conjuction with tcptx.
It will output data received from a specified host via a specified port. If no host is
specified wait for a connection from any host. The default port number is 9000.

tcptest -n/K,-k/K,-s/K,-h/K,-t/S CIN:n, POS:y, NAT:n

This is a TCP/IP test program. See its source code for details.

tcptx HOST,PORT,N CIN:n, POS:y, NAT:n

This is a TCP/IP test program to be used in conjunction with tcprx. It attempts
to send the message hello world to a specified host via a specified port. The number
of times the message is sent is given by the N argument.

testtime CIN:y, POS:y, NAT:y

This command tests the real time clock, outputting a line such as:
days=14876 hours=11 mins=59 secs=11 msecs=982

time TO/K,MSECS/S CIN:y, POS:y, NAT:y

This command outputs the current time of day to the TO file, if specified, otherwise
it is sent to the standard output stream. The MSECS options causes the time to have
higher precision. Typical output is as follows:

14:12:36.069

type FROM/A,TO,N/S CIN:y, POS:y, NAT:y

This command will output the file given by the FROM argument, sending it to the
screen unless the TO argument is given. The swirch argument N causes line numbers to
be added.

typehex FROM/A,TO/K CIN:y, POS:y, NAT:y

This will convert the file specified by FROM in hexadecimal and send the result to
the TO file if this argument is given. Its output should be compared with that generated
by the hexdump command.
154 CHAPTER 4. THE COMMAND LANGUAGE

unhold TASK/N/A CIN:n, POS:y, NAT:n

This Cintpos command resets the HOLD status bit of a specified task. That task
is then immediately available to run unless suspended of other reasons.

unpreload ,,,,,,,,,,ALL/S CIN:y, POS:y, NAT:y

This command will remove up to 10 specified preloaded commands from the Cint-
code memory. The ALL switch will cause all preloaded commands to be removed.
Commands can be preloaded into memory using the preload which can also be used
to list all preloaded commands.

vecstats CIN:y, POS:y, NAT:y

This command output information about blocks of Cintcode memory that are cur-
rently allocated. Typical output (from Cintpos) is the following:

3: 12 4: 2 6: 1 15: 2 22: 1 23: 7

27: 4 28: 1 41: 1 80: 1 200: 2 291: 1
306: 2 316: 1 406: 1 462: 1 500: 1 506: 3
571: 1 597: 1 757: 1 982: 1 1000: 10 1006: 6
1025: 2 1901: 1 2422: 1 3303: 1 20000: 1

This indicates, for instance, that there are currently 7 blocks of requested size 23
allocated.

wait N/N,SEC=SECS/S,MIN=MINS/S,UNTIL/K CIN:y, POS:y, NAT:y

This causes the CLI to wait for a specified number of seconds or minutes, or until
a specified time is reached.

why CIN:y, POS:y, NAT:y

This command attempts to give the reason why the previous command failed. For
fun you can type why several times.

x8-bin FROM/A,TO/K CIN:y, POS:y, NAT:y

This converts a file of 32-bit words in hex into a file of the corresponding bytes.
For instance, it will convert the file:

44434241 48474645 4C4B4A49 504F4E4D 54535251 58575655 310A5A59 35343332

39383736 00000A30

to

ABCDEFGHIJKLMNOPQRSTUVWXYZ
1234567890

xcmpltest CIN:y, POS:y, NAT:y

This is a test program that checks for errors in the XBCPL compiler and extended
features in the Cintcode interpreter.
4.4. CLI.B AND CLI INIT.B 155

xcdecode FROM/A,LIST/S,BIN/S CIN:y, POS:y, NAT:y

This command is the inverse of xcencode. With the LIST option it will inspect
the FROM file listing the names of the files it contains. Without the LIST option it will
extract and decode these files. If BIN is set, files are written using binwrch so that
carriage return characters (’*c’) are not ignored. All characters before the first file
separator are ignored.

xcencode FILE,LIST/K,TO/K/A,BIN/S CIN:y, POS:y, NAT:y

This command is designed to encode one or more files in such a way that they
can be passed as the body of an email message without interferring with the email
mechanism. It uses a simple form of run length encoding to reduce the size of the
resulting file. Either FILE or LIST or both must be supplied. If given FILE is the first
filename to be encoded followed by those given in LIST file, if present. If BIN is set,
files are read using binrdch so that carriage return characters (’*c’) are not ignored.
Each encoded file is preceeded by a separator of the form:

#####filename#

followed by the encoded file in which all characters with ASCII codes in the range 33
to 126 except for ’#’, ’=’ and ’.’ are copied, spaces are replaced by dots (’.’) and
all other characters (including ’#’ ’=’ and ’.’) are encoded by #hh where hh is the
ASCII code in hex. The encoded files are broken into lines of about 50 characters. The
last file to be encoded is terminated by ######+#.
Such xencode’d files can be decoded by the xdecode command.

4.4 cli.b and cli init.b

The Command Language Interpreter is a simple program implemented in BCPL whose
source code can be found in the files sysb/cli.b and sysb/cli init.b. This section
mainly describes the Cintpos version. The CLI is the first program the interacts with
after starting the system. Under Cintpos it runs as task one (named Root Cli). It
uses variables in the global vector to hold its state during command execution. These
variables have reserved global numbers typically in the range 133 to 149. They are
declared in g/clihdr.b. Since running commands use the same global vector they can
access (and even modify) these variables – a feature that is both dangerous and useful.
Commands such as run and c rely on this feature. The CLI global variables are as
follows.

cli init CIN:y, POS:y, NAT:y

This holds the function used to initialise the CLI, and depends on which context
the CLI is to run in. It is called when the CLI is first entered using the following code.
{ LET f = cli_init(parm.pkt)
IF f DO f(result2) // Must get result2 after calling cli_init
}

As can be seen cli init must either return zero or a function that can be applied
156 CHAPTER 4. THE COMMAND LANGUAGE

to result2. The function is typically deletetask or unloadseg with result2 being

suitably set.

cli returncode, cli result2 CIN:y, POS:y, NAT:y

These hold the return code and the value of result2 of the most recently executed
command.

cli faillevel CIN:y, POS:y, NAT:y

cli data CIN:y, POS:y, NAT:y

This holds CLI data dependant on the context in which the CLI is running.

cli commanddir CIN:y, POS:y, NAT:y

cli prompt CIN:y, POS:y, NAT:y

This variable holds the current prompt which should be a writef format string
since it used in the CLI as follows:

writef(cli_prompt,
cpumsecs, // msecs used by last command
taskid, // The task number, if running under Cintpos
hours, mins, secs, msecs) // The time of day

where hours, mins and secs correspond to the current time of day. On single threaded
BCPL systems taskid is set to 1.

cli currentinput, cli currentoutput, cli standardinput, cli standardoutput

CIN:y, POS:y, NAT:y
The standard input and output streams are those that were setup when the CLI
was started. Sometimes a CLI will change its currently selected streams. For instance,
while executing a command-command the currently selected input will be from a tem-
porary file of commands. On reaching the end of file input will revert to the standard
input.

cli commandfile CIN:y, POS:y, NAT:y

This is either zero or holds the name of temporary command file used in
command-commands.

cli status CIN:y, POS:y, NAT:y

This holds a collection of bits specifying the context in which the CLI is running.
The mnemonics for these bits and their meanings are as follows.
4.4. CLI.B AND CLI INIT.B 157

clibit noprompt Do not output prompts even when not in a command-

command.
clibit eofdel Delete this task when EOF is received under Cintpos.
clibit comcom This CLI is currently in a command-command executing
commands from a temporary file.
clibit maincli This CLI is the task 1 CLI under Cintpos or the main CLI
under other systems.
clibit newcli This CLI was created by the newcli command under Cint-
pos.
clibit runcli This CLI was created by the run command under Cintpos.
clibit mbxcli This CLI was created by the mbxcli command under Cint-
pos.
clibit tcpcli This CLI was created by the tcpcli command under Cint-
pos.
clibit endcli The endcli command has been executed on this CLI under
Cintpos.

cli background CIN:y, POS:y, NAT:y

This is an obsolete variable that mainly controlled the generation of prompts. It is
to be superceded by the noprompt bit in cli status.

cli defaultstack CIN:y, POS:y, NAT:y

This holds the size of the coroutine stack that the CLI creates every time it runs a
command. Its value can be changed by the stack command.

cli commandname CIN:y, POS:y, NAT:y

This holds the name of the current command

cli module CIN:y, POS:y, NAT:y

This is either zero or the module of loaded code corresponding to the currently
executing command. It is used by the CLI to unload commands after they have been
run.
158 CHAPTER 4. THE COMMAND LANGUAGE
Chapter 5

Console Input and Output

When cintsys or cintpos is started a stream is opened to receive input from standard
input which is normally the keyboard and a second stream is opened to allow output
to standard output which is normally the screen. This combination of keyboard and
screen is called the console. The treatment of console streams depends on whether
cintsys or cintpos is being used.

5.1 Cintsys console streams

The stream control block for the keyboard is obtained by calling findinput("**").
The stream is created the first time it is called. Subsequent calls yield exactly the same
stream control block. This stream has a buffer large enough to hold 4096 characters.
Characters are read from the keyboard using sardch which reads and echoes each
character to the screen. Exceptionally, ctrl-c (code 3) causes a SIGINT interrupt,
RUBOUT (code 127) is translated to backspace (code 8), ctrl-j, ctrl-m and the ENTER
(or RETURN) key all yield code 10 (the BCPL newline character) but they all echo
carriage return and linefeed to the screen.
Simple line editing of keyboard input is performed as follows. As characters are
typed they are normally transferred into the buffer, but if a backspace is received, the
latest character, is any, in the buffer is removed and its echoed symbol removed from
the screen. The contents of the buffer is not made available to the user until either a
newline character is received or the buffer becomes full.
A user can receive keyboard characters as soon as they are typed using calls of
sardch. It is also possible to read keyboard characters by polling them using the call
sys(Sys pollsardch). This yields the next character if one is available, otherwise it
returns pollingch=-3, allowing the program to do other work before trying again.
The program BCPL/bcplprogs/test/inputtst.b can be used to demonstate some
of the features of console input.
The stream control block for the screen is obtained by calling findoutput("**").
The stream is created the first time it is called. Subsequent calls yield exactly the same
stream control block. This stream has a buffer large enough to hold 4096 characters.
Calls of wrch places characters in this buffer, and when a newline or newpage character

159
160 CHAPTER 5. CONSOLE INPUT AND OUTPUT

is written, or when the buffer becomes full, or a call of deplete is made, the contents
of the buffer is transmitted to the screen by calls of sawrch.

5.2 Cintpos console streams

Under Cintpos interaction with the console is somewhat more complicated since Cintpos
can have several tasks all wishing to communicate with the keyboard and screen. This
interaction is controlled by a task called the Console Handler (typically task 3). Tasks
wishing to read from the keyboard or write to the screen must send request packets to
this task where they will be properly scheduled.
The call findinput("**") yields a new stream control block connected to the
keyboard. Initially it has no buffer. When the client task tries to read from this
stream, a read request packet is sent to the console handler which will in due course
return with a buffer of one or more characters or an indication that the keyboard stream
is exhausted. Keyboard read requests can be sent simultaneously from several tasks
and, indeed, a single task can send multiple requests. These are queued in the console
handler and processed on a first come first served basis.
The console handler obtains characters from the keyboard by sending ttyin request
packets to the keyboard device (typically device -2). This device returns keyboard
characters to the console handler as they are typed without echoing them to the screen.
It does no translation except that the characters ctrl-j, ctrl-m and the ENTER key all
yield code 10 (the BCPL newline character). Keyboard characters received by the
console handler are normally packed into an input buffer to form input lines. Simple
line editing is performed using the backspace key (code 8 or 127) which causes the most
recent character in the line buffer to be removed. When a newline is received or the
buffer is full or the escape sequence @e is typed, the line buffer is ready to send to the
currently selected task. Initially this is task 1 (the main CLI task) but can be changed
by the user using the escape mechanism described below. While a user is typing an
input line, it will appear on the screen and other screen output requests will be held
until the input line is complete. At any time if there is a completed input line for a
task that has sent a read request packet, it will be returned to the client with the line
buffer and number of characters in its two result fields. Lines that have not yet been
requested are queued as are read requests that are not yet satisfied. Note that a simple
way to temporally stop output to the screen is to type a character such as SPACE, and
then delete it later using backspace.
Cintpos console input has the following escape mechanism. All escape sequence
start with an at sign (@) and their effects are shown in the following table.
5.2. CINTPOS CONSOLE STREAMS 161

Sequence Purpose
@A Set flag 1 in the currently selected task
@B Set flag 2 in the currently selected task
@C Set flag 3 in the currently selected task
@D Set flag 4 in the currently selected task
@E Send the current incomplete line to the currently selected
task
@F Throw away the current incomplete line and all outstanding
completed lines
@H Hold the currently selected task
@L Throw away the current incomplete line
@Sdd Set the currently selected task to task dd and allow output
from any task
@Tdd Set the currently selected task to task dd and only allow
output from task dd
@U Unhold the currently selected task
@Xhh Input the character with hex code hh
@Y Toggle message tagging. When tagging is enabled every line
of output identifies the originating task
@Z Toggle echo mode. When echoing is off subsequent charac-
ters are not echoed to the screen. This is useful for typing
passwords.
@ddd Input the character with octal code ddd
@@ Input @

5.2.1 Devices
The input and output device intentifiers may be inspected and changed by the following
call:
old_in_devid := sendpkt(notinuse, console_task, Action_devices,
?, ?,
new_in_devid,
new_out_devid)
old_out_devid := result2
The device identifiers are only changed if the new identifiers are non zero. This call
is used, for instance, by the record command to change replace the screen output
device with a task that forwards each character to the screen while recording timing
information. For details, see the programs com/record.b and com/recordtask.b

5.2.2 Exclusive Input

The console handler can be set to exclusive input mode by the call:
sendpkt(notinuse, console_task, Action_exclusiveinput,
?, ?,
TRUE)
While in exclusiveinput mode normal input line editing by the console handler is
162 CHAPTER 5. CONSOLE INPUT AND OUTPUT

suspended and client tasks have direct access to the keyboard input device on a first
come first served basis by the call:

ch := sendpkt(notinuse, console_task, Action_exclusiverdch,

?, ?)

Sending an exclusiveinput request with argument FALSE returns the console handler
to its normal line editing mode and causes all outstanding exclusiverdch requests to
return end-of-file characters (-1) to their client tasks.

5.2.3 Direct access to the screen and keyboard

Although it is not recommended, client task can send read (Action ttyin) and write
(Action ttyout) requests to keyboard and screen devices. These will be serviced in a
first come first served basis and since the console handler is making such requests you
can expect strange results.
Finally the functions sardch and sawrch provide direct access to the keyboard and
screen but are mainly only used for system debugging particularly when the console
handler is not running. Note that sawrch is the character output function used by
sawritef whose output may be merged with output from the console handler.
The following test programs can be used to demonstate some of the console handlers
features.

Cintpos/posprogs/test/inputtst.b
Cintpos/posprogs/test/sardchtst.b
Cintpos/posprogs/test/devrdchtst.b
Cintpos/posprogs/test/xintst.b
Chapter 6

Cintpos Devices

Cintpos allows asynchronous communication with peripheral devices using the qpkt
and taskwait functions. If the pkt id field of packet given to qpkt is negative, the
packet is sent to the identified device. It is returned when the device has completed
the requested operation. Most devices have device control blocks (DCBs) that contain
device related data. There is a device table pointed to by rootnode!rtn_devtab
whose upper bound is held in its zeroth element. The nth element of the device table
is zero if the device does not exist, otherwise it points to the DCB of device -n. Most
devices are implemented using threads of the host operating system, but some devices
such as the clock and screen are special and use a polling mechanism implemented
entirely within the interpreter thread. The extra overhead for this is small since the
interpreter only performs the polling operation about once every 10000 or so Cintcode
instructions. This figure is typically adjusted to cause polling to take place about once
per millisecond. When Cintpos has no work to do it should enter the Idle task and
stop executing Cintcode instructions so that other programs can run. For the polling
mechanism to work, such suspensions must be short. This is normally implemented
using the waitirq sys function with a short timeout. Each time waitirq returns, a
counter in the intepreter is set to zero to cause the polling mechanism to be activated.
The resident Cintpos devices are described below.

6.0.1 The Clock Device

This device has identifier -1 and is treated specially by both qpkt and the inter-
preter. The pkt arg1 field of its packet holds the number of milliseconds that the
packet should remain with the clock before being returned. The time stamp of when
it should be returned is calculated by qpkt and placed in the pkt res1 and pkt res2
fields of the packet. It is then inserted into the time ordered clock queue held in
rootnode!rtn_clwkq. Every time the interpreter performs the polling operation it
tests the packets at the start of the clock queue returning though that have expired to
their task.

163
164 CHAPTER 6. CINTPOS DEVICES

6.0.2 The Keyboard Device

This device has identifier -2 and is currently not treated specially, and so it has a DCB,
and a device thread that is continually trying to read character from standard input
which is normally the keyboard. Packets for this device are placed on the end of the
work queue held in the dcb wkq field of the DCB. When a character becomes available
it is placed in the pkt res1 field of the first packet in the queue before returning the
packet to its task.
It is planned to modify keyboard packets to allow them to handle timeouts. This
will be done by setting the pkt arg1 field to a timeout value. If it is is negative no
timeout is used and the packet will remain with the device until a character is received,
otherwise it specifies a timeout in milliseconds. If no character is received within that
time, pollingch (=-2) is returned in the res1 field, but if a character becomes available
within that time it it returned in the normal way.

6.0.3 The Screen Device

This device has identifier -3 and is treated specially. The pkt arg1 field of the packet
holds the next character to send to the screen and when this transfer is complete the
packet is returned to the client task. Normally output to the screen causes no real time
delay.

6.0.4 TCP/IP Devices

TCP/IP devices provide a mechanism to communicate with other machines over the
internet. The pkt type field specified the TCP/IP operation required and the argument
field provide additional information about the request. The possible packet type are
as follows.

Tcp name2ipaddr arg1: name

This looks up the URL name and returns its IP address. Names such as 127.0.0.1
are allowed.

Tcp name2port arg1: name

This looks up the the given port name and returns its its number.

Tcp socket
This attempts to create a port for a two way byte stream using the IPv4 protocol.
If the result is -1 there was an error, otherwise it returns the number of the new socket.

Tcp reuseaddr arg1: sock arg2: flag

If flag=1 this modifies the socket sock to allow reuse of local addresses, otherwise
these are disallowed. A result of zero indicates success.

Tcp sndbufsz arg1: sock arg2: size

This sets the send buffer size of the given socket to size bytes. A zero result indicates
success.
 165

Tcp rcvbufsz arg1: sock arg2: sz

This sets the receive buffer size of the given socket to size bytes. A zero result
indicates success.

Tcp bind arg1: sock arg2: ipaddr arg3: port

This assigns local host and port numbers to the specified socket. A zero result
indicates success.

Tcp connect arg1: sock arg2: ipaddr arg3: port arg4: timeout
This attempts to establish a connection to a remote host via the given socket within
the given timeout. If timeout is greater than zero it specifies a timeout time in milli-
seconds, if it is zero there is no timeout and if it is -1 polling will be used but this
is not yet implemented. The result is zero if a connection was established, otherwise
it is negative and the second result indicates why the connection was not established.
A value greater than zero indicates an error, the value -1 the connection was closed
by the remote host, -2 indicates that the connection was not established within the
timeout period, and -3 indicates that when polling the connection has not yet been
established.

Tcp listen arg1: sock arg2: n

This causes the specified socket to be willing to accept incoming calls from remote
hosts. The queue limit for incoming connections is specified by n. A zero result indicates
success.

Tcp accept arg1: sock arg2: tcp, arg4: timeout

BEWARE: the implementation does not yet quite match the following specification.
This attempts to accept a connection from a remote host via a listening socket within
a specified timeout period. If timeout is greater than zero it is the timeout period in
milli-seconds, if it is zero there is no timeout and if it is negative the packet is returned
immediately having accepted a connection if possible. A positive result indicates success
and is the number of a new socket to to be used by the connection. A negative result
indicates failure with a reason in the second result. A second result of -1 indicates the
connection was closed by the remote host, -2 means a connection was not accepted
within the timeout period, and -3 indicates that there is currently no connection to
accept when polling.

Tcp recv arg1: sock arg2: buf arg3: len arg4: timeout
This attempts to read up to len bytes into the given buffer from the specified socket
within a specified timeout period. If timeout is greater than zero it is the timeout
period in milli-seconds, if it is zero there is no timeout and if it is negative the packet
is returned immediately with as many characters as are currently available. A negative
result indicates failure with a reason given in the second result, otherwise it is the
number of bytes actually read.

Tcp send arg1: sock arg2: buf arg3: len arg4: timeout
This attempts to send len bytes from the given buffer via the specified socket within
166 CHAPTER 6. CINTPOS DEVICES

a specified timeout period. If timeout is greater than zero it is the timeout period in
milli-seconds, if it is zero there is no timeout and if it is negative the packet is returned
immediately having written as many bytes as are currently possible. A negative result
indicates failure with a reason given in the second result, otherwise it is the number of
bytes actually sent.

Tcp close arg1:sock

This closes the specified socket. A zero result indicates success.
Chapter 7

The Debugger

Both Cintsys and Cintpos have interactive debuggers but these are slightly different
and so will be described separately.

7.1 The Cintsys Debugger

When the Cintsys starts up, control first passes to BOOT which initialises the system
and creates a running environment for the command language interpreter (CLI). This
is run by a recursive invocation of the interpreter and so when faults occur control
returns to BOOT which then enters an interactive debugger. This allows the user to
inspect the state of the registers and memory, and perform other debugging operations
on the faulted program. The debugger can also be entered using the abort command,
as follows:

560> abort
!! ABORT 99: User requested
*

The asterisk (*) is the debugger’s prompt character. A brief description of the available
debug commands can be display using the query (?) command.

167
168 CHAPTER 7. THE DEBUGGER

* ?
? Print list of debug commands
Gn Pn Rn Vn Variables
G P R V Pointers
n #b101 #o377 #x7FF ’c Constants
*e /e %e +e -e |e &e ^e Dyadic operators
!e Subscription
< > Shift left/right one place
$b $c $d $e $f $o $s $u $x Set the print style
SGn SPn SRn SVn Store in variable
= Print current value
TRn Trace the next n instructions
Tn Print n consecutive locations
I Print current instruction
N Print next instruction
Q Quit
B 0Bn eBn List, Unset or Set breakpoints
C Continue execution
X Equivalent to G4B9C
Z Equivalent to P1B9C
\ Execute one instruction
, Move down one stack frame
. ; [ ] Move to current/parent/first/next coroutine
*

The debugger has a current value that can be loaded, modified and displayed. For
example:

* 12 Set the current value to 12

* -2 Subtract 2
* *3 Multiply by 3
* = 30 Display the current value
* < Shift left one place
* = 60 Display the current value
* 12 -2 *3 < = 60 Do it all on one line
*

Four areas of memory, namely: the global vector, the current stack frame, the Cint-
code register, and 10 scratch variables are easily accessed using the letters G, P, R, V,
respectively.

* 10sv1 11sv2 Put 10 and 11 in variables 1 and 2

* vt5 Display the first 5 variables
V 0: 0 10 11 0 0
*
* v1*50+v2= 511 A calculation using variables
* g0= 1000 Display global zero (globsize)
* g= 3615 Display the address of global zero
* ! = 1000 Indirect and display
* gt10 Display the first 10 globals
G 0: 1000 start stop sys clihook
G 5: GLOB 5 changec 6081 6081 52
*
7.1. THE CINTSYS DEBUGGER 169

Notice that values that appear to be entry points display the first 7 characters of
the function’s name. Other display styles can be specified by the commands $C, $D, $F,
$B, $O, $S, $U or $X. These respectively display values as characters, decimal number,
in function style (the default), binary, octal, string, unsigned decimal and hexadecimal.
It is possible to display Cintcode instructions using the commands I and N. For
example:

* g4= clihook Get the entry to clihook

* n 3340: K4G 1 Call global 1, incremeting P by 4
* n 3342: RTN Return from the function
*

A breakpoint can be set at the first instruction of clihook and debugged program
re-entered by the following:

* g4= clihook Get the entry to clihook

* b9 Set break point 9
* c Resume execution
20>

The X command could have been used since it is a shorhand for G4B9C. The function
clihook is defined in BLIB and is called whenever a command is invoked. For example:

10> echo ABC Invoke the echo command

!! BPT 9: clihook Break point hit
A= 0 B= 0 3340: K4G 1
*

Notice that the values of the Cintcode registers A and B are displayed, followed by the
program counter PC and the Cintcode instruction at that point. Single step execution
is possible, for example:

* \A= 0 B= 0 24228: LLP 4

* \A= 6097 B= 0 24230: SP3
* \A= 6097 B= 0 24231: SP 89
* \A= 6097 B= 0 24233: L 80
* \A= 80 B= 6097 24235: SP 90
* \A= 80 B= 6097 24237: LLL 24272
* \A= 6068 B= 80 24239: LG 78
* \A= rdargs B= 6068 24241: K 85
* \A= 6068 B= 6068 5480: LP4
*

At this point the first instruction of rdargs is about to be executed. Its return address
is in P1, so a breakpoint can be set to catch the return, as follows:

* p1b8
* c
!! BPT 8: 24243
A= createc B= 1 24243: JNE0 24254
*
170 CHAPTER 7. THE DEBUGGER

A breakpoint can be set at the start of sys, as follows:

* g3b1 Set breakpoint 1

* b Display the currently set of breakpoints
1: sys
8: 24243
9: clihook
* 0b8 0b9 Unset breakpoints 8 and 9
* b Display the remaining breakpoint
1: sys
*

The next three calls of sys will be to write the characters ABC. The following example
steps through these and displays the state of the runtime stack just before the third
call, before leaving the debugger.

* c
!! BPT 1: sys
A= 11 B= 65 21188: SYS
* c
A
!! BPT 1: sys
A= 11 B= 66 21188: SYS
* c
B
!! BPT 1: sys
A= 11 B= 67 21188: SYS
* . 42844: Active coroutine clihook Size 20000 Hwm 127
43284: sys 11 67 312 43228
* , 43268: cnslwrf 37772
* , 43248: wrch 67 32
* , 43228: writes 42915 67
* , 42888: start 42904 42912 0 4407873
* , 42872: clihook 0
* , Base of stack
* 0b1c Clear breakpoint 1 and resume
C
210>

The following debugging commands allow the coroutine structure to be explored.

Command Effect
. Select current coroutine
, Display next stack frame
; Select parent coroutine
[ Select first coroutine
] Select next coroutine

Finally, the command Q causes a return from the Cintcode system.

7.2. THE CINTPOS DEBUGGER 171

7.2 The Cintpos Debugger

Under Cintpos, the interactive debugger can be entered by connecting the console to
task 2 (using @s02). This allows debugging to take place while other tasks are running.
Alternatively, the debugger is automatically entered in standalone mode when a fault
is encountered or by an explicit call of abort. Most of its facilities are the same as
for the Cintsys version, however a few more operations are available to access Cintpos
features. The ? command prints the following.

? Print list of debug commands

Gn Pn Rn Vn Wn An Variables
G P R V W A Pointers
123 #o377 #FF03 ’c Constants
*e /e %e +e -e |e &e ^e Dyadic operators
!e Subscription
< > Shift left/right one place
$b $c $d $f $o $s $u $x Set the print style
SGn SPn SRn SVn SWn SAn Store current value
Sn Select task n
S. Select current task
H Hold/Release selected task
K Disable/Enable clock interrupts
= Print current value
TRn Trace the next n instructions
Tn Print n consecutive locations
I Print current instruction
N Print next instruction
D Dump Cintcode memory to DUMP.mem
Q Quit -- leave the cintpos system
M Set/Reset memory watch address
B 0Bn eBn List, Unset or Set breakpoints
X (G4B9C) Set breakpoint 9 at start of clihook
Z (P1B9C) Set breakpoint 9 at return of current function
C Continue normal execution
\ Single step execute one Cintcode instruction
. ; [ ] Move to current/parent/first/next coroutine
, Move down one stack frame
a1#

The main additions as Sn to select a task, S. to select the current task and H
to hold or unhold the currently selected task. Since interrupts (particularly from the
clock device) interfere with single stepping of Cintcode instructions, the K command is
provided to turn clock interrupts on and off. The address of the task control block of
the currently selected task is given by W. Thus the first locations of the control block
can be printed by the command Wt10.
The debugger prompt contains a letter indicating whether the next instruction is
to be executed in user mode (a), in kernel mode (k) or within the interrupt service
routine (i). It also contains a number indicating which user task was running.
172 CHAPTER 7. THE DEBUGGER

7.3 Debugging Techniques

This section explores techniques that can be used to find and eliminate errors in pro-
grams. To ensure the this process is realistic a program called com/rast2wav.b of
about 1000 lines has been chosen as a case study. This program contains various pairs
of lines one correct and the other containing a a bug. Normally the line with the bug
is commented out. By changing which line is commented, it is possible to see the effect
of a bug and demonstrate how it can be found.
The program is intended to create a .wav sound file based on raster data created by
the rastering version of the BCPL system called rastsys with the aid of the command
raster. Raster data in the file RASTER can be created by the following sequence of
commands.

rastsys
c b testrast -- Compile testraster.b.
raster -- Cause the next command to
-- generate raster data.
testraster -- Actually generate the data.

This creates the raster data file RASTER representing the accesses of memory locations
during the execution of the program testraster.b whose source is:

GET "libhdr"

LET start() = VALOF

{ FOR p = 1 TO 250000 DO IF !p LOOP
RESULTIS 0
}

As can be seen this is a simple test program that that reads every Cintcode memory
word from 0 to 250000. Under 32-bit BCPL these correspond to byte addresses in the
range 0 to 1000000. The resulting file RASTER starts as follows:

F1750051 M1000000 K1000 S8

W0B71W3073B1W858B1W23B3W562B1W2B1W396B1W43B4W1B1
W213B1W5B1W2135B2W67B3W7B6W1B1N
W70B72W7325B2W14B1N
W142B71W7254B2W14B1N
W213B72W7182B2W14B1N
W284B72W7111B2W14B1N
W356B72W7039B2W14B1N

The first line specifies the rastering parameters. F1750051 states that the program exe-
cuted 1750051 Cintcode instructions. M1000000 specifies that the highest byte address
referenced was 1000000. K1000 indicates that 1000 Cintcode instructions were executed
7.3. DEBUGGING TECHNIQUES 173

per raster line and S8 says that one unit in the raster lines correspond to 8 address
bytes. What then follows are raster lines with each indentifying which addresses have
been referenced by the previous 1000 instructions. They use run length encoding with
Wn indicating that none the next n units of address space have been referenced and Bn
states that all the next n have been referenced. Each raster line is terminated by an N.
This file and others, some hand written, are used as test data for rast2wav.
The output generated by the program is a .wav file and so it is necessary to fully
understand the format of such a file. The structure is quite simple with a small header
block that describes such things as whether mono or stereo is being used and what the
sample rate is. This block is followed by 16-bit samples. Luckily it is easy to check that
the .wav file structure is correct using the freely available audacity program. This
allows the user inspect, edit and play .wav files.
Probably the most important advice on debugging is to spend sufficient time proof
reading the source code with great care. This is likely to save time in the long run. It
is, of course, essential to thoroughly understand the meaning of every construct in the
code. Misunderstanding the meaning of a statement can lead to bugs that are hard
to find. Luckily BCPL is simple and is easy to learn. Additionally, there are compile
options that help the user to check the meaning of any construct. The precedence of
expression operators such as +, -, * and / are fairly intuitive, and can be checked by
console sessions such as the following.

0.000>
0.000> type t24.b
LET f(x,y,z) = x * y / z
0.012> c b t24 tree
bcpl t24.b to t24 tree

BCPL (3 Sep 2019) 32 bit with the FLT feature

bcpl compiling to file: t24
Parse Tree
LET t24.b[1]
*-FNDEF t24.b[1]
! *-NAME: f
! *-COMMA
! ! *-NAME: x
! ! *-COMMA
! ! *-NAME: y
! ! *-NAME: z
! *-DIV
! *-MUL
! ! *-NAME: x
! ! *-NAME: y
! *-NAME: z
*-Nil
Code size = 36 bytes of 32-bit little ender Cintcode
0.044>
174 CHAPTER 7. THE DEBUGGER

This shows that x*y is computed before dividing by z. Using integer arithmetic the
result would often be different if the division was done first. This difference is significant
in the meaning of q := memvupb * (n+1) / (C7-C2+1) taken from rast2wav.b. The
D1 option is also sometimes helpful, as in:

00.000> c b t24 d1
bcpl t24.b to t24 d1

BCPL (3 Sep 2019) 32 bit with the FLT feature

bcpl compiling to file: t24
0: DATAW #x00000000
4: DATAW #x0000DFDF
8: DATAW #x2020660B
12: DATAW #x20202020
16: DATAW #x20202020
// Entry to: f
20: L1:
20: LP4
21: MUL
22: LP5
23: DIV
24: RTN
25: L2:
28: DATAW #x00000000
32: DATAW #x00000000
Code size = 36 bytes of 32-bit little ender Cintcode
0.045>

Note that the three arguments of f are held in positions 3, 4 and 5 relative to the P
pointer.
The precedence of non arithmetic operators are not so intuitive and, indeed,
tend to be different in different languages. A typical BCPL error is the belief that
IF a&7 = b&7 DO means IF (a&7)=(b&7) DO. BCPL uses operators such as & and |
for both Boolean and bit pattern operations and gives them the precedence normally
given to Boolean operators.
The transformations performed by the FLT feature and not always understood but
can be checked using the TREE2 compiler option that outputs the parse tree after these
transformations have been done by the translation phase. As an example study the
following console session.

0.000> type t25.b

LET f(x, FLT y, z) BE
x, y, z +:= 1, FLOAT x + y * 2, FIX y / z
0.012> c b t25 tree2
bcpl t25.b to t25 tree2
7.3. DEBUGGING TECHNIQUES 175

BCPL (3 Sep 2019) 32 bit with the FLT feature

bcpl compiling to file: t25

Parse Tree after calling translate

LET t25.b[1]
*-RTDEF t25.b[1]
! *-NAME: f
! *-COMMA
! ! *-NAME: x
! ! *-COMMA
! ! *-FLT
! ! ! *-NAME: y
! ! *-NAME: z
! *-SEQ
! *-ASSADD t25.b[2]
! ! *-NAME: x
! ! *-NUMBER: 1
! *-SEQ
! *-ASSFADD t25.b[2]
! ! *-NAME: y
! ! *-FADD
! ! *-FLOAT
! ! ! *-NAME: x
! ! *-FMUL
! ! *-NAME: y
! ! *-FNUM: 2.000000
! *-ASSADD t25.b[2]
! *-NAME: z
! *-DIV
! *-FIX
! ! *-NAME: y
! *-NAME: z
*-Nil
Code size = 68 bytes of 32-bit little ender Cintcode
0.047>

This shows that the so called simultaneneous assignment is, in fact, a sequence of three
assignments with the second one promoted to floating point. It shows that FLOAT and
FIX are monadic operators more binding that multiplication and division. It also shows
that FLOAT x+y*2 is transformed to FLOAT x#+y#*2.0.
Another useful debugging aid is BCPL’s cross referencing facility. A cross refer-
ence file xrast2wav can be created by the command make xrast2wav. This uses the
following commands in Makefile.

xrast2wav: allcompiled com/rast2wav.b

cintsys -c c bc rast2wav xref >rawxref
176 CHAPTER 7. THE DEBUGGER

cintsys -c sortxref rawxref to xrast2wav

rm rawxref

A few lines from xrast2wav are as follows:

fcount G:226 DEF com/rast2wav.b[97] fcount=

fcount G:226 LG com/rast2wav.b[643]
line_fsecs#:=ftotalsecs#*FLOAT fcount#/fmaxfcount
fcount G:226 LG com/rast2wav.b[705] fcount:=fcount+kval
fcount G:226 LG com/rast2wav.b[706] line_fsecs#:=pos2secs(fcount)
fcount G:226 SG com/rast2wav.b[360] fcount:=0
fcount G:226 SG com/rast2wav.b[440] fcount:=maxfcount
fcount G:226 SG com/rast2wav.b[705] fcount:=fcount+kval
filter G:209 DEF com/rast2wav.b[71] filter=
filter G:209 LG com/rast2wav.b[725] filter(notev,C7)
filter G:209 RT com/rast2wav.b[923] LET filter(v,upb)BE..

This shows that fcount was declared as global variable 226 on line 97 of rast2wav.b.
This variable is used to hold the number of Cintcode instructions obeyed to reach
the current raster line. On line 643 it is used to compute the time in seconds as a
floating point number corressponding to the time of the current raster line. The actual
statement in rast2wav.b is:

line_fsecs := ftotalsecs * FLOAT fcount / fmaxfcount

The last three lines show that the function filter was declared to be global 209
and was defined on line 923 and used just once on line 725. We can see that the
arguments in the call matches the parameters in its definition. Careful reading of the
cross reference listing can sometime find errors in the program. This is worth doing
occasionally as the program is being developed.
Another vital tool to assist debugging is the interactive debugger. This is entered
automatically when a fault is detected but can also be entered explicitly by the user.
At an early stage of debugging the following sequence of commands are useful.

0.000> abort

!! ABORT 99: User requested

* x
Breakpoint 9 at start of clihook

0.011> rast2wav

!! BPT 9: clihook
A= 0 B= 0 25100: K4G 1
*
7.3. DEBUGGING TECHNIQUES 177

This sets breakpoint 9 to be in clihook which is in the resident system. It causes

a breakpoint just as the rast2wav command is about to be entered after it has been
loaded into memory and initialised. The instruction K4G 1 is about to call the function
start in rast2wav.b. At this point we can inspect the global vector, as in:

* g+200t15

G 200: smoot’mples wrsample mark0 mark1 rdn

G 205: read_’arams read_’lines testmem notecofn filter
G 210: addnote note2str #G0212# #G0213# #G0214#
*

This shows that the 13 global functions in rast2wav.b have been correctly inition-
alised. These are useful since they allow the user to set breakpoints at the the the first
instruction of any of these functions, as in:

* g209= #G0209#
* x
Breakpoint 9 at start of clihook

0.011> rast2wav

!! BPT 9: clihook
A= 0 B= 0 25100: K4G 1
* g209= filter
* b1
* b
1: filter
9: clihook
* c
Converting file RASTER to RASTER.wav
sample_rate = 44100
mono 16-bit samples
Total time with the extra second: 11 seconds

maxaddress = 460876
maxfcount = 58862328
kval=1000 sval=8
c2=0 C3=0 C4=24 C5=36 C6=48 C7=60
Data bytes = 970184
Total number of samples: 440992

debugnote=-1

!! BPT 1: filter
178 CHAPTER 7. THE DEBUGGER

A= 145563 B= 60 63920: LM1

*

Another good way to enter the debugger is to insert calls of abort into the code
usually preceded by a call of writef or sawritef to output the values of some relevant
variables. In the early stages of debugging it is useful to call abort after the command
arguments have been decoded. For example:

0.000> c bc rast2wav
bcpl com/rast2wav.b to cin/rast2wav

BCPL (3 Sep 2019) 32 bit with the FLT feature

bcpl compiling to file: cin/rast2wav
Code size = 5516 bytes of 32-bit little ender Cintcode
0.151> rast2wav
Converting file RASTER to RASTER.wav
sample_rate = 44100
mono 16-bit samples
Total time with the extra second: 11 seconds

maxaddress = 460876
maxfcount = 58862328
kval=1000 sval=8

!! ABORT 8889: Unknown fault

*

This method means we do not need to set the breakpoint in clihook. Giving the call
of abort an essentially random argument makes it easier to find the call in the source
code later.
More to follows.
Chapter 8

The Design of OCODE

BCPL was designed to be a portable language with a compiler that is easily transferred
from machine to machine. To help to achieve this, the compiler is structured as shown
in figure 8.1 so that the codegenerator (CG), which is inherently machine dependent, is
separated from the rest of the compiler. The front end of the compiler performs syntax
analysis producing a parse tree (Tree) which is then translated by the translation phase
(TRN) to produce an intermediate form (OCODE) suitable for code generation.

BCPL LEX SYN Tree TRN OCODE CG Target

code

Figure 8.1: The structure of the compiler

8.1 Representation of OCODE

Since OCODE is output by TRN to be read in by CG, there is little need for it to
be readable by humans and so is encoded as a sequence of integers which, in the
current Cintcode implementation the OCODE is buffered in memory, however if the
compiler is not given the TO argument it does not invoke the codegenerator but, instead,
outputs the OCODE data to the file ocode in text form as a sequence of signed decimal
numbers. This numerical representation of OCODE can be transformed to a more
readable mnemonic form using the procode commands, described on page 146. As an

179
180 CHAPTER 8. THE DESIGN OF OCODE

example, if the file test.b is the following:

GET "libhdr"
LET start() BE { LET a, b, c = 1, 0, -1
writef("Answer is %n*n", a+b+c)
}

then the command: bcpl test.b would write the following text to the file ocode.:

85 2 94 1 5 115 116 97 114 116 95 3 42 1 42 0 42 -1 92 91 9 43

13 65 110 115 119 101 114 32 105 115 32 37 110 10 40 4 40 3 14
40 5 14 41 74 51 6 97 91 3 103 91 3 90 2 92 76 1 1 1

These numbers encode the OCODE statements in a natural way as can be verified
by comparing them with the following more readable form of the same statements,
generated by the procode command:

JUMP L2
ENTRY L1 5 ’s’ ’t’ ’a’ ’r’ ’t’
SAVE 3 LN 1 LN 0 LN -1 STORE STACK 9
LSTR 13 ’A’ ’n’ ’s’ ’w’ ’e’ ’r’ ’ ’ ’i’ ’s’ ’ ’ ’%’ ’n’ 10
LP 4 LP 3 ADD LP 5 ADD LG 74 RTAP 6 RTRN STACK 3
ENDPROC STACK 3 LAB L2 STORE GLOBAL 1 1 L1

8.2 The OCODE Abstract Machine

OCODE was specifically designed for BCPL and is a compromise between the desire
for simplicity and the conflicting demands of efficiency and machine independence.
OCODE is an assembly language for an abstract stack based machine that has a global
vector and an area of memory for program and static data as shown in figure 8.2.

Global vector Current stack frame

S
G P
Memory for program and static data

Li Lj

Figure 8.2: The BCPL abstract machine

The OCODE machine has five registers G, P, PC, A and L. G points to the global
vector and P points to the stack frame of the currently executing function. PC points
to the next instruction to execute. The current size of the stack frame is held in S but
S does not require a physical register since its value at every point in the program is
8.3. LOADING AND STORING VALUES 181

known by both the frontend of the compiler and the codegenerator. They both know
the effect on S of every OCODE statement.
Static variables, tables and string constants are allocated space embedded in the
compiled code and are referenced using labels such as L36 and L92. All global, local
and static variables are of the same size and, on most modern implementations, they
hold 32 bit values. More recently 64 bit versions of BCPL are becoming more common.
OCODE is normally encoded as a compacted sequence of integers since it is gen-
erated by the frontend of the compiler and read by the codegenerator. For human
consumption a more readable form can be created using the procode command de-
scribed on page 146.. An OCODE statement consists of a function or directive code
possibly followed by operands that are either optionally signed integers, quoted char-
acters or labels of the form Ln where n is a label number. The following are examples
of mnemonic OCODE statements:

LSTR 5 ’H’ ’e’ ’l’ ’l’ ’o’

LP 3
GETBYTE
SL L36

There are OCODE statements for loading and storing values, for applying expres-
sion operators, for the implementation of functions and routines, and to control the
flow of execution. There are also directives for the allocation of storage and to allow
information to be passed to the codegenerator.

8.3 Loading and Storing values

A variables may be local, global or static, and may be accessed in three ways depending
on its context, and so there are 9 statements for accessing variables as shown in the
following table.

Statement Meaning
LP n P!S := P!n; S := S+1
LG n P!S := G!n; S := S+1
LL Ln P!S := Ln; S := S+1
LLP n P!S := @P!n; S := S+1
LLG n P!S := @G!n; S := S+1
LLL Ln P!S := @Ln; S := S+1
SP n S := S-1; P!n := P!S
SG n S := S-1; G!n := P!S
SL Ln S := S-1; Ln := P!S
RSTACK n P!n := A; S := n+1

The RSTACK statement is used in conjunction with the RES statement in the com-
pilation of VALOF expressions. See page 186 for details.
The following tables shows the six statements for loading constants.
182 CHAPTER 8. THE DESIGN OF OCODE

Statement Meaning
LF Ln P!S := entry point Ln; S := S+1
LN n P!S := n; S := S+1
FNUM m e P!S := <make float>(m, e); S := S+1
TRUE P!S := TRUE; S := S+1
FALSE P!S := FALSE; S := S+1
QUERY P!S := ?; S := S+1
LSTR n C1 . . . Cn P!S := "C1 . . . Cn "; S := S+1

LF Ln loads the entry point of a non global function onto the stack. LN n loads the
signed integer constant n onto the stack. FNUM m e loads a floating point approximation
of a number with signed integer mantissa m and signed integer decimal exponent e onto
the stack. The statements TRUE and FALSE are present to improve portability between
machines that use different representations for the integers. For instance, on machines
using ones complement or sign and modulus arithmetic, TRUE is not equivalent to LN
-1. QUERY loads an undefined value onto the stack, and the LSTR statement allocates a
string in static memory and loads a pointer to it onto the stack.
Indirect assignments and assignments to elements of word and byte arrays normally
use the statements STIND and PUTBYTE whose meanings are given in table 5.3.

Statement Meaning
STIND !(P!(S-1)) := P!(S-2); S := S-2
PUTBYTE (P!(S-2))%(P!(S-1)) := P!(S-3); S := S-3

Assuming ptr is in global 200, the following assignments:

!ptr := 12; ptr!3 := 99; ptr%3 := 65

translate into the following OCODE:

LN 12 LG 200 STIND
LN 99 LG 200 LN 3 ADD STIND
LN 65 LG 200 LN 3 PUTBYTE

8.4 Field Selection Operators

Accessing and updating fields as required by the OF operator are implemented using
the OCODE operators SELLD and SELST.

SELLD takes two argments len and sh. It effect is equivalent to

P!(S-1) := !(P!(S-1)) >> sh & mask
where mask is a bit pattern containing len right justified ones. If em len is zero no
masking is done.

SELST takes three argments op, len and sh. If op is zero, its effect is equivalent to
SLCT len:sh:0 OF (P!(S-1)) := P!(S-2); S := S-2
8.5. EXPRESSION OPERATORS 183

but if op is non zero it represents and assignment operator (assop) and the statement
is equivalent to:
SLCT len:sh:0 OF (P!(S-1)) assop:= P!(S-2); S := S-2
The mapping between op and assop is given by the following table.

op assop op assop op assop

0 none 1 ! 2 #*
3 #/ 4 #MOD 5 #+
6 #- 7 * 8 /
9 MOD 10 + 11 -
12 << 13 >> 14 &
15 | 16 EQV 17 XOR

The floating-point assignment operators are only allowed when the specified field
is a full word, typically with len and sh both zero. The SELST operator with len and
sh both zero is used in the compilation assop:= assignments where the left hand side
is a simple variable or a subscripted expression. For instance, the assigment v!3+:=1
might generate the following OCODE.

LN 1
LN 3 LG 200 ADD
SELST ADD 0 0

8.5 Expression Operators

The monadic expression operators only affect the topmost item of the stack and do not
change the value of S. They are shown in the next table.

Statement Meaning
RV P!(S-1) := ! P!(S-1)
ABS P!(S-1) := ABS P!(S-1)
FABS P!(S-1) := FABS P!(S-1)
FLOAT P!(S-1) := FLOAT P!(S-1)
FIX P!(S-1) := FIX P!(S-1)
NEG P!(S-1) := - P!(S-1)
FNEG P!(S-1) := #- P!(S-1)
NOT P!(S-1) := ∼ P!(S-1)

All dyadic expression operators take two operands from stack replacing them the
result and decrementing S by 1. These operators are shown in the following table.
184 CHAPTER 8. THE DESIGN OF OCODE

Statement Meaning
GETBYTE S := S-1; P!(S-1) := P!(S-1) % P!S
MUL S := S-1; P!(S-1) := P!(S-1) * P!S
FMUL S := S-1; P!(S-1) := P!(S-1) #* P!S
DIV S := S-1; P!(S-1) := P!(S-1) / P!S
FDIV S := S-1; P!(S-1) := P!(S-1) #/ P!S
MOD S := S-1; P!(S-1) := P!(S-1) MOD P!S
ADD S := S-1; P!(S-1) := P!(S-1) + P!S
FADD S := S-1; P!(S-1) := P!(S-1) #+ P!S
SUB S := S-1; P!(S-1) := P!(S-1) - P!S
FSUB S := S-1; P!(S-1) := P!(S-1) #- P!S
EQ S := S-1; P!(S-1) := P!(S-1) = P!S
FEQ S := S-1; P!(S-1) := P!(S-1) #= P!S
NE S := S-1; P!(S-1) := P!(S-1) ∼= P!S
FNE S := S-1; P!(S-1) := P!(S-1) #∼= P!S
LS S := S-1; P!(S-1) := P!(S-1) < P!S
FLS S := S-1; P!(S-1) := P!(S-1) #< P!S
GR S := S-1; P!(S-1) := P!(S-1) > P!S
FGR S := S-1; P!(S-1) := P!(S-1) #> P!S
LE S := S-1; P!(S-1) := P!(S-1) <= P!S
FLE S := S-1; P!(S-1) := P!(S-1) #<= P!S
GE S := S-1; P!(S-1) := P!(S-1) >= P!S
FGE S := S-1; P!(S-1) := P!(S-1) #>= P!S
LSHIFT S := S-1; P!(S-1) := P!(S-1) << P!S
RSHIFT S := S-1; P!(S-1) := P!(S-1) >> P!S
LOGAND S := S-1; P!(S-1) := P!(S-1) & P!S
LOGOR S := S-1; P!(S-1) := P!(S-1) | P!S
EQV S := S-1; P!(S-1) := P!(S-1) EQV P!S
XOR S := S-1; P!(S-1) := P!(S-1) XOR P!S

Vector subscription (E1 !E2 is implemented using PLUS and RV. The value of x MOD
y is either zero or it has the same sign as x and its magnitude is less than ABS y. Shifts
by a negative amounts yield undefined results, but on some versions of BCPL the shift
direction is reversed. Shifts greater than the word length yield zero.

8.6 Functions and Routines

The design of the OCODE statements for the implementation of function and routine
calls have been designed with care to allow code generators as much freedom as possible.
The mechanism allows some arguments to be passed in registers if this is worthwhile,
and the distribution of work between the code for a call and the code at the entry
point is up to the implementer. In a typical program there are about five calls for each
function or routine and so there is some incentive to keep the size of the call small by
transferring some of the work to the save sequence.
8.6. FUNCTIONS AND ROUTINES 185

The compilation of a function or routine definition generates an OCODE sequence

of the following form:

ENTRY Li n C1 . . . Cn
SAVE s
body of function or routine
ENDPROC

Li is the label allocated for the entry point. As a debugging aid, the length of the
function or routine name is given by n and its characters by the C1 . . . Cn . The SAVE
statement specifies the initial setting of S, which is just the save space size (=3) plus
the number of formal parameters. For functions defined using pattern matching the
number of formal parameters is determined by the patterns in the match list. The
state of the stack just after entry is shown in figure 8.3.

Save space Procedure arguments

A1 A2 An
S
P
Figure 8.3: The stack frame on function or routine entry

The save space is used to hold the previous value of P, the return address and the
function entry address. Thus, the first argument of a function is always at position 3
relative to the P pointer. On some older versions of BCPL the size of the save space
may be different.
The end of the body is marked by an ENDPROC statement which is non executable
but allows the code generator to keep track of nested definitions. In early versions of
OCODE, the first two arguments of ENTRY were interchanged and ENDPROC was given
a numerical argument.
The language insists that arguments are laid out in consecutive locations on the
stack and that there is no limit to their number. This suggests that a good strategy is
to place the arguments of a call in the locations they must occupy when the function or
routine is entered. Thus, a typical call E(E1 , . . . , En ) is compiled by first incrementing
S to leave room for the save space in the new stack frame, then generate code to evaluate
the arguments E1 , . . . , En before generating code for E. The state is then as shown
in figure 8.4. Finally, either FNAP k or RTAP k is generated, depending on whether a
function or routine call is being compiled. Notice that k is the distance between the
old and new stack frames.
The return from a routine is performed by RTRN which restores the previous value
of P and resumes execution from the return address. The return from a function is
performed by FNRN just after the function result has been evaluated on the top of the
186 CHAPTER 8. THE DESIGN OF OCODE

Old stack frame New stack frame

E1 E2 En E

k
S
P

Figure 8.4: The moment of calling E(E1,E2,...En)

stack. FNRN performs the same action as RTRN, after placing the function result in a
special register (A) ready for FNAP to store it in the required location in the previous
stack frame.

8.7 Control
The PC register holds the address of the next instructions to be executed. For most
OCODE statements it is incremented by the size of the instruction, but for control
statements PC is set specifically as needed as needed by, for instance, conditional jumps
or SWITCHON commands. The OCODE statements concerned with function and rou-
tine calls have already been described above. The remaining control statements are
described here.
LAB Ln Ln := PC
This sets the value of label Ln to the current position in the compiled code.
JUMP Ln PC := Ln
JT Ln s := S-1; IF P!S DO PC := Ln
JF Ln S := S-1; UNLESS P!S DO PC := Ln
JUMP causes an unconditional jump to the instruction labelled by Ln. Both JT and
JF pop the top item from the stack then they conditionally jump to label Ln depending
on its value.
GOTO S := S-1; PC := P!S
This is used in the translation of the BCPL GOTO command.
RES Ln S := S-1; A := P!S; PC := Ln
This is used in the compilation of RESULTIS commands. The result is evaluated
and placed on the top of the stack, followed by a RES statement which pops the result
from the stack and places the it in register A before jumping to Ln. This label points
to the first instruction after the code for the VALOF block where there is an RSTACK
statement will push A onto the top of the stack after setting S appropriately for this
point in the program.
If the VALOF block is the body of a function, the compiled code for its RESULTIS
commands is optimised using FNRN rather than RES.
SWITCHON n LdK1 L1 . . . Kn Ln
8.8. DIRECTIVES 187

This is used in the compilations of switches. It makes a jump determined by the value
on the top of the stack. Its first argument (n) is the number of cases in the switch and
the second argument (Ld) is the default label. K1 to Kn are the case constants and L1
to Ln are the corresponding labels.
FINISH Ln S := S-1; A := P!S; PC := Ln
This statement is the compilation of the BCPL FINISH command. It should be con-
verted by the codegenerator into code equivalent to stop(0) by the code generator.
Users are strongly discourage from using FINISH

8.8 Directives
Sometimes the size of the stack frame changes other than in the course of expression
evaluation. This happens, for instance, when control leaves a block in which local
variables were declared. The statement STACK s informs the code generator that the
size of the current stack frame is now s.
The STORE statement is used to inform the code generator that the point separating
the declarations and body of a block has been reached and that any anonymous results
on the stack are actually initialised local variables and so should be stored in their true
stack locations.
Static variables and tables are allocated space in the program area using statements
of the form ITEMN n, where n is the initial value of the static cell. The elements of
table are placed in consecutive locations by consective ITEMN statements. A label may
be set to the address of a static cell by preceding the ITEMN statement by a statement
of the form DATALAB Ln.
The SECTION and NEEDS directives in a BCPL program translate into SECTION and
NEEDS statements of the form:

SECTION n C1 . . . Cn
NEEDS n C1 . . . Cn

where C1 to Cn are the characters of the SECTION or NEEDS name and n is the length.
The end of an OCODE module is marked by the GLOBAL statement which con-
tains information about global functions, routines and labels. The form of the GLOBAL
statement is as follows:

GLOBAL n K1 L1 . . . Kn Ln

where n is the number of items in the global initialisation list. Ki is the global number
and Li is its label. When a module is loaded its global entry points must be initialised.

8.9 Discussion
A very early version of OCODE used a three address code in which the operands were
allowed to be the sum of up to three simple values with a possible indirection. The
intention was that reasonable code should be obtainable even when codegenerating
188 CHAPTER 8. THE DESIGN OF OCODE

one statement at a time. It was soon found more convenient to use an intermediate
code that separates the accessing of values from the application of operators. This
improved portability by making it possible to implement very simple non optimising
codegenerators. Optimising codegenerators could absorb several OCODE statements
before emitting compiled code.
The TRUE and FALSE statements were added in 1968 to improve portability to
machines using sign and modulus or one’s complement arithmetic. Luckily two’s com-
plement arithmetic has now become the norm. Other extensions to OCODE, notably
the ABS, QUERY, GETBYTE and PUTBYTE statements were added as the corresponding
constructs appeared in the language.
In 1980, the BCPL changed slightly to permit position independent code to be
compiled. This change specified that non global functions, routines and labels were
no longer variables, and the current version of OCODE reflects this change by the
introduction of the LF statement and the removal of the old ITEML statement that used
to allocate static cells for such entry points.
Another minor change in this version of OCODE is the elimination of the ENDFOR
statement that was provided to fix a problem on 16-bit word addressed machines with
more than 64 Kbytes of memory.
Chapter 9

The Design of Cintcode

The original version of Cintcode was a byte stream interpretive code designed to be
both compact and capable of efficient interpretation on small 16 bit machines based
on 8 bit micro processors such as the Z80 and 6502. Versions that ran on the BBC
Microcomputer and under CP/M were marketed by RCP Ltd [2]. The current version
of Cintcode was extended for 32 bit implementations of BCPL and mainly differs from
the original by the provision of 32 bit operands and the removal of a size restriction of
the global vector.
There is now also a version of Cintcode for 64-bit implementations of BCPL. This is
almost identical to the 32-bit version. A nineth Cintcode register (MW) has been added.
This is normally zero but can be set by a new Cintcode instruction (MW), see below.
On 64-bit implementations, the instructions that take four byte immediate operands,
namely KW, LLPW, LW, LPW, SPW, APW, and AW, sign extend the four byte immediate
operand before adding the MW register into the senior half of the 64-bit result before
resetting the MW to zero. In this version static variables are allocated in 64-bit 8 byte
aligned locations.
The Cintcode machine has nine registers as shown in figure 9.1.

Stack frame Global vector Program area

Registers

A
B
C
P
G
ST
PC
Count
MW

Figure 9.1: The Cintcode machine

189
190 CHAPTER 9. THE DESIGN OF CINTCODE

The registers A and B are used for expression evaluation, and C is used in in byte
subscription. P and G are pointers to the current stack frame and the global vector,
respectively. ST is used as a status register in the Cintpos version of Cintcode, and PC
points to the first byte of the next Cintcode instruction to execute. Count is a register
used by the debugger. While it is positive, Count is decremented on each instruction
execution, raising an exception (code 3) on reaching zero. When negative, it causes a
second (faster) interpreter to be used.
Cintcode encodes the most commonly occurring operations as single byte instruc-
tions, using multi-byte instructions for rarer operations. The first byte of an instruction
is the function code. Operands of size 1, 2 or 4 bytes immediately follow some function
bytes. The two instructions used to implement switches have inline data following the
function byte. Cintcode modules also contains static data for stings, integers, tables
and global initialisation data.

9.1 Designing for Compactness

To obtain a compact encoding, information theory suggests that each function code
should occur with approximately equal frequency. The self compilation of the BCPL
compiler, as shown in figure 4.2, was the main benchmark test used to generate fre-
quency information and a summary of how often various operations are used during this
test is given in table 9.1. This data was produced using the tallying feature controlled
by the stats command, described on page 151.
The statistics from different programs vary greatly, so while encoding the common
operations really compactly, there is graceful degradation for the rarer cases ensuring
that even unusual programs are handled reasonably well. There are, for instance,
several one byte instructions for loading small integers, while larger integers are handled
using 2, 3 and 5 byte instructions. The intention is that small changes in a source
program should cause small small changes in the size of the corresponding compiled
code.
Having several variant instructions for the same basic operation does not greatly
complicate the compiler. For example the four variants of the AP instruction that adds
a local variable into register A is dealt with by the following code fragment taken from
the codegenerator.

TEST 3<=n<=12 THEN gen(f_ap0 + n)

ELSE TEST 0<=n<=255
THEN genb(f_ap, n)
ELSE TEST 0<=n<=#xFFFF
THEN genh(f_aph, n)
ELSE genw(f_apw, n)

It is clear from table 9.1 that accessing variables and constants requires special care,
and that conditional jumps, addition, calls and indirection are also important. Since
access to local variables accounts for about a quarter of the operations performed, about
this proportion of codes were allocated to instructions concerned with local variables.
Local variables are allocated words in the stack starting at position 3 relative to the P
9.1. DESIGNING FOR COMPACTNESS 191

Operation Executions Static count

Loading a local variable 3777408 1479

Updating a local variable 1965885 1098
Loading a global variable 5041968 1759
Updating a global variable 796761 363
Using a positive constant 4083433 1603
Using a negative constant 160224 93
Conditional jumps (all) 2013013 488
Conditional jumps on zero 494282 267
Unconditional direct jump 254448 140
Unconditional indirect jumps 152646 93
Procedure calls 1324206 1065
Procedure returns 1324204 381
Binary chop switches 43748 12
Label vector switches 96461 17
Addition 2135696 574
Subtraction 254935 111
Other expression operations 596882 74
Loading a vector element 1356315 429
Updating a vector element 591268 137
Loading a byte vector element 476688 53
Updating a byte vector element 405808 29
Table 9.1: Counts from the BCPL self compilation test

pointer and, as one would expect, small numbered locals are used far more frequently
than the others, so operations on low numbered locals often have single byte codes.
Although not shown here, other statistics, such as the distribution of relative ad-
dressing offsets and operand values, influenced the design of Cintcode.

9.1.1 Global Variables

Global variables are referenced as frequently as locals and therefore have many function
codes to handle them. The size of the global vector in most programs is less than 512,
but Cintcode allows this to be as large are 65536 words. Each operation that refers to a
global variable is provided with three related instructions. For instance, the instructions
to load a global into register A are as follows:
192 CHAPTER 9. THE DESIGN OF CINTCODE

LG b B := A; A := G!b

LG1 b B := A; A := G!(b+256)

LGH h B := A; A := G!h
Here, b and h are unsigned 8 and 16 bit values, respectively.

9.1.2 Composite Instructions

Compactness can be improved by combining commonly occurring pairs (and triples)
of operations into a single instructions. Many such composite instructions occur in
Cintcode; for instance, AP3 adds local 3 to the A register, and L1P6 will load v!1 into
register A, assuming v is held in local 6.

9.1.3 Relative Addressing

A relative addressing mechanism is used in conditional and unconditional jumps and
the instructions: LL, LLL, SL and LF. All these instructions refer to locations within
the code and are optimised for small relative distances. To simplify the codegenerator
all relative addressing instructions are 2 bytes in length. The first being the function
code and the second being an 8 bit relative address.

Direct J a dest = x + a

PC x

Indirect J$ b hh dest = q + hh

PC x q = (x & #xFFFFFFFE) + 2*b

Figure 9.2: The relative addressing mechanism

All relative addressing instructions have two forms: direct and indirect, depending
on the least significant bit of the function byte. The details of both relative address
calculations are shown in figure 9.2, using the instructions J and J$ as examples. For
the direct jump (J), the operand (a) is a signed byte in the range -128 to +127 which
is added to the address (x) of the operand byte to give the destination address (dest).
For the indirect jump, J$, the operand (b) is an unsigned byte in the range 0 to 255
which is doubled and added to the rounded version of x to give the address (q) of a
16 bit signed value hh which is added to q to give the destination address (dest).
The compiler places the resolving half word as late as possible to increase the chance
that it can be shared by other relative addressing instructions to the same desination,
as could happen when several ENDCASE statements occur in a large SWITCHON
9.2. THE CINTCODE INSTRUCTION SET 193

command. The use of a 16 bit resolving word places a slight restriction on the maximum
size of relative references. Any Cintcode module of less than 64K bytes will have no
problem.

9.2 The Cintcode Instruction Set

The resulting selection of function codes is shown in Table 9.2 and they are described
in the sections that follow. In the remaining sections of this chapter the following
conventions hold:
Symbol Meaning
n An integer encoded in the function byte.
Ln The one byte operand of a relative addressing instruction.
b An unsigned byte, range 0 ≤ b ≤ 255.
h An unsigned halfword, range 0 ≤ h ≤ 65535.
w A signed 32 bit word.
filler Optional filler byte to round up to a 16 bit boundary.
A The Cintcode A register.
B The Cintcode B register.
C The Cintcode C register.
P The Cintcode P register.
G The Cintcode G register.
PC The Cintcode PC register.
MW The Cintcode MW register used in 64-bit Cintcode.

9.2.1 Byte Ordering and Alignment

A Cintcode module is a vector of 32 bit words containing the compiled code and static
data of a section of program. The first word of a module holds its size in words that
is used as a relative address to the end of the module where the global initialisation
data is placed. The last word of a module holds the highest referenced global number,
and working back, there are pairs of words giving the global number and relative entry
address of each global function or label defined in the module. A relative address of
zero marks the end of the initialisation data. See section 8.3 for more details.
The compiler can generate code for either a big- or little-endian machine. These
differ only in the byte ordering of bytes within words. For a little endian machine, the
first byte of a 32 bit word is at the least significant end, and on a big-endian machine, it
is the most significant byte. This affect the ordering of bytes in 2 and 4 byte immediate
operands, 2 byte relative address resolving words, 4 byte static quantities and global
initialisation data. Resolving words are aligned on 16 bit boundaries relative to the
start of the module, and 4 byte statics values are aligned on 32 bit boundaries. The 2
and 4 byte immediate operands are not aligned.
For efficiency reasons, the byte ordering is chosen to suit the machine on which the
code is to be interpreted. The compiler option OENDER causes the BCPL compiler to
194 CHAPTER 9. THE DESIGN OF CINTCODE

0 32 64 96 128 160 192 224

0 - K LLP L LP SP AP A
1 FLTOP KH LLPH LH LPH SPH APH AH
2 BRK KW LLPW LW LPW SPW APW AW
3 K3 K3G K3G1 K3GH LP3 SP3 AP3 L0P3
4 K4 K4G K4G1 K4GH LP4 SP4 AP4 L0P4
5 K5 K5G K5G1 K5GH LP5 SP5 AP5 L0P5
6 K6 K6G K6G1 K6GH LP6 SP6 AP6 L0P6
7 K7 K7G K7G1 K7GH LP7 SP7 AP7 L0P7
8 K8 K8G K8G1 K8GH LP8 SP8 AP8 L0P8
9 K9 K9G K9G1 K9GH LP9 SP9 AP9 L0P9
10 K10 K10G K10G1 K10GH LP10 SP10 AP10 L0P10
11 K11 K11G K11G1 K11GH LP11 SP11 AP11 L0P11
12 LF S0G S0G1 S0GH LP12 SP12 AP12 L0P12
13 LF$ L0G L0G1 L0GH LP13 SP13 XPBYT S
14 LM L1G L1G1 L1GH LP14 SP14 LMH SH
15 LM1 L2G L2G1 L2GH LP15 SP15 BTC MDIV
16 L0 LG LG1 LGH LP16 SP16 NOP CHGCO
17 L1 SG SG1 SGH SYS S1 A1 NEG
18 L2 LLG LLG1 LLGH SWB S2 A2 NOT
19 L3 AG AG1 AGH SWL S3 A3 L1P3
20 L4 MUL ADD RV ST S4 A4 L1P4
21 L5 DIV SUB RV1 ST1 XCH A5 L1P5
22 L6 MOD LSH RV2 ST2 GBYT RVP3 L1P6
23 L7 XOR RSH RV3 ST3 PBYT RVP4 L2P3
24 L8 SL AND RV4 STP3 ATC RVP5 L2P4
25 L9 SL$ OR RV5 STP4 ATB RVP6 L2P5
26 L10 LL LLL RV6 STP5 J RVP7 L3P3
27 FHOP LL$ LLL$ RTN GOTO J$ ST0P3 L3P4
28 JEQ JNE JLS JGR JLE JGE ST0P4 L4P3
29 JEQ$ JNE$ JLS$ JGR$ JLE$ JGE$ ST1P3 L4P4
30 JEQ0 JNE0 JLS0 JGR0 JLE0 JGE0 ST1P4 SELLD
31 JEQ0$ JNE0$ JLS0$ JGR0$ JLE0$ JGE0$ MW SELST
Table 9.2: The Cintcode function codes
9.2. THE CINTCODE INSTRUCTION SET 195

compile code with the opposite endianess to that of the machine on which the compiler
is running, see the description of the bcpl command on page 131.

9.2.2 Loading Values

The following instructions are used to load constants, variables, the addresses of vari-
ables and function entry points. Notice that all loading instructions save the old value
of register A in B before updating A. This simplifies the translation of dyadic expression
operators.

Ln 0 ≤ n ≤ 10 B := A; A := n
LM1 B := A; A := -1
L b B := A; A := b
LM b B := A; A := -b
LH h B := A; A := h
LMH h B := A; A := -h
LW w B := A; A := w
MW w MW := w

These instructions load integer constants. Constants are in the range -1 to 10 are the
most common and have single byte instructions. The other cases use successively larger
instructions. The MW instruction is only used in 64-bit Cintcode. See page 189 for more
details.

LPn 3 ≤ n ≤ 16 B := A; A := P!n
LP b B := A; A := P!b
LPH h B := A; A := P!h
LPW w B := A; A := P!w

These instructions load local variables and anonymous results addressed relative to P.
Offsets in the range 3 to 16 are the most common and use single byte instructions. The
other cases use succesively larger instructions.

LG b B := A; A := G!b
LG1 b B := A; A := G!(b + 256)
LGH h B := A; A := G!h

LG loads the value of a global variables in the range 0 to 255, LG1 load globals in the
range 256 to 511, and LGH can load globals up to 65535. Global numbers must be in
the range 0 to 65535.

LL Ln B := A; A := variable Ln
LL$ Ln B := A; A := variable Ln
LF Ln B := A; A := entry point Ln
LF$ Ln B := A; A := entry point Ln
196 CHAPTER 9. THE DESIGN OF CINTCODE

LL loads the value of a static variable and LF loads the entry address of a function,
routine or label in the current module.

LLP b B := A; A := @P!b
LLPH h B := A; A := @P!h
LLPW w B := A; A := @P!w
LLG b B := A; A := @G!b
LLG1 b B := A; A := @G!(b + 256)
LLGH h B := A; A := @G!h
LLL Ln B := A; A := @(variable Ln)
LLL$ Ln B := A; A := @(variable Ln)

These instructions load the BCPL pointers to local, global and static variables.

9.2.3 Indirect Load

GBYT A := B%A
RV A := A!0
RVn 1≤n≤6 A := A!n
RVPn 3≤n≤7 A := P!n!A
L0Pn 3 ≤ n ≤ 12 B := A; A := P!n!0
L1Pn 3≤n≤6 B := A; A := P!n!1
L2Pn 3≤n≤5 B := A; A := P!n!2
L3Pn 3≤n≤4 B := A; A := P!n!3
L4Pn 3≤n≤4 B := A; A := P!n!4
LnG b 0≤n≤2 B := A; A := G!b!n
LnG1 b 0≤n≤2 B := A; A := G!(b+256)!n
LnGH h 0≤n≤2 B := A; A := G!h!n

These instructions are used in the implementation of byte and word indirection oper-
ators % and ! in right hand contexts.

9.2.4 Expression Operators

NEG A := -A
NOT A := ~A

These instructions implement the three monadic expression operators.

MUL A := B * A
DIV A := B / A
MOD A := B MOD A
ADD A := B + A
SUB A := B - A
LSH A := B << A
RSH A := B >> A
9.2. THE CINTCODE INSTRUCTION SET 197

AND A := B & A
OR A := B | A
XOR A := B XOR A
These instructions provide for all the normal arithmetic and bit pattern dyadic opera-
tors. The instructions DIV and MOD generate exception 5 if the divisor is zero. Evalua-
tion of relational operators in non conditional contexts involve conditional jumps and
the FHOP instruction, see page 200. Addition is the most frequently used arithmetic
operation and so there are various special instructions improve its efficiency.
An 1≤n≤5 A := A + n
Sn 1≤n≤4 A := A - n
A b A := A + b
AH h A := A + h
AW w A := A + w
S b A := A - b
SH h A := A - h
These instructions implement addition and subtraction by constant integer amounts.
There are single byte instructions for incrementing by 1 to 5 and decremented by 1 to
4. For other values longer instructions are available.
APn 3 ≤ n ≤ 12 A := A + P!n
AP b A := A + P!b
APH h A := A + P!h
APW w A := A + P!w
AG b A := A + G!b
AG1 b A := A + G!(b+256)
AGH h A := A + G!h
These instructions allow local and global variables to be added to A. Special instructions
for addition by static variables are not provided, and subtraction by a variable is not
common enough to warrant special treatment.

9.2.5 Simple Assignment

SPn 3 ≤ n ≤ 16 P!n := A
SP b P!b := A
SPH h P!h := A
SPW w P!w := A
SG b G!b := A
SG1 b G!(b+256) := A
SGH h G!h := A
SL Ln variable Ln := A
SL$ Ln variable Ln := A
These instructions are used in the compilation of assignments to named local, global
and static variables. The SP instructions are also used to save anonymous results and
to layout function arguments.
198 CHAPTER 9. THE DESIGN OF CINTCODE

9.2.6 Indirect Assignment

PBYT B%A := C
XPBYT A%B := C
ST A!0 := B
STn 1≤n≤3 A!n := B
ST0Pn 3≤n≤4 P!n!0 := A
ST1Pn 3≤n≤4 P!n!1 := A
STPn 3≤n≤5 P!n!A := B
S0G b G!b!0 := A
S0G1 b G!(b+256)!0 := A
S0GH h G!h!0 := A

These instructions are used in assignments in which % or ! appear as the leading

operator on the left hand side.

9.2.7 Function and Routine Calls

At the moment a function or routine is called the state of the stack is as shown in
figure 9.3. At the entry point of a function or routine the first argument, if any, will
be in register A and in memory P!3.

Old stack frame New stack frame

E2 En

Figure 9.3: The moment of calling E(E1,E2,...En)

9.2. THE CINTCODE INSTRUCTION SET 199

Kn 3 ≤ n ≤ 11
K b
KH h
KW w

These instructions call the function or routine whose entry point is in A and whose first
argument (if any) is in B. The new stack frame at position k relative to P where k is n,
b, h or w depending on which instruction is used. The effect of these instructions is as
follows:

P!k := P // Save the old P pointer

P := P+k // Set its new value
P!1 := PC // Save the return address
PC := A // Set PC to the entry point
P!2 := PC // Save it in the stack for debugging
A := B // Put the first argument in A
P!3 := A // Save it in the stack

As can be seen, three words of link information (the old P pointer, the return address
and entry address) are stored in the base of the new stack frame.

KnG b 3 ≤ n ≤ 11
KnG1 b 3 ≤ n ≤ 11
KnGH h 3 ≤ n ≤ 11

These instructions deal with the common situation where the entry point of the function
is in the global vector and the stack increment is in the range 3 to 11. The global number
gn is b, b + 256 or h depending on which function code is used and stack increment k
is n. The first argument (if any) is in A. The effect of these instructions is as follows:

P!k := P // Save the old P pointer

P := P+k // Set its new value
P!1 := PC // Save the return address
PC := G!gn // Set the new PC value from the global value
P!2 := PC // Save it in the stack for debugging
P!3 := A // Save the first argument in the stack

RTN

This instruction causes a return from the current function or routine using the previous
P pointer and the return address held in P!0 and P!1. The effect of the instruction is
as follows:

PC := P!1 // Set PC to the return address

P := P!0 // Restore the old P pointer

When returning from a function the result will be in A.

200 CHAPTER 9. THE DESIGN OF CINTCODE

9.2.8 Flow of Control and Relations

The following instructions are used in the compilation of conditional and unconditional
jumps, and relational expressions. The symbol rel denotes EQ, NE, LS, GR, LE or GE
indicating the relation being tested.

J Ln PC := Ln
J$ Ln PC := Ln
Jrel Ln IF B rel A DO PC := Ln
Jrel$ Ln IF B rel A DO PC := Ln
Jrel0 Ln IF A rel 0 DO PC := Ln
Jrel0$ Ln IF A rel 0 DO PC := Ln

The destinations of these jump instructions are computed using the relative addressing
mechanism described in Section 9.1.3. Notice than when the comparison is with zero,
A holds the left operand of the relation.

GOTO PC := A

This instruction is only used in the compilation of the GOTO command.

FHOP A := 0; PC := PC+1

The FHOP instruction is only used in the compilation of relational expressions in non
conditional contexts as in the compilation. The assignment: x := y < z is typically
compiled as follows:

LP4 Load y
LP5 Load z
JLS 2 Jump to the LM1 instruction if y<z
FHOP A := FALSE; and hop over the LM1 instruction
LM1 A := TRUE
SP3 Store in x

9.2.9 Switch Instructions

The instructions are used to implement switches are SWL and SWB, switching on the
value held in A. They both assume that all case constants are in the range 0 to 65535,
with the compiler taking appropriate action when this constraint is not satisfied.

SWL filler n dlab L0 . . . Ln−1

This instruction is used when there are sufficient case constants all within a small
enough range. It performs the jump by selecting an element from a vector of 16 bit
resolving half words. The quantities n, dlab, and L0 to Ln−1 are 16 bit half words,
aligned on 16 bit boundaries by the optional filler byte. If A is in the range 0 to n − 1 it
uses the appropriate resolving half word LA , otherwise it uses the resolving half word
9.2. THE CINTCODE INSTRUCTION SET 201

dlab to jump to the default label. See Section 9.1.3 for details on how resolving half
words are interpreted.

SWB filler n dlab K1 L1 . . . Kn Ln

This instruction is used when the range of case constants is too large for SWL to be
economical. It performs the jump using a binary chop strategy. The quantities n, dlab,
K1 to Kn and L1 to Ln are 16 bit half words aligned on 16 bit boundaries by the
option filler byte. This instruction successively tests A with the case constants in the
balanced binary tree given in the instruction. The tree is structured in a way similar
to that used in heapsort with the children of the node at position i at positions 2i and
2i + 1. References to nodes beyond n are treated as null pointers. Within this tree, Ki
is greater than all case constants in the tree rooted at position 2i, and less than those
in the tree at 2i + 1. The search starts at position 1 and continues until a matching
case constant is found or a null pointer is reached. If A is equal to some Ki then PC is
set using the resolving half word Li , otherwise it uses the resolving half word dlab to
jump to the default label. See Section 9.1.3 for details on how resolving half words are
interpreted.
The use of this structure is particularly good for the hand written machine code
interpreter for the Pentium where there are rather few central registers. Cunning use
can be made of the add with carry instruction (adcl). In the following fragment of
code, %esi points to n, %eax holds i and A is held in %eab. There is a test elsewhere
to ensure that A is in the range 0 to 65535.

swb1: cmpw (%esi,%eax,4),%bx ; { compare A with Ki

je swb3 ; Jump if A=Ki
adcl %eax,%eax ; IF A>Ki THEN i := 2i
; ELSE i := 2i+1
cmpw (%esi),%ax ;
jle swb1 ; } REPEATWHILE i<=n

The compiler ensures that the tree always has at least 7 nodes allowing the code can
be further improved by preceeding this loop with two copies of:

cmpw (%esi,%eax,4),%bx ; compare Ki with A

je swb3 ; Jump if match found
adcl %eax,%eax ; IF A>Ki THEN i := 2i
; ELSE i := 2i+1

The above code is a great improvement on any straightforward implementation of the

standard binary chop mechanism.

9.2.10 Miscellaneous
XCH Exchange A and B
ATB B := A
ATC C := A
BTC C := B
202 CHAPTER 9. THE DESIGN OF CINTCODE

These instructions are used move values between register A, B and C.

NOP

This instruction has no effect.

SYS

This instruction is used in body of the hand written library routine sys. If A is zero,
the interpreter returns with exception code P!4.
If A is -1 it sets register count to P!4, setting A to the previous value of count.
Changing the value of count may change which of the two interpreters is used. For
more details see Section 4.3.
Otherwise, it performs a system operation returning the result in A. In the C im-
plementation of the interpreter this is done by the following code:

c = dosys(p, g);

MDIV

This instruction is used as the one and only instruction in the body of the hand written
library routine muldiv, see Section 3.3. It divides P!5 into the double length product of
P!3 and P!4 placing the result in A and the remainder in the global variable result2.
It then performs a function return (RTN). Its effect is as follows:

A := <the result>
G!Gn_result2 := <the remainder>
PC := P!1 // PC := P!1
P := P!0 // P := P!0

CHGCO

This instruction is used in the implementation of coroutines. It is the one and only
instruction in the body of the hand written library routine changeco(val,cptr) where
val is passed in Cintcode register A and cptr is in P!4. Its effect, which is rather subtle,
is shown below. For more information see page 56.

G!Gn_currco!0 := P!0 // !currco := !P -- changeco’s old P pointer

PC := P!1 // PC := P!1 -- changeco’s return address
G!Gn_currco := P!4 // currco := cptr
P := P!4!0 // P := !cptr

BRK

This instruction is used by the debugger to implement break points. It causes the
interpreter to return with exception code 2.
9.2. THE CINTCODE INSTRUCTION SET 203

9.2.11 Floating-point Instructions

Floating-point operations other than those performed by SELST are provided by the
FLTOP instruction. They are as follows.

FLTOP 1 b A := floating point(A × 10b )

FLTOP 3 A := FLOAT A
FLTOP 4 A := FIX A
FLTOP 5 A := #ABS A
FLTOP 6 A := A #* B
FLTOP 7 A := A #/ B
FLTOP 8 A := A #+ B
FLTOP 9 A := A #- B
FLTOP 10 A := #+A
FLTOP 11 A := #-A
FLTOP 12 A := A #= B
FLTOP 13 A := A #~= B
FLTOP 14 A := A #< B
FLTOP 15 A := A #> B
FLTOP 16 A := A #<= B
FLTOP 17 A := A #>= B

In the above table, b is a signed byte representing a decimal exponent in the range -128
to +127. Floating point numbers with exponents outside this range can be generated
using sys(Sys flt, fl mk, x, e) as described on page 3.3.

9.2.12 Select Instructions

Access to fields and some op:= assignment are performed using the following instruc-
tions.

SELLD len sh A := SLCT len:sh:0 OF A

SELST 0 len sh SLCT len:sh:0 OF A := B
SELST op len sh SLCT len:sh:0 OF A op:= B

The mapping between op and its corresponding expression operator is given by the
table on page 183.

9.2.13 Undefined Instructions

There is now only one undefined instruction and it code is 0. It will cause the interpreter
to return with exception code of 1.
204 CHAPTER 9. THE DESIGN OF CINTCODE

9.2.14 Corruption of B
To improve the efficiency of some hand written machine code interpreters, the following
instructions are permitted to corrupt the value held in B:

K KH KW Kn KnG KnG1 KnGH

SWL SWB MDIV CHGCO

All other instructions either set B explicitly or leave its value unchanged.

9.2.15 Exceptions
When an exception occurs, the interpreter saves the Cintcode registers in its register
vector and yields the exception number as result. For exceptions caused by non existent
instructions, BRK, DIV or MOD the program counter is left pointing to the offend-
ing instruction. For more details see the description of sys(Sys interpret,...) on
page 80.
Chapter 10

The Design of Sial

Sial is an internal intermediate assembly language designed for BCPL. The first version
was called Cial (Compact Internal Assembly Language) was pronounced “seal”. It was
essentially an assembly language for Cintcode with the same function code mnemonics
and the same abstract machine registers. It was soon found that rather than having a
variety of codes to load an integer constant (such as L0, L1, L2, LM1, LW, LH or L), it was
better to have one function code to load positive integers and another for negative ones
with the values specified by operands. This form is more convenient for translation and
easier to compress. The new language is called Sial (also pronouced “seal”) with the S
standing for smaller. Sial therefore has fewer function codes than Cintcode and most
of them take operands but still uses the same abstract machine registers. Although
Cintcode load instructions save the value of the A register in B before setting A, Sial
loads typically do not. The current version of Sial has not yet been updated to deal
with the extended BCPL features such as floating point and op:= assignments.
As as example of the use of Sial, consider the program com/hello.b which is as
follows:

GET "libhdr"
LET start() = VALOF
{ writef("Hello*n")
RESULTIS 0
}

This can be translated into Sial using bcpl2sial com/hello.b to hello.sial. The
resulting file is:
F104
F113 K5 C115 C116 C97 C114 C116
F111 L1
F112 M9001
F32 P3 G94
F11 K0
F77
F107 M9001 K6 C72 C101 C108 C108 C111 C10
F106 K1 G1 L1 G94
F105

205
206 CHAPTER 10. THE DESIGN OF SIAL

This can be converted into something slightly more readable using the command:
sial-sasm hello.sial to * giving: This can be translated into Sial using the
bcpl2sial command as follows.

0.010> sial-sasm hello.sial to *

Converting hello.sial to *
MODSTART
//Entry to: start
ENTRY K5 C115 C116 C97 C114 C116
LAB L1
LSTR M9001
KPG P3 G94
L K0
RTN
STRING M9001 K6 C72 C101 C108 C108 C111 C10
GLOBAL K1
G1 L1
G94
MODEND
Conversion complete
0.000>

Alternatively, the Sial can be translated, statement by statement, into the assembly
language of a machine such as the Pentium as follows.

0.000> sial-386 hello.sial to hello.s

Converting hello.sial to hello.s
Conversion complete
0.010> type hello.s
# Code generated by sial-386
.text
.align 16
# MODSTART
# Entry to: start
# ENTRY K5 C115 C116 C97 C114 C116
# LAB L1
LA1:
movl %ebp,0(%edx)
movl %edx,%ebp
popl %edx
movl %edx,4(%ebp)
movl %eax,8(%ebp)
movl %ebx,12(%ebp)
# LSTR M9001
leal MA9001,%ebx
shrl $2,%ebx
# KPG P3 G94
movl 376(%esi),%eax
leal 12(%ebp),%edx
10.1. THE SIAL SPECIFICATION 207

call *%eax
# L K0
xorl %ebx,%ebx
# RTN
movl 4(%ebp),%eax
movl 0(%ebp),%ebp
jmp *%eax
# STRING M9001 K6 C72 C101 C108 C108 C111 C10
.data
.align 4
MA9001:
.byte 6
.byte 72
.byte 101
.byte 108
.byte 108
.byte 111
.byte 10
.text
# GLOBAL K1
.globl prog
.globl _prog
prog:
_prog:
movl 4(%esp),%eax
# G1 L1
movl $LA1,4(%eax)
# G94
ret
# MODEND
0.020>

Sial was designed as an experiment in the compact representation of algorithms

that can be just-in-time compiled easily into code for any target machine. Its sec-
ondary purpose was to allow an easy way to generate native code translations of BCPL
programs giving typically a ten fold speedup over the Cintcode interpretive version. An
experienced programmer can normally modify an existing Sial translator to generate
reasonable code for a new target in one or two days.
The following sections give a specification of Sial and an outline of how the trans-
lator sial-686 works.

10.1 The Sial Specification

Sial consists of a stream of directives and instructions each starting with an opcode
followed by operands. Both opcodes and operands and encoded using integers each
prefixed by a letter specifying what kind of value it represents. The prefixes are as
follows:
208 CHAPTER 10. THE DESIGN OF SIAL

F An opcode or directive
P A stack offset, 0 to #xFFFFFF
G A global variable number, 0 to 65535
K A 24-bit unsigned constant, often small in value
W A signed integer, used for static data and large constants
C A byte in range 0 to 255
L A label generated by translation phase
M A label generated by the Sial codegenerator
The instructions are for an abstract machine with the following internal registers.
a The main accumulator, function first arg and result register
b The second accumulator used in dyadic operations
c Register used by pbyt and xpbyt, and possibly currupted by
some other instructions, such as mul, div, rem, xdiv and xrem
P Pointer to the base of the current stack frame
G Pointer to the base of the Global Vector
PC Set by jump and call instrunctions
The opcodes and directives are as follows:
Mnemonic Operand(s) Meaning

lp Pn a := P!n
lg Gn a := G!n
ll Ln a := !Ln
llp Pn a := @ P!n
llg Gn a := @ G!n
lll Ln a := @ !Ln
lf Ln a := address of entry point Ln
l Kn a := n
lm Kn a := - n
sp Pn P!n := a
sg Gn G!n := a
sl Ln !Ln := a
ap Pn a := a + P!n
ag Gn a := a + G!n
a Kn a := a + n
s Kn a := a - n
10.1. THE SIAL SPECIFICATION 209

lkp Kk Pn a := P!n!k
lkg Kk Gn a := G!n!k
rv a := ! a
rvp Pn a := P!n!a
rvk Kn a := a!k
st !a := b
stp Pn P!n!a := b
stk Kn a!n := b
stkp Kk Pn P!n!k := a
skg Kk Gn G!n!k := a
xst !b := a
k Pn Call a(b,...) incrementing P by n
leaving b in a
kpg Pn Gg Call Gg(a,...) incrementing P by n
neg a := - a
not a := ~ a
abs a := ABS a
xdiv a := a / b; c := ?
xmod a := a MOD b; c := ?
xsub a := a - b; c := ?
mul a := b * a; c := ?
div a := b / a; c := ?
mod a := b MOD a; c := ?
add a := b + a
sub a := b - a
eq a := b = a
ne a := b ~= a
ls a := b < a
gr a := b > a
le a := b <= a
ge a := b >= a
eq0 a := a = 0
ne0 a := a ~= 0
ls0 a := a < 0
gr0 a := a > 0
le0 a := a <= 0
ge0 a := a >= 0
210 CHAPTER 10. THE DESIGN OF SIAL

lsh a := b << a
rsh a := b >> a
and a := b & a
or a := b | a
xor a := b XOR a
eqv a := b EQV a
gbyt a := b % a
xgbyt a := a % b
pbyt b % a := c
xpbyt a % b := c
swb Kn Ld K1 L1 ... Kn Ln Binary chop switch, Ld default
swl Kn Ld L1 ... Ln Label vector switch, Ld default
xch Swap a and b
atb b := a
atc c := a
bta a := b
btc c := b
atblp Pn b := a; a := P!n
atblg Gn b := a; a := G!n
atbl Kk b := a; a := k
j Ln Jump to Ln
rtn Function or routine return
goto PC := a
ikp Kk Pn a := P!n + k; P!n := a
ikg Kk Gn a := G!n + k; G!n := a
ikl Kk Ln a := !Ln + k; !Ln := a
ip Pn a := P!n + a; P!n := a
ig Gn a := G!n + a; G!n := a
il Ln a := !Ln + a; !Ln := a
jeq Ln Jump to Ln if b = a
jne Ln Jump to Ln if b ~= a
jls Ln Jump to Ln if b < a
jgr Ln Jump to Ln if b > a
jle Ln Jump to Ln if b <= a
jge Ln Jump to Ln if b >= a
jeq0 Ln Jump to Ln if a = 0
jne0 Ln Jump to Ln if a ~= 0
jls0 Ln Jump to Ln if a < 0
jgr0 Ln Jump to Ln if a > 0
jle0 Ln Jump to Ln if a <= 0
jge0 Ln Jump to Ln if a >= 0
jge0m Mn Jump to Mn if a >= 0
10.1. THE SIAL SPECIFICATION 211

brk Breakpoint instruction

nop No operation
chgco Change coroutine
mdiv a := muldiv(P!3, P!4, P!5)
sys System function
section Kn C1 ... Cn Name of section
modstart Start of module
modend End of module
global Kn G1 L1 ... Gn Ln Global initialisation data
string Ml Kn C1 ... Cn String constant
const Mn Ww Large integer constant
static Ln Kk W1 ... Wk Static variable or table
mlab Mn Destination of jge0m
lab Lm Program label
lstr Mn a := Mn (pointer to string)
entry Kn C1 ... Cn Start of a function
The following Sial operators were added in August 2014 to allow native code com-
pilation of the floating point code. All floating point operators may corrupt global 11
(tempval).
float a := FLOAT a; b := ?
fix a := FIX a; b := ?
fabs a := #ABS a; b := ?
fneg a := #- a; b := ?
fmul a := b #* a; b := ?
fdiv a := b #/ a; b := ?
fmod a := b #MOD a; b := ?
fadd a := b #+ a; b := ?
fsub a := b #- a; b := ?
feq a := b #= a; b := ?
fne a := b #~= a; b := ?
fls a := b #< a; b := ?
fgr a := b #> a; b := ?
fle a := b #<= a; b := ?
fge a := b #>= a; b := ?
feq0 a := a #= 0; b := ?
fne0 a := a #~= 0; b := ?
fls0 a := a #< 0; b := ?
fgr0 a := a #> 0; b := ?
fle0 a := a #<= 0; b := ?
fge0 a := a #>= 0; b := ?
The floating point conditional jump instructions are as follows.
212 CHAPTER 10. THE DESIGN OF SIAL

jfeq Ln Jump to Ln if b #= a; b := ?
jfne Ln Jump to Ln if b #~= a; b := ?
jfls Ln Jump to Ln if b #< a; b := ?
jfgr Ln Jump to Ln if b #> a; b := ?
jfle Ln Jump to Ln if b #<= a; b := ?
jfge Ln Jump to Ln if b #>= a; b := ?
jfeq0 Ln Jump to Ln if a #= 0; b := ?
jfne0 Ln Jump to Ln if a #~= 0; b := ?
jfls0 Ln Jump to Ln if a #< 0; b := ?
jfgr0 Ln Jump to Ln if a #> 0; b := ?
jfle0 Ln Jump to Ln if a #<= 0; b := ?
jfge0 Ln Jump to Ln if a #>= 0; b := ?
Notice that all floating point instructions currently leave register b undefined, but
this may be changed later. They may also may corrupt global 11 (tempval).
A second example of the use of Sial is the following program (com/fact.b):

SECTION "fact"
GET "libhdr"
LET start() = VALOF
{ FOR i = 1 TO 5 DO writef("fact(%n) = %i4*n", i, fact(i))
RESULTIS 0
}
AND fact(n) = n=0 -> 1, n*fact(n-1)

It translation in Sial code is as follows:

F104
F103 K4 C102 C97 C99 C116
F113 K5 C115 C116 C97 C114 C116
F111 L1
F11 K1
F13 P3
F111 L4
F3 P3
F69
F9 L2
F31 P9
F13 P9
F3 P3
F13 P8
F112 M1
F32 P4 G94
F79 K1 P3
F75 K5
F89 L4
F11 K0
F77
F107 M1 K15 C102 C97 C99 C116 C40 C37 C110
10.1. THE SIAL SPECIFICATION 213

C41 C32 C61 C32 C37 C105 C52 C10

F113 K4 C102 C97 C99 C116
F111 L2
F92 L5
F11 K1
F77
F111 L5
F12 K1
F16 P3
F69
F9 L2
F31 P4
F73 P3
F39
F77
F106 K1 G1 L1 G94
F105

Using the sial-sasm command we obtain the following more readable version:

MODSTART
SECTION K4 C102 C97 C99 C116
//Entry to: start
ENTRY K5 C115 C116 C97 C114 C116
LAB L1
L K1
SP P3
LAB L4
LP P3
ATB
LF L2
K P9
SP P9
LP P3
SP P8
LSTR M1
KPG P4 G94
IKP K1 P3
ATBL K5
JLE L4
L K0
RTN
STRING M1 K15 C102 C97 C99 C116 C40 C37 C110 C41 C32
C61 C32 C37 C105 C52 C10
//Entry to: fact
ENTRY K4 C102 C97 C99 C116
LAB L2
JNE0 L5
L K1
RTN
LAB L5
LM K1
AP P3
ATB
214 CHAPTER 10. THE DESIGN OF SIAL

LF L2
K P4
ATBLP P3
MUL
RTN
GLOBAL K1
G1 L1
G94
MODEND

This can be translated into assembly language using the program com/sial-686.b
which is a simple program based on sial-sasm.b. This version can now compile the
floating point instructions recently added to Sial. It generates the readable version
of the Sial source as comments interspersed with the corresponding Pentium assembly
code. For the example program given above, it outputs the following assembly language.

# Code generated by sial-686

.text
.align 16
# MODSTART
# SECTION K4 C102 C97 C99 C116
# Entry to: start
# ENTRY K5 C115 C116 C97 C114 C116
# LAB L1
LA1:
movl %ebp,0(%edx)
movl %edx,%ebp
popl %edx
movl %edx,4(%ebp)
movl %eax,8(%ebp)
movl %ebx,12(%ebp)
# L K1
movl $1,%ebx
# SP P3
movl %ebx,12(%ebp)
# LAB L4
LA4:
# LP P3
movl 12(%ebp),%ebx
# ATB
movl %ebx,%ecx
# LF L2
leal LA2,%ebx
# K P9
movl %ebx,%eax
movl %ecx,%ebx
leal 36(%ebp),%edx
call *%eax
# SP P9
movl %ebx,36(%ebp)
# LP P3
movl 12(%ebp),%ebx
10.1. THE SIAL SPECIFICATION 215

# SP P8
movl %ebx,32(%ebp)
# LSTR M1
leal MA1,%ebx
shrl $2,%ebx
# KPG P4 G94
movl 376(%esi),%eax
leal 16(%ebp),%edx
call *%eax
# IKP K1 P3
movl 12(%ebp),%ebx
incl %ebx
movl %ebx,12(%ebp)
# ATBL K5
movl %ebx,%ecx
movl $5,%ebx
# JLE L4
cmpl %ebx,%ecx
jle LA4
# L K0
xorl %ebx,%ebx
# RTN
movl 4(%ebp),%eax
movl 0(%ebp),%ebp
jmp *%eax
# STRING M1 K15 C102 C97 C99 C116 C40 C37 C110 C41 C32
# C61 C32 C37 C105 C52 C10
.data
.align 4
MA1:
.byte 15
.byte 102
.byte 97
.byte 99
.byte 116
.byte 40
.byte 37
.byte 110
.byte 41
.byte 32
.byte 61
.byte 32
.byte 37
.byte 105
.byte 52
.byte 10
.text
# Entry to: fact
# ENTRY K4 C102 C97 C99 C116
# LAB L2
LA2:
movl %ebp,0(%edx)
movl %edx,%ebp
popl %edx
216 CHAPTER 10. THE DESIGN OF SIAL

movl %edx,4(%ebp)
movl %eax,8(%ebp)
movl %ebx,12(%ebp)
# JNE0 L5
orl %ebx,%ebx
jne LA5
# L K1
movl $1,%ebx
# RTN
movl 4(%ebp),%eax
movl 0(%ebp),%ebp
jmp *%eax
# LAB L5
LA5:
# LM K1
movl $-1,%ebx
# AP P3
addl 12(%ebp),%ebx
# ATB
movl %ebx,%ecx
# LF L2
leal LA2,%ebx
# K P4
movl %ebx,%eax
movl %ecx,%ebx
leal 16(%ebp),%edx
call *%eax
# ATBLP P3
movl %ebx,%ecx
movl 12(%ebp),%ebx
# MUL
movl %ecx,%eax
imul %ebx
movl %eax,%ebx
# RTN
movl 4(%ebp),%eax
movl 0(%ebp),%ebp
jmp *%eax
# GLOBAL K1
.globl fact
.globl _fact
fact:
_fact:
movl 4(%esp),%eax
# G1 L1
movl $LA1,4(%eax)
# G94
ret
# MODEND

When implementing sial-686 it was necessary to decide how the Intel registers
were to be used and what the BCPL calling sequence should be. The chosen register
allocation was as follows:
10.1. THE SIAL SPECIFICATION 217

Intel register Use

%eax A work register

%ebx The A register
%ecx The B register
%edx The C register
%esi The G pointer
%edi A work register
%ebp The P pointer
%st(0) The X register used in compilation of floating point operations
the
%sp(0) The S register used point numbers between X and A or B
when transferring float-
ing
The chosen BCPL calling sequence is as follows:

# Entry address must be in %eax

# The first argument must be in %ebx
leal <stack increment>(%ebp),%edx # Set %edx to the new P pointer
call *%eax # Subroutine jump to the entry point

The entry sequence is as follows:

# The first argument is in %ebx(=A)

# The new P pointer is in %edx(=C)
movl %ebp,0(%edx) # C!0 := P
movl %edx,%ebp # P := C
popl %edx # Get the return address
movl %edx,4(%ebp) # P!1 := return address
movl %eax,8(%ebp) # P!2 := entry address
movl %ebx,12(%ebp) # P!3 := the first argument

The return sequence is as follows:

# The result is in %ebx(=A)

movl 4(%ebp),%eax # Get the return address
movl 0(%ebp),%ebp # P := the saved P pointer
jmp *%eax # Jump to the return address

The structure of sial-686 is simple. It mainly consists of a large switch within the
function scan that has a case for each Sial function code and directive. For example,
the case for the function code kpg is approximately as follows:

CASE f_kpg: cvfpg("KPG") // Call Gg(a,...) incrementing P by n

writef("*n movl %n(%%esi),%%eax", 4*gval)
writef("*n leal %n(%%ebp),%%edx", 4*pval)
writef("*n call **%%eax")
ENDCASE
218 CHAPTER 10. THE DESIGN OF SIAL

The call cvfpg("KPG") reads the Sial statement knowing it is of the form: KPG
Pk Gn. This outputs the statement as an assembly language comment after placing k
and n in pval and gval, respectively. The three writef calls then output the three
assembly language instructions for the KPG operation, and ENDCASE transfers control to
where the next Sial statement is processed. All the other cases are equally simple.
To improve the efficiency of the floating point code, instructions that normally load
a value into A or B are delay until it is known how the value is to be used. If a floating
point operation is about to be performed it is better to load the value into X. Where
possible, sial-686.b remembers what value (such as which local or global) is currently
held in A, B, X and S.
The section name of the program, which must be present, compiles into a C callable
function that initialises the BCPL global vector with the entry points defined within
this module. To complete the 686 implementation, there is a short handwritten assem-
bly language library natbcpl/sysasm/mlib.s that defines the BCPL callable func-
tions sys, changeco and muldiv. The program must be linked the compiled ver-
sions of the BCPL library modules BLIB and DLIB, and also clib whose source is in
natbcpl/sysc/clib.c.
Every section must contain the definition of a function to initialise the global vector
with the entry points of functions defined in the section. For a section defined in
BCPL, the name of the initialisation function is the section name which must have been
specified in the BCPL source. A C program such as initprog.c must be provided to
with a definition of a function called initsections that calls the initialisation function
of every section of the program. The command makeinit, described on page 143, can
be used to create the initialiation program. For the program prog.b given above, the
following command:

makeinit prog.b to initprog.c

will create the file initprog.c which is as follows:

// Initialisation file written by makeinit version 2.0

#include "bcpl.h"
WORD stackupb=50000;
WORD gvecupb=1000;
// BCPL sections
extern BLIB(WORD *); // file (run-time library)
extern DLIB(WORD *); // file (system dependent library)
extern prog(WORD *); // file prog.b
void initsections(WORD *g) {
BLIB(g); // file (run-time library)
DLIB(g); // file (system dependent library)
prog(g); // file prog.b
return;
}
10.2. COMPACTION OF SIAL 219

If needed, the runtime stack size and the size of the global vector can be specified
by arguments to makeinit. The compilation of initprog must be linked in with the
object code of all the other sections needed by prog.b when building its executable.
Assuming the executable is placed in bin/prog, it can be executed by the bash shell
command ./bin/prog or possibly just prog if the PATH environment variable is suitably
set.

10.2 Compaction of Sial

In order to transmit program to a device such as a mobile phone or space probe over a
slow connection it is useful to have a compact representation of the code. Sial is both
target machine independent and can be compacted with ease. This section gives a brief
overview of an experimental compaction technique that seems to performs well.
Since the types of operands and their number depend only on the Sial operator,
an Sial stream can be split into several streams of which the main one is the stream
of Sial operators. Others are streams holding global variable numbers, local variable
offsets, program label numbers, data labels, integer constants, character codes and a
some others. These streams can be separately compressed taking advantage of the
special properties of each. Some ideas are given below.
Local variable offsets have a very skew distribution and so are susceptible to Huff-
man (or possibly arithmetic) coding after some preprocessing to deal with large values
and the implementation of a mechanism to take advantage of the observation that, if
an offset is used once, the same offset is likely to be used again in the near future. This
might suggest the use of move-to-front buffering.
Program labels have the property that, in any section, they are each only set
once using a LAB or ENTRY statement. If they are systematically renumbered so that
successive label setting statements take successive label numbers, there is no need for
these statements to take a label argument. The remaining labels in the stream are
typically nearly monotonic the compaction algorithm can take advantage of this.
The operation code stream often contains repeated patterns that are susceptible to
the conventional techniques used to compress text, and the same applies to the stream
of characters. It might be worth separating out the integers representing the character
string lengths from other integers and place them either in a stream of their own or
insert them into the stream of characters.
Some preliminary experiments on Sial compression can be found in the directory
bcplprogs/sial in the standard BCPL distribution.
220 CHAPTER 10. THE DESIGN OF SIAL
Chapter 11

The MC Package

This chapter describes the MC package which provides a machine independent way to
generate and execute native machine code at runtime. The work on this package started
in January 2008 and is still under development, however, it currently works well enough
to run the n-queens problem on i386 machines about 24 times faster than the normal
Cintcode interpretive version. MC package development is performed in the directory
BCPL/bcplprogs/mc/ and fairly stable versions are copied to BCPL/cintcode/g/mc.h,
BCPL/cintcode/com/mci386.b and BCPL/cintcode/cin/mci386 which can be used
from any working directory. Currently the MC package does not have any floating
point operations. This will be rectified in due course.
The package is based on a simple machine independent abstract machine code
called MC which is easily translated into machine instructions for most architectures.
Although native code is generated by MC calls such as mcRDX(mc add, mc b, 20,
mc d), MC has a corresponding assembly language to assist debugging. The assembly
form of the instruction generated by the previous call is ADD B,20(D) meaning set
register B to the sum of B and the contents of the memory location whose address is 20
plus the value of register D. MC instructions are fairly low level and typically translate
into single native code instructions for most architectures. This example translates into
the i386 GNU statement: addl 20(%edx),%ebx.
The first operand is the destination for any instruction that updates a regis-
ter or memory location. Thus assignments are always from right to left as in
most programming languages but unlike many assembly codes where, for instance,
movl 20(%edx),%ebx updates the second operand.
The MC machine has six registers A, B, C, D, E and F that are directly available
to the programmer, and also a program counter, stack pointer, stack frame pointer and
a condition code register, although these cannot be accessed explicitly.

11.1 MC Example
The following program is a simple demonstration of the i386 version of the MC package.

GET "libhdr"
GET "mc.h"

221
222 CHAPTER 11. THE MC PACKAGE

MANIFEST {
A=mc_a; B=mc_b; C=mc_c; D=mc_d; E=mc_e; F=mc_f
a1=1; a2; a3
}
LET start() = VALOF
{ // Load the dynamic code generation package for i386 machines.
LET mcseg, mcb, n = globin(loadseg("mci386")), 0, 0
UNLESS mcseg DO
{ writef("Trouble with MC package: mci386*n")
GOTO fin
}
// Create an MC instance for 10 functions with a data space
// of 100 words and code space of 4000 words.
mcb := mcInit(10, 100, 4000)
UNLESS mcb DO
{ writef("Unable to create an mci386 instance*n")
GOTO fin
}
mc := 0 // Currently no selected MC instance.
mcSelect(mcb) // Select the new MC instance.
mcK(mc_debug, #b0011) // Trace comments and MC instructions.
mcKKK(mc_entry, 1, 3, 5) // Entry point for function 1
// having 3 arguments and 5 local variables
mcK(mc_debug, #b1111) // Trace comments, MC instructions, target
// instructions and the compiled code.
mcRA(mc_mv, A, a1) // A := <arg 1>
mcRA(mc_add, A, a2) // A := A + <arg 2>
n := mcNextlab()
mcL(mc_lab, n) // Ln:
mcRA(mc_add, A, a3) // A := A + <arg 3>
mcR(mc_dec, A) // A := A - 1
mcRK(mc_cmp, A, 100)
mcJS(mc_jlt, n) // IF A<100 JMP Ln
mcK(mc_debug, #b0011) // Trace only comments and MC instructions.
mcF(mc_rtn) // Return from function 1 with result in A.
mcF(mc_endfn) // End of function 1 code.
mcF(mc_end) // End of dynamic code generation.
writef("*nF1(10, 20, 30) => %n*n", mcCall(1, 10, 20, 30))
fin:
IF mcseg DO unloadseg(mcseg)
RESULTIS 0
}

When this program runs it outputs the following.

// ENTRY 1 3 5
// DEBUG 15
11.1. MC EXAMPLE 223

// MV A,A1
movl 20(%ebp), %eax
573: 8B 45 14
// ADD A,A2
addl 24(%ebp), %eax
576: 03 45 18
// LAB L1
lab L1
579: L1:
// ADD A,A3
addl 28(%ebp), %eax
579: 03 45 1C
// DEC A
decl %eax
582: 48
// CMP A,$100
cmpl $100, %eax
583: 83 F8 64
// JLT L1
jl L1
586: 7C F7
// DEBUG 3
// RTN
// ENDFN
// END
F1(10, 20, 30) => 117

The result of 117 (= 10+20+(30-1)*3) shows that the body of the loop was correctly
executed three times.
The header file (mc.h) defines manifests (such as mc mv and mc add) and globals
(such as mcK and mcRA) provided by the package. The package itself must be dynami-
cally loaded (by globin(loadseg("mci386"))) and then selected (by mcSelect(mcb)).
MC instructions are compiled by calls such as mcRA(op,... or mcRK(op,... where op
specifies the instruction or directive and the letters following mc (eg RA or RK) specify
the sort of operands supplied.
A register operand is denoted by R and an integer operand by K. There are 9 possible
kinds of memory operands denoted by A, V, G, M, L, D, DX, DXs and DXsB. A denotes an
specified argument of the current function, V denotes a specified local variable of the
current function, G denotes a specified BCPL global variable, M denotes a location in
Cintcode memory specified by a BCPL pointer, L denotes the position within the data
or code areas of the compiled code corresponding to a given label, D denotes a specified
absolute machine address, DX denotes a location whose machine address is the sum of
a given byte offset and register, DXs is similar to DX only the index register is scaled
by a given factor of 1, 2, 4 or 8 and finally DXsB is like DXs but has a second specified
register added into the effective address.
The following table summarises the MC code generation functions. The first ar-
gument is always specifies the directive or instruction and the remaining arguments
specify the operands. The destination of any instruction that updates a register or
memory location is always the first operand.
224 CHAPTER 11. THE MC PACKAGE

Function Operands
mcF No operand
mcK One integer operand
mcR One MC register operand
mcA One operand specifying an argument number
mcV One operand specifying an local variable number
mcG One operand specifying a global variable number
mcM One operand giving the word address of a location in Cintcode mem-
ory
mcL One numeric label operand, defaulting to 32-bit relative
mcD One operand giving an absolute machine address
mcDX One memory operand specified by an offset added to an index register
mcDXs One memory operand specified by an offset added to an index register
scaled by s which must be 1, 2, 4 or 8
mcDXsB One memory operand specified by an offset added to a base register
and an index register scaled by s which must be 1, 2, 4 or 8
mcJS Jump instructions with near relative destinations
mcJL Jump instructions with possibly distant relative destinations
mcJR Jump instructions with destination given by resister
mcRA Two operands, R and A
mcRV Two operands, R and V
mcRG Two operands, R and G
mcRM Two operands, R and M
mcRL Two operands, R and L
mcRD Two operands, R and D
mcRDX Two operands, R and DX
mcRDXs Two operands, R and DXs
mcRDXsB Two operands, R and DXsB
mcRR Two operands, R and R
mcAR Two operands, A and R
mcVR Two operands, V and R
mcGR Two operands, G and R
mcMR Two operands, M and R
mcLR Two operands, L and R
mcDR Two operands, D and R
mcDXR Two operands, DX and R
mcDXsR Two operands, DXs and R
mcDXsBR Two operands, DXsB and R
11.2. MC LIBRARY FUNCTIONS 225

mcRK Two operands, R and K

mcAK Two operands, A and K
mcVK Two operands, V and K
mcGK Two operands, G and K
mcMK Two operands, M and K
mcLK Two operands, L and K
mcDK Two operands, D and K
mcDXK Two operands, DX and K
mcDXsK Two operands, DXs and K
mcDXsBK Two operands, DXsB and K
mcKK Two integer operands
mcKKK Three integer operands
mcPRF One printf format string and one register

11.2 MC Library Functions

mcb := mcInit(maxfno, dsize, csize)

Create an instance of the MC package, allocating space for maxfno functions, dsize
words of data space and csize words of code space. The MC control block is assigned
to mcb.

mcSelect(mcb)
Select an instance of the MC package by assigning mcb to the global variable mc.
For efficiency reasons, mcSelect copies various field in the control block to global
variables. If mc was non zero, the previous setting of the globals are saved in the
previously selected MC instance. It is thus important to set mc to zero before the first
call od mcSelect.

res := mcCall(fno, a1, a2, a3)

Call the function with number fno giving it the three arguments a1, a2, a3. The
result is assigned to res. Function fno must have been defined to expect three argu-
ments.

mcClose()
Close the currently selected MC instance deleting all its workspace and compiled
code. It also sets mc to zero.

mcPRF(mess, reg)
This function is an invaluable debugging aid which compiles code to call the C
function printf with the given format string (packed in the data area) and the value
of the specified register. All registers, including the condition code, are preserved. The
register argument may be omitted if the format string requires no register argument.
Typical use of mcPRF is as follows:

mcRK(mc_mv, D, #x01234567)
mcRK(mc_mv, A, #x89ABCDEF)
226 CHAPTER 11. THE MC PACKAGE

mcRK(mc_mv, A, #x10000000)
mcPRF("With D=%8x ", D)
mcPRF("A=%8x ", A)
mcPRF("B=%8x*n", B)
mcR(mc_div, B)
mcPRF("the instruction: DIV B*n")
mcPRF("gives D=%8x ", D)
mcPRF("A=%8x ", A)
mcPRF("B=%8x*n", B)

This causes the following output:

With D= 1234567 A=89abcdef B=10000000

the instruction: DIV B
gives D= 9abcdef A=12345678 B=10000000

n := mcNextlab()
Allocate the next available label assigning its number to n. Labels are use by
instructions that refer to static data and in jump instructions. There is essentially no
limit to the number of labels that may be allocated.

mcComment(format, a, b,..., k)
This is a debugging aid to make the compiled code more readable using writef to
write a message to the listing output during code generation if the least significant bit
of mcDebug is a one. The variable mcDebug is set by the DEBUG directive described
below.

res := mcDatap()
res := mcCodep()
These calls return the current positions in the data and code area respectively.

All the other functions compile MC directives and instructions and are described
below.

11.3 The MC Language

The MC abstract machine language is fairly low level and is somewhat influenced by
the i386 architecture. Particularly the rather small number of MC registers allowed, the
rich variety of memory addressing modes and the specification of the instructions for
multiplication, division and shifts. However, it is machine independent and reasonably
easy to compile into native machine code for most machines. Before describing the
MC instructions, a few key features will be introduced. As mentioned earlier the
MC machine has six registers named A to F which are typically mapped directly onto
machine registers of the target architecture. These can be used for any purpose except
for a few instructions such as MUL, DIV and the shifts which may implicitly use some of
them implicitly.
11.3. THE MC LANGUAGE 227

When an MC function is declared it has a specified number of arguments and local

variables (see the ENTRY statement below). When a function is called by the CALL
instruction, the required number of arguments must have already been pushed onto the
stack. On return these arguments will have been automatically popped from the stack.
If the wrong number of arguments are given, the effect is undefined. By convention,
the result of a function is returned in register A.
Numeric labels are used to refer to static data and positions in the code. They
are allocated by calls of mcNextlab, described above. Many architectures allow both
conditional and unconditional jumps to use short offsets (typically single bytes) to
specify the relative address of the destination. Jump instructions automatically use
short relative addresses for backward jumps if possible, but, for forward jumps, the
programmer is required to give a hint. Jump instructions compiled by mcJS expect
forward jumps to use short relative addresses while mcJL specifies that larger relative
addresses are to be used. If a short relative address proves insufficent and error message
is generated telling the programmer that mcJL should have been used. The function
mcJR is used when the destination address of a jump instruction is in a register.
Conditional jump instructions inspect the condition code to determine whether or
not to jump. The condition code is set by the CMP, ADD, ADDC, SUB and SUBC instructions
and preserved by jump instructions (JMP and Jcc). All other instructions (including
INC and DEC leave the condition code undefined.
All MC directives and instructions are described below in alphabetical order. The
name of the operation is given in bold caplital letters together with the list of possible
operand types. The BCPL manifest for the operation consists of the name in lower
case letters preceeded by mc . For example, mc add is the manifest constant for the
ADD operation, and since RDXs appears in its list of operand types, it can be compiled
by, for instance, mcRDXs(mc add, mc a, 20, mc d, 4).

ADD RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Add the second operand into the first and set the condition code appropriately. For
example, mcRG(mc add, mc d, 150) will compile code to add global 150 in register D.

ADDC RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Add the condition code carry bit and the second operand into the first and set the
condition code appropriately. Adding 1 into the 64-bit value held in B:A can be done
by the code generated by:

mcRK( mc_add, mc_a, 1) // Don’t use INC here!

mcRK( mc_addc, mc_b, 0)

ALIGNC K
Align the next instruction to an address which is a multiple of k which must be 2,
4 or 8.
228 CHAPTER 11. THE MC PACKAGE

ALIGND K
Align the next item of data to an address which is a multiple of k which must be
2, 4 or 8.

AND RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Perform the bit wise AND of the second operand into the first.

CALL KK
Call the function who number is the first argument with n arguments that have
already been pushed onto the stack when n is the second operand. On return these
arguments will have been popped and, by convention, the result will be in register A.

CDQ F
Sign extend register A into D. That is, if A is positive set D to zero, otherwise it is
to #xFFFFFFFF. This is normally used in conjuction with DIV.

CMP RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Set the condition code to difference between the first operand and the second. The
condition code is used by conditional jumps and conditional setting instructions. For
example,

mcRK(mc_cmp, mc_b, 100)

mcJL(mc_jle, 25)

will compile code to jump the label L25 is B<=100, using signed arithmetic.

DATAB K
Assemble one byte of data with the specified value.

DATAK K
Assemble one aligned word of data with the specified value.

DATAL L
Assemble one aligned word of data initialised with the absolute address of code or
data specified by the given label.

DEBUG K
Set the debug tracing level (mcDebug) to the specified value. The least significant
four bits of mcDebug control the level of tracing as follows.

#b0001 Output any mcComment comments.

#b0010 Output the MC instructions.
#b0100 Output the target machine instructions.
#b1000 Output the compiled binary code.
11.3. THE MC LANGUAGE 229

DEC R A V G M L D DX DXs DXsB

Decrement the specified register or memory word by 1, leaving the condition code
undefined.

DIV K R A V G M L D DX DXs DXsB

Divide the double length value in D:A by the specified operand. The result is left
in A and the remainder in D. The DIV instruction performs signed arithmetic.

DLAB L
Set the specified label to the absolute address of the next available byte in the data
area.

ENDFN F
This marks the end of the body of the current function.

END F
This directive specifies that no more code generation will be done. The system
will free all temporary work space only preseving the MC control block, the function
dispatch table, and the data and code areas.

ENTRY KKK
This specifies the entry point of the function whose number is given by the first
operand. The second operand specifies how many arguments the function takes and the
third specified how many local variables the function may use. Calls to this function
must have the required number of arguments pushed onto the stack, and on return
this number of values will be automatically popped from the stack. Functions called
directly from BCPL using mcCall always take three arguments, but functions called
using the CALL instruction can take any number of arguments.

INC R A V G M L D DX DXs DXsB

Increment the specified register or word of memory by one, leaving the condition
code undefined.

JEQ JS JL JR
Jump to the specified location if the first operand of a previous CMP instruction was
equal to its second operand.

JGE JS JL JR
Jump to the specified location if the first operand of a previous CMP instruction was
greater than or equal to its second operand using signed arithmetic.

JGT JS JL JR
Jump to the specified location if the first operand of a previous CMP instruction was
greater than its second operand using signed arithemetic.

JLE JS JL JR
Jump to the specified location if the first operand of a previous CMP instruction was
less than or equal to its second operand using signed arithmetic.
230 CHAPTER 11. THE MC PACKAGE

JLT JS JL JR
Jump to the specified location if the first operand of a previous CMP instruction was
less than its second operand using signed arithmetic.

JMP JS JL JR
Unconditionally jump to the specified location.

JNE JS JL JR
Jump to the specified location if the first operand of a previous CMP instruction was
not equal to its second operand.

LAB L
Set the specified label to the machine address of the current position in the code
area.

MV RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Move the second operand into the first.

MVB AR VR GR MR LR DR DXR DXsR DXsBR

AK VK GK MK LK DK DXK DXsK DXsBK
Move the least significant byte of the second operand into the memory byte location
specified by the first.

MVH AR VR GR MR LR DR DXR DXsR DXsBR

AK VK GK MK LK DK DXK DXsK DXsBK
Move the least significant 16 bits of the second operand into the 16-bit memory
location specified by the first.

MVSXB RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Move the sign extended byte value specified by the second operand into the first.

MVSXH RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Move the sign extended 16-bit value specified by the second operand into the first.

MVZXB RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Move the zero extended byte value specified by the second operand into the first.

MVZXH RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
11.3. THE MC LANGUAGE 231

RK AK VK GK MK LK DK DXK DXsK DXsBK

Move the zero extended 16-bit value specified by the second operand into the first.

LEA RA RV RG RM RL RD RDX RDXs RDXsB

Load the register specified by the first operand with the absolute address of the
memory location specified by the second operand.

LSH RK RR
Shift to the left the value in the register specified by the first operand by the
amount specified by the second operand. If the second operand is a register is must be
C. Vacated positions are filled with zeros. The effect is undefined if the shift distance
is not in the range 0 to 31.

MUL K R A V G M L D DX DXs DXsB

Multiply register A by the operand placing the double length result in D:A. Signed
arithmetic is used. Unsigned arithmetic is used. Immediate (K) operands may some-
times be packed in the data area.

NEG R A V G M L D DX DXs DXsB

Negate the value specified by the operand.

NOP F
Performs no operation.

NOT R A V G M L D DX DXs DXsB

Perform the bitwise complement of the value specified by the operand.

OR RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Perform the bitwise OR of the second operand into the first.

POP R A V G M L D DX DXs DXsB

Pop one word off the stack placing it in the specified register or memory location.

PUSH K R A V G M L D DX DXs DXsB

Push the specified constant, register or memory location onto the stack.

RSH RR RK
Shift to the right the value in the register specified by the first operand by the amount
specified by the second operand. If the second operand is a register is must be C.
Vacated positions are filled with zeros. The effect is undefined if the shift distance is
not in the range 0 to 31.

RTN F
This causes a return from the current function. The result, if any, should be in A.
232 CHAPTER 11. THE MC PACKAGE

SEQ R
Set the specified register to one if the first operand of a previous CMP instruction
was equal to its second operand, otherwise set it to zero.

SGE R
Set the specified register to one if the first operand of a previous CMP instruction
was greater than or equal to its second operand using signed arithmetic, otherwise set
it to zero.

SGT R
Set the specified register to one if the first operand of a previous CMP instruction
was greater than its second operand using signed arithmetic, otherwise set it to zero.

SLE R
Set the specified register to one if the first operand of a previous CMP instruction
was less than or equal to its second operand using signed arithmetic, otherwise set it
to zero.

SLT R
Set the specified register to one if the first operand of a previous CMP instruction
was less than its second operand using signed arithmetic, otherwise set it to zero.

SNE R
Set the specified register to one if the first operand of a previous CMP instruction
was not equal to its second operand, otherwise set it to zero.

SUB RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Subtract the second operand from the first, and set the condition code appropri-
ately.

SUBC RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Subtract the condition code carry bit and the second operand from the first, and
set the condition code appropriately. Subtracting 1 from the 64-bit value held in B:A
can be done by the code generated by:

mcRK( mc_sub, mc_a, 1) // Don’t use DEC here!!

mcRK( mc_subc, mc_b, 0)

UDIV K R A V G M L D DX DXs DXsB

Divide the double length value in D:A by the specified operand. The result is left
in A and the remainder in D. The UDIV instruction performs unsigned arithmetic.
11.3. THE MC LANGUAGE 233

UJGE JS JL JR
Jump to the specified location if the first operand of a previous CMP instruction was
greater than or equal to its second operand using unsigned arithmetic.

UJGT JS JL JR
Jump to the specified location if the first operand of a previous CMP instruction was
greater than its second operand using unsigned arithmetic.

UJLE JS JL JR
Jump to the specified location if the first operand of a previous CMP instruction was
less than or equal to its second operand using unsigned arithmetic.

UJLT JS JL JR
Jump to the specified location if the first operand of a previous CMP instruction was
less than its second operand using unsigned arithmetic.

UMUL K R A V G M L D DX DXs DXsB

Multiply register A by the operand placing the double length result in D:A. Unsigned
arithmetic is used. Immediate (K) operands may sometimes be packed in the data area.

USGE R
Set the specified register to one if the first operand of a previous CMP instruction
was greater than or equal to its second operand using unsigned arithmetic, otherwise
set it to zero.

USGT R
Set the specified register or memory word to one if the first operand of a previous CMP
instruction was greater than its second operand using unsigned arithmetic, otherwise
set it to zero.

USLE R
Set the specified register to one if the first operand of a previous CMP instruction
was less than or equal to its second operand using unsigned arithmetic, otherwise set
it to zero.

USLT R
Set the specified register to one if the first operand of a previous CMP instruction
was less than its second operand using unsigned arithmetic, otherwise set it to zero.

XCHG RR RA RV RG RM RL RD RDX RDXs RDXsB

Exchange the values specified by the two operands.

XOR RA RV RG RM RL RD RDX RDXs RDXsB

RR AR VR GR MR LR DR DXR DXsR DXsBR
RK AK VK GK MK LK DK DXK DXsK DXsBK
Exclusive OR the second operand into the first.
234 CHAPTER 11. THE MC PACKAGE

11.4 MC Debugging Aids

The primary debugging aid is to inspect the generated code and the is controlled by
the DEBUG directive which sets the tracing level held in the global variable mcDebug.
Assuming bimc are the least significant four bit of mcDebug, if c = 1, print comments
compiled by mcComment. If m = 1, print MC instructions and directives. If i = 1, print
the corresponding target instruction(s) and if b = 1, print the resulting binary code in
hexadecimal. To fully understand this output it is, of course, necessary to have a good
understanding of the target architecture being used.
A second important debugging aid is provided by the mcPRF function which compiler
code to output the value of a specified register using a given printf format string. On
return all registers including the condition code are preserved. A typical call of mcPRF
is as follows.

mcPRF("The value of register A is %8x*n", mc_a)

As an aid to debugging MC packages themselves, there is a test program called

bcplprogs/mc/mcsystest.b which systematically tests all MC instructions, directives
and addressing modes generating error messages for each error found. Each such error
message includes a test number which helps to locate the source of the of the problem.
If mcsystest is given a test number as argument, it provides a detailed compilation
trace of the specified test. This should provide sufficient information to locate the error
in the package.

11.5 The n-queens Demonstration

This section shows how the algorithm to solve the n-queens problem as described in
Section 13.3 on page 248 can be reimplemented using the MC package. The MC version
of the program is as follows.

GET "libhdr"
GET "mc.h"
MANIFEST {
// Register mnemonics
ld = mc_a
col = mc_b
rd = mc_c
poss = mc_d
p = mc_e
count = mc_f
}
LET start() = VALOF
{ // Load the dynamic code generation package
LET argv = VEC 50
LET lo, hi, dlevel = 1, 16, #x0000
LET mcname = "mci386" // Default setting
LET mcseg = 0
LET mcb = 0
11.5. THE N-QUEENS DEMONSTRATION 235

UNLESS rdargs("mc,lo/n,hi/n,-c/s,-m/s,-a/s,-b/s", argv, 50) DO

{ writef("Bad arguments for mcqueens*n")
RESULTIS 0
}
IF argv!0 DO mcname := argv!0 // mc
IF argv!1 DO lo := !argv!1 // lo/n
IF argv!2 DO hi := !argv!2 // hi/n
IF argv!3 DO dlevel := dlevel | #b0001 // -c/s comments
IF argv!4 DO dlevel := dlevel | #b0010 // -m/s mc instructions
IF argv!5 DO dlevel := dlevel | #b0100 // -a/s assembler
IF argv!6 DO dlevel := dlevel | #b1000 // -b/s binary
mcseg := globin(loadseg(mcname))
UNLESS mcseg DO
{ writef("Trouble with MC package: mci386*n")
GOTO fin
}
// Create an MC instance for hi functions with a data space
// of 10 words and code space of 4000
mcb := mcInit(hi, 10, 40000)
UNLESS mcb DO
{ writef("Unable to create an mci386 instance*n")
GOTO fin
}
mc := 0 // Currently no selected MC instance
mcSelect(mcb)
mcK(mc_debug, dlevel)
FOR n = lo TO hi DO
{ mcComment("*n*n// Code for a %nx%n board*n", n, n)
gencode(n) // Compile the code for an nxn board
}
mcF(mc_end)
writef("Code generation complete*n")
FOR n = lo TO hi DO
{ LET k = 0
writef("Calling mcCall(%n)*n", n)
k := mcCall(n)
writef("Number of solutions to %i2-queens is %i9*n", n, k)
}
fin:
IF mc DO mcClose()
IF mcseg DO unloadseg(mcseg)
writef("*n*nEnd of run*n")
236 CHAPTER 11. THE MC PACKAGE

}
AND gencode(n) BE
{ LET all = (1<<n) - 1
mcKKK(mc_entry, n, 3, 0)
mcRK(mc_mv, ld, 0)
mcRK(mc_mv, col, 0)
mcRK(mc_mv, rd, 0)
mcRK(mc_mv, count, 0)
cmpltry(1, n, all) // Compile the outermost call of try
mcRR(mc_mv, mc_a, count) // return count
mcF(mc_rtn)
mcF(mc_endfn)
}
AND cmpltry(i, n, all) BE
{ LET L = mcNextlab()
mcComment("*n// Start of code from try(%n, %n, %n)*n", i, n, all)
mcRR(mc_mv, poss, ld) // LET poss = (~(ld | col | rd)) & all
mcRR(mc_or, poss, col)
mcRR(mc_or, poss, rd)
mcR (mc_not, poss)
mcRK(mc_and, poss, all)
mcRK(mc_cmp, poss, 0) // IF poss DO
TEST n-i<=2
THEN mcJS(mc_jeq, L) // (use a short jump if near the last row)
ELSE mcJL(mc_jeq, L)
TEST i=n
THEN { // We can place a queen in the final row.
mcR(mc_inc, count) // count := count+1
}
ELSE { // We can place queen(s) in a non final row.
LET M = mcNextlab()
mcL (mc_lab, M) // { Start of REPEATWHILE loop
mcRR(mc_mv, p, poss) // LET p = poss & -poss
mcR (mc_neg, p)
mcRR(mc_and, p, poss) // // p is a valid queens position
mcRR(mc_sub, poss, p) // poss := poss - p

mcR (mc_push, ld) // Save current state

mcR (mc_push, col)
mcR (mc_push, rd)
mcR (mc_push, poss)
// Call try((ld+p)<<1, col+p, (rd+p)>>1)
mcRR(mc_add, ld, p)
mcRK(mc_lsh, ld, 1) // ld := (ld+p)<<1
11.5. THE N-QUEENS DEMONSTRATION 237

mcRR(mc_add, col, p) // col := col+p

mcRR(mc_add, rd, p)
mcRK(mc_rsh, rd, 1) // rd := (rd+p)>>1
cmpltry(i+1, n, all) // Compile code for row i+1
mcR (mc_pop, poss) // Restore the state
mcR (mc_pop, rd)
mcR (mc_pop, col)
mcR (mc_pop, ld)
mcRK(mc_cmp, poss, 0)
mcJL(mc_jne, M) // } REPEATWHILE poss
}
mcL(mc_lab, L)
mcComment("// End of code from try(%n, %n, %n)*n*n",
i, n, all)
}

In this implementation all the working variables are held in registers and all re-
cursive calls are unwound knowing that the depth of recursion will be limited, in this
case to no more than 16. The stack is used to save the state at the moment when a
recursive call would have been made in the original program. An optimisation is done
based on the knowledge that if a queen can be placed on the nth row of n × n board
then the solution count can be incremented.
When running on a Pentium IV this implementation executes approximately 24
times faster than the normal interpretive Cintcode version and 25% faster than the
corresponding optimised C version of the algorithm.
238 CHAPTER 11. THE MC PACKAGE
Chapter 12

Installation

The implementation of BCPL described in this report is freely available via my Home
Page [3] to individuals for private use and to academic institutions. If you install the
system, please send me an email (to [email protected]) so I can keep a record of
who is interested in it.
This implementation is designed to be machine independent being based on an in-
terpreter written in C. There are, however, hand written assembly language versions
of the interpreter for several architectures (including i386, MIPS, ALPHA and Hitachi
SH3), although these are now little used and are becoming out of date. For Win-
dows XP and Windows 10 there are precompiled .exe files such as wincintsys.exe
and winrastsys.exe. These files should be copied into the appropriate bin directory
and renamed as cintsys.exe and rastsys.exe. For all the other architectures it is
necessary to rebuild the system, but this should be reasonably easy to do.
The simplest installation is for 32-bit Linux machines which will be covered in detail
here.

12.1 Linux Installation

This section describes how to install the BCPL Cintcode System on a Linux machine
using an Intel 386 or later Intel processors. It can be installed on both 32 and 64 bit
architectures, and the size of the BCPL word can be either 32 or 64 bits. To rebuild
the BCPL Cintcode system perform the followwing steps.
First create a directory typically named distribution in your home directory
($HOME) and extract the BCPL distribution files in bcpl.tgz by, typically, typing the
commands:

ch $HOME/distribution
tar zxvf ../Downloads/bcpl.tgz

This creates the directory BCPL containing all the files needed to rebuild the system.
Next enter the cintcode directory by typing the following command.

239
240 CHAPTER 12. INSTALLATION

cd BCPL/cintcode

You are now ready to rebuild the system, but first you must ensure that the envi-
ronment variables BCPLROOT, BCPLHDRS, BCPLPATH, BCPLSCRIPTS are properly defined.
For convenience, there is a bash shell script in os/Linux/setbcplenv. This script also
adds the directory cintcode/bin to the PATH variable. To set all the variables run the
following command.

. $(HOME)/distribution/BCPL/cintcode/os/linux/setbcplenv

It is probably even better to place this line near the end of ~/.bashrc so that the
environment variables are setup every time a new shell window is created.
You are now ready to rebuild the BCPL system by typing the following commands.

cd $(HOME)/distribution/BCPL/cintcode
make clean
make

This should recompile all the C and BCPL code required by the BCPL system and
leave it waiting for the user to type a BCPL Cmmand Language (CLI) command. To
test it type the following commands.

type com/hello.b
c bc hello
hello
bcpl com/bcpl.b to junk
junk com/bcpl.b to junk
c bc bcpl
bench100
c bc cmpltest
cmpltest

What the make command did perhaps needs some explanation. Without arguments
make reads the file Makefile from the current directory and performs the first action
it finds in this file. This first causes bin/cintsys to be created by compiling and
linking all the source files needed to build cintsys. But before doing this it creates
the #include file sysc/INT.h by compiling and running sysc/mkint-h.c. INT.h con-
tains several #define macros that that allow the C programs to determine important
properties of the host machine, such as the C types for signed and unsigned characters.
The C source files for cintsys are all in the directory sysc/ and are: cintsys.c,
cinterp.c, kblib.c, cfuncs.c, joyfn.c, sdlfn.c, glfn.c and extfn.c.
Although cintsys can now be called, it will only work if precompiled Cintcode
compilations of sysb/boot.b, sysb/blib.b, sysb/dlib.b, sysb/cli.b, are placed in
the directorie cin/syscin/. The hand written Cintcode file syslib must also be placed
there for the (trivial) definitions of the functions sys, changeco and muldiv. Compiled
12.1. LINUX INSTALLATION 241

versions of the commands abort, c, echo and bcpl are then placed in cin/. Finally
several scripts such as b, bc and bs are placed in cintcode/. Most of these files have
different versions depending on whether the host is a big or little ender machine.
Finally, make causes the command c compall to be executed on the newly created
system. This compiles all the resident system components contained in sysb and the
standard commands in com/. The system is then ready for use.
You will notice that directory BCPL contains BCPL/cintcode, BCPL/bcplprogs and
BCPL/natbcpl. The directory BCPL/cintcode contains all the source files of the BCPL
Cintcode System, BCPL/bcplprogs contains a collection of directories holding demon-
stration programs, and BCPL/natbcpl contains a version of BCPL that compiles into
native code (for Intel and ALPHA machines) using a mechanism based on the Sial
abstract machine code.
Once the system hase been built it is normally entered using the command cintsys
which can be called when in any directory. If anything has gone wrong various debug-
ging aids can be turned on using either

cintsys -f -v

or

cintsys -f -vv

The output should be studied in conjunction with sysc/cintsys.c and sysb/boot.b.

Hopefully, there will be enough information there to diagnose and correct the problem.
It includes, in particular, a trace of all uses of the shell environment variables which
are a common source of trouble.
Read the documentation in cintcode/doc/ and any README files you can
find. A log of recent changes can be found in cintcode/doc/changes. A log
of recent tests under different machines and operating systems can be found in
cintcode/CintsysTestsLog.txt. The current version of this BCPL manual is avail-
able from my home page as a .pdf file. There is an extensive demonstration script of
commands in cintcode/doc/notes.
To create the 64-bit version of Cintcode BCPL, type the following.

make clean64
make sys64
cintsys64

The resulting system is almost identical to the standard 32-bit Cintcode BCPL system
but uses a BCPL word length of 64 bits rather that the normal 32.
242 CHAPTER 12. INSTALLATION

Other versions of the system that can be created using other make files, for instance:
make -f MakefileSDL clean
make -f MakefileSDL
This will provide a version with an interface to the SDL graphics library. An interface
to the OpenGL graphics library is provided if MakefileGL is used. The GL version can
be demonstrated by the follow sequence of commands.
cd $(HOME)
cd ../bcplprogs/raspi
cintsys
sysinfo
c b engine
engine
c b dragon
dragon
c b bucket
bucket
c b gltst
gltst

When you enter cintsys you can choose one of two Cintcode interpreters. These
can be selected by the commands fast and slow. The slow interpreter performs
more runtime checks and has mode debugging aids than the fast interpreter and is
thus somewhat slower. Both interpreters are compilations of the same source file
sysc/cinterp.c with the differences controlled by conditional compilation statements
such as #ifdef FASTERPyes.
The make command actually creates rastsys in addition to cintsys. This is a ve-
rion of the system that allows the user to generate raster data that can be used to make
graphs such as the one in Figure 4.2 on page 149 showing memory references during
the compilation of a BCPL compiler. This version is built from the same the source
programs in C using conditional compilation statements such as #ifdef RASTERPyes.
There is a different but related system called cintpos that is closely related to
cintsys. It is a Cintcode based implementation of the Tripos Portable Operating
System originally implemented at Cambridge in the late 1970s. This system allows the
user to create tasks which in modern terminology would be called threads since they
all use the same address space. Information can be sent from one task to another using
the call qpkt(pkt). This appends the packet on the end of a work queue belonging to
the destination task. A task can extract the first packet on its work queue using a call
of taskwait(). If the work queue is empty the task becomes suspended. Every task
has a distinct integer priority and there is a scheduler that ensures the highest priority
task that can run is given control. As with the BCPL distribution, Cintpos has its own
directory ~/distribution/Cintpos and all its files are contained in cintpos.tgz.
Two of the main programs of Cintpos are called cintpos.c and cinterp.c. These
have much in common with cintsys.c and cinterp.c of the BCPL distribution and
the plan is make the C programs in Cintpos identical to the corresponding ones in
the BCPL distribution with the the differences controlled by conditional compilation
statements such as #ifdef CINTSYSyes and #ifdef CINTPOSyes. This change is still
under development.
12.2. COMMAND LINE ARGUMENTS 243

12.2 Command Line Arguments

The commands cintsys, cintsys64 and cintpos that invoke the Cintcode interpreter
can be given various arguments. These are:

-m n Set the Cintcode memory size to n words.

-t n Set the tally vector size to n words.
-s Enter the Cintcode system giving the name of this
file as the command for the CLI to run.
-c text Enter cintsys with standard input
setup to read the characters from text followed by
an end-of-stream character.
-- text Enter cintsys with standard input
setup to read the characters in text followed by
the characters of the old standard input.
-f Trace the use of environment variables in pathinput
-v Trace the bootstrapping process
-vv As -v, but also include some Cincode level tracing
-h Output some help information.

The rastering versions of the interpreter rastsys, rastsys64 can receive the same
arguments.

12.3 Installation on Other Machines

Carry out steps 1 to 4 above. In the directory BCPL/cintcode/sysasm you will find
directories for different architectures, e.g. ALPHA, MIPS, SUN4, SPARC, MSDOS,
MAC, OS2, BC4, Win32, CYGWIN32 and shWinCE. These contain files that are
architecture (or compiler) dependent, typically including cintasm.s (or cintasm.asm).
For some old versions of Linux, it is necessary to change _dosys to dosys (or vice-versa)
in the file sysasm/LINUX/cintasm.s.
Edit Makefile (typically by adding and removing comment symbols) as necessary
for your system/machine and then execute make in the cintcode directory, e.g:

make

Variants of the above should work for the other architectures running Unix.

12.4 Installation for Windows XP

The files wincintsys.exe and winrastsys.exe are included in the standard distribu-
tion and should work under many versions of the Windows operating systems (such as
Windows XP) just by typing the command:

wincintsys
244 CHAPTER 12. INSTALLATION

It may be more convenient to move them into a different directory and rename
them as cintsys.exe and rastsys.exe.
I have recently upgraded the Windows version of BCPL so that it can be compiled
and run using the freely available Microsoft C compiler and libraries. On a new PC
I installed the freely available .NET Framework 3.5 and the corresponding SDK 3.5.
This provided amongst many other things a C compiler and all the relevant libraries.
I then created a shortcut on the desktop with
Target: %SystemRoot%\system32\cmd.exe /q /k os\windows\VC8env.bat
and
Start in: E:\distribution\BCPL\cintcode
Double clicking on this shortcut opens a Shell window with the required environ-
ment variable all set up C compilation and the BCPL running environment. If they
are not correct you may have to edit VC8env.bat. The BCPL system was then rebuilt
by the commands:
nmake /f os/windows/MakefileVC clean
nmake /f os/windows/MakefileVC
This should recompile and link all the C code of the BCPL Cintcode system and
then recompile all the standard BCPL system programs and commands. For good
measure, once the BCPL Cintcode system has been entered, recompile all the BCPL
code again by typing:
c compall

12.5 Installation using Cygwin

I recommend using the GNU development tools and utilities for Windows that are
available from http://sourceware.cygnus.com/cygwin/.
Edit the cintcode/Makefile to comment out the LINUX version
CC = gcc -O9 -DforLINUX -DSOUND -DCALLC -lm
SYSM = ../cintcode/sysasm/linux
and enable the CYGWIN32 version
CC = gcc -O9 -DforCYGWIN32 -DSOUND -DCALLC -lm
SYSM = ../cintcode/sysasm/CYGWIN32
Then type:
make
This should recompile the system and create the executable cintsys.exe.
Remember to include the cintcode directory in your PATH and BCPLPATH shell
variables, so that the cintsys can be run in any directory.
Careful inspection of the Makefile and directories in cintcode/sysasm will show
that versions also exist that use Microsoft C++ 5.0 and Borland C4.0, but these are
likely to be out of date and their use is not recommended.
12.6. INSTALLATION FOR WINDOWS CE2.0 245

12.6 Installation for Windows CE2.0

A version of the BCPL Cintcode System is available for handheld machines run-
ning Windows CE version 2.0. For installation details see the README file in
sysasm/shwince. This system provides a scrollable window for interaction with the
CLI. It also provides a simple graphical facilities using a graphics window. The system
has only been tested on an HP 620LX handheld machine.

12.7 The Native Code Version

A BCPL native mode system for 686/Pentium based machines is in directory
BCPL/natbcpl. It can be re-built and tested by changing to the directory BCPL/natbcpl
and running make. If you have the SDL libraries installed (see bcpl4raspi.pdf), you
could try

make -f MakefileSDL clean

make -f MakefileSDL bucket
./bucket

A version (64 bit) for the DEC Alpha is also available but is now out of date and
has not been tested recently. To re-build it, it is necessary to comment out the lines
for Linux and uncomment the lines for the ALPHA in Makefile, before running make.
Recently, a version for the ARM processor has been added, particularly for the
Raspberry Pi machine. In directory BCPL/natbcpl on the Raspberry Pi, try typing

make -f MakefileRaspi clean

make -f MakefileRaspi

If you have the SDL libraries installed (see bcpl4raspi.pdf), you could try

make -f MakefileRaspiSDL clean

make -f MakefileRaspiSDL bucket
./bucket

It is useful to know how the make commands such as those above work. Here is a
brief explanation.
The command make clean just deletes all previously built executables together
with all files in the directories obj, sial, temps and tempc since these can easily be
recreated.
The call make prog causes the required BCPL programs to be compiled, if neces-
sary, into Pentium assembly language by executing the following CLI commands. This
also ensures the C program tempc/initprog.c is up to date.
246 CHAPTER 12. INSTALLATION

bcpl2sial ./prog.b to sial/prog.sial noselst

sial-686 -t sial/prog.sial to temps/prog.s
bcpl2sial sysb/blib.b to sial/blib.sial noselst
sial-686 -t sial/blib.sial to temps/blib.s
bcpl2sial ../cintcode/sysb/dlib.b to sial/dlib.sial noselst
sial-686 -t sial/dlib.sial to temps/dlib.s
makeinit prog.b to tempc/initprog.c

If necessary make prog also updates the header file tempc/INT.h needed by clib.c
using the following bash commands.

gcc -o mkint-h sysc/mkint-h.c

./mkint-h >sysc/INT.h
rm -f mkint-h
cp sysc/INT.h tempc
cp sysc/bcpl.h tempc

Finally it updates the executable prog, if necessary, by compiling and linking all
the required C and assembly language programs.

gcc -O9 -DforLINUX -o obj/initprog.o -c tempc/initprog.c

gcc -O9 -DforLINUX -o obj/clib.o -c sysc/clib.c
gcc -O9 -DforLINUX -o obj/kblib.o -c sysc/kblib.c
gcc -O9 -DforLINUX -o obj/sdlfn.o -c sysc/sdlfn.c
gcc -O9 -DforLINUX -o obj/prog.o -c temps/prog.s
gcc -O9 -DforLINUX -o obj/blib.o -c temps/blib.s
gcc -O9 -DforLINUX -o obj/dlib.o -c temps/dlib.s
gcc -O9 -DforLINUX -o obj/mlib.o -c i386/mlib.s
gcc -O9 -DforLINUX -o prog
obj/initprog.o obj/prog.o
obj/mlib.o obj/clib.o obj/blib.o
obj/dlib.o obj/kblib.o obj/sdlfn.o -lm

The native code program can now be executed in a bash shell using the command
prog or possibly ./prog.
Chapter 13

Example Programs

13.1 Coins
The following program prints out how many different ways a sum of money can be
composed from coins of various denominations.

GET "libhdr"
LET coins(sum) = c(sum, (TABLE 200, 100, 50, 20, 10, 5, 2, 1, 0))
AND c(sum, t) = sum<0 -> 0,
sum=0 -> 1,
!t=0 -> 0,
c(sum, t+1) + c(sum-!t, t)
LET start() = VALOF
{ writes("Coins problem*n")
t(0); t(1); t(2); t(5); t(21); t(100); t(200)
RESULTIS 0
}
AND t(n) BE writef("Sum = %i3 number of ways = %i6*n", n, coins(n))

247
248 CHAPTER 13. EXAMPLE PROGRAMS

13.2 Primes
The following program prints out a table of all primes less than 1000, using the sieve
method.
GET "libhdr"
GLOBAL { count: ug }
MANIFEST { upb = 999 }
LET start() = VALOF
{ LET isprime = getvec(upb)
count := 0
FOR i = 2 TO upb DO isprime!i := TRUE // Until proved otherwise.
FOR p = 2 TO upb IF isprime!p DO
{ LET i = p*p
UNTIL i>upb DO { isprime!i := FALSE; i := i + p }
out(p)
}
writes("*nend of output*n")
freevec(isprime)
RESULTIS 0
}
AND out(n) BE
{ IF count MOD 10 = 0 DO newline()
writef(" %i3", n)
count := count + 1
}

13.3 Queens
The following program calculates the number of ways n queens can be placed on a n×n
chess board without any two occupying the same row, column or diagonal.
GET "libhdr"
GLOBAL { count:200; all:201 }
LET try(ld, col, rd) BE TEST col=all
THEN count := count + 1
ELSE { LET poss = all & ~(ld | col | rd)
UNTIL poss=0 DO
{ LET p = poss & -poss
poss := poss - p
try(ld+p << 1, col+p, rd+p >> 1)
}
}
13.4. FRIDAYS 249

LET start() = VALOF

{ all := 1
FOR i = 1 TO 16 DO
{ count := 0
try(0, 0, 0)
writef("Number of solutions to %i2-queens is %i9*n", i, count)
all := 2*all + 1
}
RESULTIS 0
}

13.4 Fridays
The following program prints a table of how often the 13th day of the month lies on
each day of the week over a 400 year period. Since there are an exact number of weeks
in 4 centuries, program shows that the 13th is most of a Friday!

GET "libhdr"
MANIFEST { mon=0; sun=6; jan=0; feb=1; dec=11 }
LET start() = VALOF
{ LET count = TABLE 0, 0, 0, 0, 0, 0, 0
LET daysinmonth = TABLE 31, ?, 31, 30, 31, 30,
31, 31, 30, 31, 30, 31
LET days = 0
FOR year = 1973 TO 1973+399 DO
{ daysinmonth!feb := febdays(year)
FOR month = jan TO dec DO
{ LET day13 = (days+12) MOD 7
count!day13 := count!day13 + 1
days := days + daysinmonth!month
}
}
FOR day = mon TO sun DO
writef("%i3 %sdays*n",
count!day,
select(day,
"Mon","Tues","Wednes","Thurs","Fri","Sat","Sun")
)
RESULTIS 0
}
AND febdays(year) = year MOD 400 = 0 -> 29,
year MOD 100 = 0 -> 28,
year MOD 4 = 0 -> 29,
28
AND select(n, a0, a1, a2, a3, a4, a5, a6) = n!@a0
250 CHAPTER 13. EXAMPLE PROGRAMS

13.5 Lambda Evaluator

The following program is a simple parser and evaluator for lambda expressions.

GET "libhdr"
MANIFEST {
// selectors
H1=0; H2; H3; H4
// Expression operators and tokens
Id=1; Num; Pos; Neg; Mul; Div;Add; Sub
Eq; Cond; Lam; Ap; Y
Lparen; Rparen; Comma; Eof
}
GLOBAL {
space:200; str; strp; strt; ch; token; lexval
}
LET lookup(bv, e) = VALOF
{ WHILE e DO { IF bv=H1!e RESULTIS H2!e
e := H3!e
}
writef("Undeclared name %c*n", H2!bv)
RESULTIS 0
}
AND eval(x, e) = VALOF SWITCHON H1!x INTO
{ DEFAULT: writef("Bad exppression, Op=%n*n", H1!x)
RESULTIS 0
CASE Id: RESULTIS lookup(H2!x, e)
CASE Num: RESULTIS H2!x
CASE Pos: RESULTIS eval(H2!x, e)
CASE Neg: RESULTIS - eval(H2!x, e)
CASE Add: RESULTIS eval(H2!x, e) + eval(H3!x, e)
CASE Sub: RESULTIS eval(H2!x, e) - eval(H3!x, e)
CASE Mul: RESULTIS eval(H2!x, e) * eval(H3!x, e)
CASE Div: RESULTIS eval(H2!x, e) / eval(H3!x, e)
CASE Eq: RESULTIS eval(H2!x, e) = eval(H3!x, e)
CASE Cond: RESULTIS eval(H2!x, e) -> eval(H3!x, e), eval(H4!x, e)
CASE Lam: RESULTIS mk3(H2!x, H3!x, e)
CASE Ap: { LET f, a = eval(H2!x, e), eval(H3!x, e)
LET bv, body, env = H1!f, H2!f, H3!f
RESULTIS eval(body, mk3(bv, a, env))
}
CASE Y: { LET bigf = eval(H2!x, e)
// bigf should be a closure whose body is an
// abstraction eg Lf Ln n=0 -> 1, n*f(n-1)
LET bv, body, env = H1!bigf, H2!bigf, H3!bigf
// Make a closure with a missing environment
LET yf = mk3(H2!body, H3!body, ?)
// Make a new environment including an item for bv
LET ne = mk3(bv, yf, env)
H3!yf := ne // Now fill in the environment component
RESULTIS yf // and return the closure
}
}
13.5. LAMBDA EVALUATOR 251

// *************** Syntax analyser ***********************

// Construct Corresponding Tree
// a ,.., z --> [Id, ’a’] ,.., [Id, ’z’]
// dddd --> [Num, dddd]
// x y --> [Ap, x, y]
// Y x --> [Y, x]
// x * y --> [Times, x, y]
// x / y --> [Div, x, y]
// x + y --> [Plus, x, y]
// x - y --> [Minus, x, y]
// x = y --> [Eq, x, y]
// b -> x, y --> [Cond, b, x, y]
// Li y --> [Lam, i, y]
LET mk1(x) = VALOF { space := space-1; !space := x; RESULTIS space }
AND mk2(x,y) = VALOF { mk1(y); RESULTIS mk1(x) }
AND mk3(x,y,z) = VALOF { mk2(y,z); RESULTIS mk1(x) }
AND mk4(x,y,z,t) = VALOF { mk3(y,z,t); RESULTIS mk1(x) }
AND rch() BE
{ ch := Eof
IF strp>=strt RETURN
strp := strp+1
ch := str%strp
}
AND parse(s) = VALOF
{ str, strp, strt := s, 0, s%0
rch()
RESULTIS nexp(0)
}
252 CHAPTER 13. EXAMPLE PROGRAMS

AND lex() BE SWITCHON ch INTO

{ DEFAULT: writef("Bad ch in lex: %c*n", ch)
CASE Eof: token := Eof
RETURN
CASE ’ ’:
CASE ’*n’ :rch(); lex(); RETURN
CASE ’a’:CASE ’b’:CASE ’c’:CASE ’d’:CASE ’e’:
CASE ’f’:CASE ’g’:CASE ’h’:CASE ’i’:CASE ’j’:
CASE ’k’:CASE ’l’:CASE ’m’:CASE ’n’:CASE ’o’:
CASE ’p’:CASE ’q’:CASE ’r’:CASE ’s’:CASE ’t’:
CASE ’u’:CASE ’v’:CASE ’w’:CASE ’x’:CASE ’y’:
CASE ’z’:
token := Id; lexval := ch; rch(); RETURN
CASE ’0’:CASE ’1’:CASE ’2’:CASE ’3’:CASE ’4’:
CASE ’5’:CASE ’6’:CASE ’7’:CASE ’8’:CASE ’9’:
token, lexval := Num, 0
WHILE ’0’<=ch<=’9’ DO
{ lexval := 10*lexval + ch - ’0’
rch()
}
RETURN
CASE ’-’: rch()
IF ch=’>’ DO { token := Cond; rch(); RETURN }
token := Sub
RETURN
CASE ’+’: token := Add; rch(); RETURN
CASE ’(’: token := Lparen; rch(); RETURN
CASE ’)’: token := Rparen; rch(); RETURN
CASE ’**’: token := Mul; rch(); RETURN
CASE ’/’: token := Div; rch(); RETURN
CASE ’L’: token := Lam; rch(); RETURN
CASE ’Y’: token := Y; rch(); RETURN
CASE ’=’: token := Eq; rch(); RETURN
CASE ’,’: token := Comma; rch(); RETURN
}
13.5. LAMBDA EVALUATOR 253

AND prim() = VALOF

{ LET a = TABLE Num, 0
SWITCHON token INTO
{ DEFAULT: writef("Bad expression*n"); ENDCASE
CASE Id: a := mk2(Id, lexval); ENDCASE
CASE Num: a := mk2(Num, lexval); ENDCASE
CASE Y: RESULTIS mk2(Y, nexp(6))
CASE Lam: lex()
UNLESS token=Id DO writes("Id expected*n")
a := lexval
RESULTIS mk3(Lam, a, nexp(0))
CASE Lparen: a := nexp(0)
UNLESS token=Rparen DO writef("’)’ expected*n")
lex()
RESULTIS a
CASE Add: RESULTIS mk2(Pos, nexp(3))
CASE Sub: RESULTIS mk2(Neg, nexp(3))
}
lex()
RESULTIS a
}

AND nexp(n) = VALOF { lex(); RESULTIS exp(n) }

AND exp(n) = VALOF
{ LET a, b = prim(), ?
{ SWITCHON token INTO
{ DEFAULT: BREAK
CASE Lparen:
CASE Num:
CASE Id: UNLESS n<6 BREAK
a := mk3(Ap, a, exp(6)); LOOP
CASE Mul: UNLESS n<5 BREAK
a := mk3(Mul, a, nexp(5)); LOOP
CASE Div: UNLESS n<5 BREAK
a := mk3(Div, a, nexp(5)); LOOP
CASE Add: UNLESS n<4 BREAK
a := mk3(Add, a, nexp(4)); LOOP
CASE Sub: UNLESS n<4 BREAK
a := mk3(Sub, a, nexp(4)); LOOP
CASE Eq: UNLESS n<3 BREAK
a := mk3(Eq, a, nexp(3)); LOOP
CASE Cond: UNLESS n<1 BREAK
b := nexp(0)
UNLESS token=Comma DO writes("Comma expected*n")
a := mk4(Cond, a, b, nexp(0)); LOOP
}
} REPEAT
RESULTIS a
}
254 CHAPTER 13. EXAMPLE PROGRAMS

AND try(expr) BE
{ LET v = VEC 2000
space := v+2000
writef("Trying %s*n", expr)
writef("Answer: %n*n", eval(parse(expr), 0))
}
AND start() = VALOF
{ try("(Lx x+1) 2")
try("(Lx x) (Ly y) 99")
try("(Ls Lk s k k) (Lf Lg Lx f x (g x)) (Lx Ly x) (Lx x) 1234")
try("(Y (Lf Ln n=0->1,n**f(n-1))) 5")
RESULTIS 0
}

13.6 Fast Fourier Transform

The following program is a simple demonstration of the algorithm for the fast fourier
transform. Instead of using complex numbers, it uses integer arithmetic modulo 65537
with an appropriate Nth root of unity.

GET "libhdr"
MANIFEST {
modulus = #x10001 // 2**16 + 1
$$ln10 // Set condition compilation flag to select data size
//$$walsh
$<ln16 omega = #x00003; ln = 16 $>ln16 // omega**(2**16) = 1
$<ln12 omega = #x0ADF3; ln = 12 $>ln12 // omega**(2**12) = 1
$<ln10 omega = #x096ED; ln = 10 $>ln10 // omega**(2**10) = 1
$<ln4 omega = #x08000; ln = 4 $>ln4 // omega**(2**4) = 1
$<ln3 omega = #x0FFF1; ln = 3 $>ln3 // omega**(2**3) = 1
$<walsh omega=1 $>walsh // The Walsh transform
N = 1<<ln // N is a power of 2
upb = N-1
}
STATIC { data=0 }
13.6. FAST FOURIER TRANSFORM 255

LET start() = VALOF

{ writef("fft with N = %n and omega = %n modulus = %n*n*n",
N, omega, modulus)
data := getvec(upb)
UNLESS omega=1 DO // Unless doing Walsh tranform
check(omega, N) // check that omega and N are consistent
FOR i = 0 TO upb DO data!i := i
pr(data, 7)
// prints -- Original data
// 0 1 2 3 4 5 6 7
fft(data, ln, omega)
pr(data, 7)
// prints -- Transformed data
// 65017 26645 38448 37467 30114 19936 15550 42679
fft(data, ln, ovr(1,omega))
FOR i = 0 TO upb DO data!i := ovr(data!i, N)
pr(data, 7)
// prints -- Restored data
// 0 1 2 3 4 5 6 7
RESULTIS 0
}

AND fft(v, ln, w) BE // ln = log2 n w = nth root of unity

{ LET n = 1<<ln
LET vn = v+n
LET n2 = n>>1
// First do the perfect shuffle
reorder(v, n)
// Then do all the butterfly operations
FOR s = 1 TO ln DO
{ LET m = 1<<s
LET m2 = m>>1
LET wk, wkfac = 1, w
FOR i = s+1 TO ln DO wkfac := mul(wkfac, wkfac)
FOR j = 0 TO m2-1 DO
{ LET p = v+j
WHILE p<vn DO { butterfly(p, p+m2, wk); p := p+m }
wk := mul(wk, wkfac)
}
}
}
AND butterfly(p, q, wk) BE { LET a, b = !p, mul(!q, wk)
!p, !q := add(a, b), sub(a, b)
}
256 CHAPTER 13. EXAMPLE PROGRAMS

AND reorder(v, n) BE
{ LET j = 0
FOR i = 0 TO n-2 DO
{ LET k = n>>1
// j is i with its bits is reverse order
IF i<j DO { LET t = v!j; v!j := v!i; v!i := t }
// k = 100..00 10..0000..00
// j = 0xx..xx 11..10xx..xx
// j’ = 1xx..xx 00..01xx..xx
// k’ = 100..00 00..0100..00
WHILE k<=j DO { j := j-k; k := k>>1 } //) "increment" j
j := j+k //)
}
}
AND check(w, n) BE
{ // Check that w is a principal nth root of unity
LET x = 1
FOR i = 1 TO n-1 DO { x := mul(x, w)
IF x=1 DO writef("omega****%n = 1*n", i)
}
UNLESS mul(x, w)=1 DO writef("Bad omega**%n should be 1*n", n)
}
AND pr(v, max) BE
{ FOR i = 0 TO max DO { writef("%I5 ", v!i)
IF i MOD 8 = 7 DO newline()
}
newline()
}
AND dv(a, m, b, n) = a=1 -> m,
a=0 -> m-n,
a<b -> dv(a, m, b MOD a, m*(b/a)+n),
dv(a MOD b, m+n*(a/b), b, n)

AND inv(x) = dv(x, 1, modulus-x, 1)

AND add(x, y) = VALOF
{ LET a = x+y
IF a<modulus RESULTIS a
RESULTIS a-modulus
}
AND sub(x, y) = add(x, neg(y))
AND neg(x) = modulus-x
AND mul(x, y) = x=0 -> 0,
(x&1)=0 -> mul(x>>1, add(y,y)),
add(y, mul(x>>1, add(y,y)))
AND ovr(x, y) = mul(x, inv(y))
Bibliography

[1] D.T. Ross et al. AED-0 programmer’s guide and user kit. Technical report, Elec-
tronic Systems Laboratory M.I.T, 1964.

[2] C. Jobson and J.M. Richards. BCPL for the BBC Microcomputer. Acornsoft Ltd,
Cambridge, 1983.

[3] M. Richards. My WWW Home Page. www.cl.cam.ac.uk/users/mr/.

[4] M. Richards. The Implementation of CPL-like programming languages. Phd thesis,

Cambridge University, 1966.

[5] M. Richards, A.R. Aylward, P. Bond, R.D. Evans, and B.J. Knight. The Tripos
Portable Operating System for Minicomputers. Software-Practice and Experience,
9:513–527, June 1979.

[6] Christopher Strachey. A General Purpose Macrogenerator. Computer Journal,

8(3):225–241, 1965.

257
258 BIBLIOGRAPHY
Appendix A

BCPL Syntax Diagrams

The syntax of standard BCPL is specified using the transition diagrams given in fig-
ures A.1, A.2, A.3 and A.4. In extended BCPL the floating point operators have
the same precedence as the corresponding integer ones, and the op:= operators are
syntactically identical to the := operator. The syntax of the more binding sequenc-
ing operator (<>) requires some new diagrams to be drawn. This will be done in
due course. It is sufficient to know that <> is more binding than DO, THEN, ELSE,
REPEAT, REPEATWHILE, REPEATUNTIL, and colon. Within the diagrams the syntactic
categories program, section, declaration, command and expressionn are represented
by the rounded boxes: program , section , D , C , En , respectively.
The rectangular boxes are called test boxes and can only be traversed if the con-
dition labelling the box matches the current input. When the label is a token, as in
WHILE and := , it must match the next input token for the test to succeed. The
test box eof is only satisfied if the end of file has been reached. Sometimes the
test box contains a side condition, as in REM n<6 , in which case the side condition
must also be satisfied. The only other test boxes are is call and is name which
are only satisfied if the most recently read expression is syntactically a function call
or a name, respectively. By setting n successively from 0 to 8 in the definition of the
category En , we obtain the definitions of E0 to E8 . Starting from the
definition of program , we can construct an infinite transition diagram containing
only test boxes by simply replacing all rounded boxes by their definitions, recursively.
The parsing algorithm searches through this infinite diagram for a path with the same
sequence of tokens as the program being parsed. In order to eliminate ambiguities,
the left hand branch at a branch point is tried first. Notice how this rule causes the
command

IF i>10 DO i := i/2 REPEATUNTIL i<5

to be equivalent to

IF i>10 DO { i := i/2 REPEATUNTIL i<5 }

259
260 APPENDIX A. BCPL SYNTAX DIAGRAMS

and not

{ IF i>10 DO i := i/2 } REPEATUNTIL i<5

A useful property of these diagrams is that, once a test box has been successfully
traversed, previous branching decisions need not be reconsidered and so the parser
never needs to backtrack.

program

section eof

section

SECTION string ;

NEEDS string ;

MANIFEST name = E0 ; } ;

STATIC {

name : E0 ; }

GLOBAL {

AND
, E0
=
LET name ( name )
BE C

Figure A.1: Program, Section

 261

MANIFEST name = E0 ; }

STATIC {

name : E0 ; }

GLOBAL {

AND
,
= E0
LET name ( name )
BE C

= VEC E0

, ,

, name = E0

Figure A.2: Declarations

262 APPENDIX A. BCPL SYNTAX DIAGRAMS

BREAK REPEATWHILE E0

LOOP REPEATUNTIL

ENDCASE REPEAT

RETURN

FINISH

SKIP

GOTO E0

RESULTIS

TEST E0 THEN C ELSE C

DO

FOR name = E0 TO E0 BY E0 THEN C

DO
IF E0

UNLESS

WHILE

UNTIL

{ D ; C ; }

SWITCHON E0 INTO {

, ,

E0 , E0 := E0

:= E0

is call

is name : C

CASE E0

DEFAULT

Figure A.3: Commands

 263

En
,

TRUE ( n<9 E0 )

FALSE

! n<8 E8
?
% n<8
name
OF n<8
number
* n<6 E6
character
/ n<6
string
MOD n<6

( E0 ) + n<5 E5

− n<5
! E7
= n<4 = n<4
@
~= n<4 ~= n<4
+ E5
< n<4 < n<4
−
> n<4 > n<4
ABS
<= n<4 <= n<4
NOT E3
>= n<4 >= n<4
,
E4
TABLE E0

<< n<4 E4
VALOF C
>> n<4

SLCT E9 & n<3 E3

: E9 | n<2 E2

EQV n<1 E1
: E9
XOR n<1

−> n<1 E0 , E0

Figure A.4: Expressions

264 APPENDIX A. BCPL SYNTAX DIAGRAMS