Shorter flattened names for nested $anons #852

retronym · 2012-07-08T21:24:01Z

The effect of the change is best described by the diff of flatten.check

Can you suggest additional interesting cases for test/files/run/flatten.scala?
Do we care that by unnesting the names we make it harder to mentally link the mangled name to its source code? (Incidentally, it would be great if we could teach YourKit to do this for us automatically, to save us either guessing / messing around with javap to figure it out).
I've done this for $anonFun, $anon: is there anything else that could have the same treatment?
The code to walk the rawowner chain was a bit fiddly to write and should be scrutinized. I didn't find an existing method on Symbol that did what I needed.

References SI-3623.

The following commit will change that phase to elide the noisy `$anon1$anon2` intermediaries from flat names of nested anonymous functions and classes.

This avoids ugly class names like Foo$anon1$anon2$anon3 when defining anonymous functions or classes in a nested location.

scala-jenkins · 2012-07-08T21:50:10Z

Started jenkins job pr-scala-testsuite-linux-opt at https://scala-webapps.epfl.ch/jenkins/job/pr-scala-testsuite-linux-opt/459/

scala-jenkins · 2012-07-08T22:43:11Z

jenkins job pr-scala-testsuite-linux-opt: Failed - https://scala-webapps.epfl.ch/jenkins/job/pr-scala-testsuite-linux-opt/459/

retronym · 2012-07-09T05:56:39Z

Build failure. SIGILL, que?

  [partest] testing: [...]/files/jvm/serialization-new.scala                      [FAILED]
  ...
  [partest] 
  [partest] x10 = Monday
  [partest] y10 = Monday
  [partest] x10 eq y10: true, y10 eq x10: true
  [partest] x10 equals y10: true, y10 equals x10: true
  [partest] 
  [partest] x9 eq x10: false, x10 eq x9: false
  [partest] x9 equals x10: false, x10 equals x9: false
  [partest] x9 eq y10: false, y10 eq x9: false
  [partest] x9 equals y10: false, y10 equals x9: false
  [partest] 
  [partest] f1 = <na>
  [partest] _f1 = <na>
  [partest] f1(2): 4, _f1(2): 4
  [partest] 
  [partest] xs0 = List(1, 2, 3)
  [partest] _xs0 = List(1, 2, 3)
  [partest] xs0 eq _xs0: false, _xs0 eq xs0: false
  [partest] xs0 equals _xs0: true, _xs0 equals xs0: true
  [partest] 
  [partest] xs1 = List()
  [partest] _xs1 = List()
  [partest] xs1 eq _xs1: true, _xs1 eq xs1: true
  [partest] 
  [partest] o1 = None
  [partest] _o1 = None
  [partest] o1 eq _o1: true, _o1 eq o1: true
  [partest] 
  [partest] o2 = Some(1)
  [partest] _o2 = Some(1)
  [partest] o2 eq _o2: false, _o2 eq o2: false
  [partest] o2 equals _o2: true, _o2 equals o2: true
  [partest] 
  [partest] s1 = 'hello
  [partest] _s1 = 'hello
  [partest] s1 eq _s1: true, _s1 eq s1: true
  [partest] s1 equals _s1: true, _s1 equals s1: true
  [partest] 
  [partest] t1 = (BannerLimit,12345)
  [partest] _t1 = (BannerLimit,12345)
  [partest] t1 eq _t1: false, _t1 eq t1: false
  [partest] t1 equals _t1: true, _t1 equals t1: true
  [partest] 
  [partest] x = BitSet(1, 2)
  [partest] y = BitSet(1, 2)
  [partest] x equals y: true, y equals x: true
  ...
  [partest] x = UnrolledBuffer(one, two)
  [partest] y = UnrolledBuffer(one, two)
  [partest] x equals y: true, y equals x: true
  [partest] 
  [partest] x = ParArray(abc, def, etc)
  [partest] y = ParArray(abc, def, etc)
  [partest] #
  [partest] # A fatal error has been detected by the Java Runtime Environment:
  [partest] #
  [partest] #  SIGILL (0x4) at pc=0xf77173f2, pid=28430, tid=2922589040
  [partest] #
  [partest] # JRE version: 6.0_31-b04
  [partest] # Java VM: Java HotSpot(TM) Server VM (20.6-b01 mixed mode linux-x86 )
  [partest] [thread -1372046480 also had an error]
  [partest] x equals y: true, y equals x: true
  [partest] # Problematic frame:
  [partest] # C  [libc.so.6+0x1563f2]  __libc_thread_freeres+0x36672
  [partest] #
  [partest] # An error report file with more information is saved as:
  [partest] # /localhome/hudson/workspace/pr-scala-testsuite-linux-opt/hs_err_pid28430.log
  [partest] #
  [partest] # If you would like to submit a bug report, please visit:
  [partest] #   http://java.sun.com/webapps/bugreport/crash.jsp

Works for me:


topic/elide-anon ~/code/scala ./test/partest test/files/jvm/serialization.scala  
Testing individual files
testing: [...]/files/jvm/serialization.scala                          [  OK  ]

All of 1 tests were successful (elapsed time: 00:00:12)
topic/elide-anon ~/code/scala ./test/partest test/files/jvm/serialization-new.scala 

Testing individual files
testing: [...]/files/jvm/serialization-new.scala                      [  OK  ]

All of 1 tests were successful (elapsed time: 00:00:11)

PLS REBUILD ALL

scala-jenkins · 2012-07-09T06:06:50Z

Started jenkins job pr-scala-testsuite-linux-opt at https://scala-webapps.epfl.ch/jenkins/job/pr-scala-testsuite-linux-opt/461/

scala-jenkins · 2012-07-09T07:27:10Z

jenkins job pr-scala-testsuite-linux-opt: Success - https://scala-webapps.epfl.ch/jenkins/job/pr-scala-testsuite-linux-opt/461/

Under the new scheme, they retain the method suffix added in SI-101.

scala-jenkins · 2012-07-09T18:58:28Z

Started jenkins job pr-scala-testsuite-linux-opt at https://scala-webapps.epfl.ch/jenkins/job/pr-scala-testsuite-linux-opt/466/

scala-jenkins · 2012-07-09T20:37:11Z

jenkins job pr-scala-testsuite-linux-opt: Success - https://scala-webapps.epfl.ch/jenkins/job/pr-scala-testsuite-linux-opt/466/

paulp · 2012-07-11T19:22:31Z

src/reflect/scala/reflect/internal/Symbols.scala

I never really figured out whether it was intentional or not, but ANON_CLASS_NAME is a substring of ANON_FUN_NAME, so the above is redundant. You can just check for ANON_CLASS_NAME. Which has obvious unintended consequences:

package foo object bar { class anonymize { } } // scala> afterFlatten(intp.terms("foo").info.members filter (_.isAnonymousClass)) // res0: List[$r.intp.global.Symbol] = List(anonymous class bar$anonymize)

Yep, it's pretty loose as it stands. I was trying not to rely on that quirk, but that at least warrants a comment.

retronym · 2012-07-12T07:49:12Z

@odersky Anonymous functions owned by a method are suffixed with the method name; this still happens after this patch (see the expanded test in 9629175).

The current scheme seems to stop short of being useful for serialization/JRebel: flat names are prefixed with owner chain, but the unflattened name is used to generate fresh names in flatten, meaning that the names of all anonymous classes and of all non-method-owned anonymous functions are sensitive to the overall position in the compilation unit.

I was also considering appending the name of the first parent class to the generated name of an anonymous class, e.g. $anonRunnable. (This is problematic for a parent class named fun, though!).

I'm going to close this PR, implement some refinements, and then discuss further on scala-internals. This can wait until after 2.10.

retronym added 2 commits July 8, 2012 22:54

A test case that scrutinises the results of the flatten phase.

d561f35

The following commit will change that phase to elide the noisy `$anon1$anon2` intermediaries from flat names of nested anonymous functions and classes.

Elide non top-level owners from flat name of $anon{Fun}

9689094

This avoids ugly class names like Foo$anon1$anon2$anon3 when defining anonymous functions or classes in a nested location.

Expand flatten test with method local anons.

9629175

Under the new scheme, they retain the method suffix added in SI-101.

ghost assigned paulp Jul 11, 2012

paulp reviewed Jul 11, 2012
View reviewed changes

retronym closed this Jul 12, 2012

retronym mentioned this pull request Apr 30, 2016

SI-3623 Reduce maxClassfileName default value to 128 #5088

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Shorter flattened names for nested $anons #852

Shorter flattened names for nested $anons #852

Uh oh!

retronym commented Jul 8, 2012

Uh oh!

scala-jenkins commented Jul 8, 2012

Uh oh!

scala-jenkins commented Jul 8, 2012

Uh oh!

retronym commented Jul 9, 2012

Uh oh!

scala-jenkins commented Jul 9, 2012

Uh oh!

scala-jenkins commented Jul 9, 2012

Uh oh!

scala-jenkins commented Jul 9, 2012

Uh oh!

scala-jenkins commented Jul 9, 2012

Uh oh!

paulp Jul 11, 2012

Uh oh!

retronym Jul 12, 2012

Uh oh!

retronym commented Jul 12, 2012

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Shorter flattened names for nested $anons #852

Shorter flattened names for nested $anons #852

Uh oh!

Conversation

retronym commented Jul 8, 2012

Uh oh!

scala-jenkins commented Jul 8, 2012

Uh oh!

scala-jenkins commented Jul 8, 2012

Uh oh!

retronym commented Jul 9, 2012

Uh oh!

scala-jenkins commented Jul 9, 2012

Uh oh!

scala-jenkins commented Jul 9, 2012

Uh oh!

scala-jenkins commented Jul 9, 2012

Uh oh!

scala-jenkins commented Jul 9, 2012

Uh oh!

paulp Jul 11, 2012

Choose a reason for hiding this comment

Uh oh!

retronym Jul 12, 2012

Choose a reason for hiding this comment

Uh oh!

retronym commented Jul 12, 2012

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants