Overhaul Windows install docs #536
Conversation
MacOS X and Mac OS X changed to OS X.
tr is always installed.
Cygwin is a flavour of Unix. Coupled with the fact that it is built using the normal Unix configure route, it adds to the confusion of the Windows ports (especially the fact that the four native ports need a Cygwin environment to be compiled) including its instructions in the native port. Various out-of-date instructions also removed: Cygwin hasn't provided GCC 4.5.3 for a long time and FlexDLL can be recompiled from sources without alteration since version 0.32.
Since it does work and is being maintained!
Script intended to make the compilation instructions for the Microsoft Visual C++ Compiler ports a little simpler. This script allows the user to start Cygwin from the appropriate Visual Studio Command Prompt (e.g. by issuing `C:\cygwin\bin\mintty -`) and then, from the OCaml sources directory, issue ```bash eval `tools/msvs-promote-path` ``` This script emits an error if the Microsoft Linker cannot be found in any of the directories in `PATH` or if the `link` command already refers to the Microsoft Linker. If, as will usually be the default, `link` refers to the coreutils `link` command, `PATH` is altered to move the first entry containing the Microsoft Linker to the start of `PATH`. This has the effect of overriding the Unix interpretation of the `link` command, but this will never affect OCaml's build process. This script is taken from msvs-tools (https://github.com/metastack/msvs-tools) and re-licensed by the copyright holder.
Simplified the instructions by de-duplicating the 32/64-bit instructions now that every port is available for both architectures. Removed out-of-date Cygwin information. Included details on the new FlexDLL bootstrapping procedure.
|
/cc @msprotz FlexDLL bootstrapping may be of interest for the binary installer? |
|
Not sure why... I'm happy to use Alain's binary. Is there a reason why I On 4/8/2016 7:52 AM, David Allsopp wrote:
|
|
Potentially simpler build? It would also mean that the x64 distribution would have a 64-bit compiled version of flexlink, rather than a 32-bit one (see part of Alain's #388 (comment) for when this may be desirable). I did mean "may be" not "will be" though 😄 |
README.win32.adoc
Outdated
|
|
||
| make -f Makefile.nt flexdll world [bootstrap] opt opt.opt install | ||
|
|
||
| * `make -f Makefile.nt` will install FlexDLL by placing `flexlink.exe` (and the |
There was a problem hiding this comment.
Should this be make -f Makefile.nt flexdll?
There was a problem hiding this comment.
Oops - actually it should read make -f Makefile.nt install - it's describing the fact that a bootstrapped compilation also installs FlexDLL as part of OCaml.
|
Should msvs-promote-path be installed? (Yes, it might be weird to install a Cygwin specific script for a native install...) Alternatively, one could direct people to download the tool from your repository directly. |
|
Up to you - the only thing if it is installed would be whether it wants to be installed with that name (or prefixed |
|
Do you plan to release msvs-tools as a standalone package? If so, I think the best would be to refer to it as an (optional) dependency from the README (msvs-detect would also probably be useful), and let Windows binary packagers decide whether to bundle the tools with their installer. (Another approach for the "link" issue would be to address it within flexlink itself. This might be the most user friendly approach.) |
|
Where are we on this? If we want to have it in 4.03, there should be a clear consensus to merge very soon (as in, the next day or so). I would personally be tempted to just blindly trust @dra27 on the documentation changes, but I can't assess the risk of the new instructions making it worse for someone than the old. Any opinion? |
|
Sorry for the slow reply. |
|
I don't think we should re-release a version even if there is only a doc change. |
|
There is a confidence-test in the new instructions with the msvs-promote-path command which would be handy to have run by people with a system set-up that is manually configured "the old way" (@damiendoligez, @alainfrisch) - running |
|
Indeed, the ball would be in @alainfrisch and @damiendoligez 's court. Note that the plan for after the release is to switch to a shorter release cycle (as in: have another release in 2016), so missing this one wouldn't be as much of a problem. |
It works as advertised, except that |
dra27
force-pushed
the
overhaul-install-docs
branch
from
9526244 to
dcd2a09
Compare
|
Done (also squashed the typo commit and rebased) |
|
Congratulations on the release of 4.03.0! Just a small note, in case in the last-minute rush it's slipped through, that this needs merging on trunk. |
|
Thanks for the reminder, I just merged in trunk. |
remove emit_block_header_for_closure
…l#536) * Fix flambda_o3 and flambda_oclassic attributes The implementations of the `flambda_o3` and `flambda_oclassic` attributes were calling into `Clflags`, but the flambda backend implements optimization flags by circumventing `Clflags` entirely. The only obvious solution is to add a hook to `Clflags` so we can simply override what `Clflags.set_o3 ()` does. Fortunately, doing it this way means that the other two ways of setting the flags - the command line and `OCAMLPARAM` - no longer need to do anything special. Tested manually using the included .ml source file. Note that `[@@@flambda_o3]` does nothing if `-Oclassic` is on. This actually reflects existing behavior: `-Oclassic -O3` and `-O3 -Oclassic` are, as far as I can tell, effectively the same as `-Oclassic`. * Make later -O flags completely override earlier ones Also document the test more thoroughly. * Don't have -O* clobber earlier options Now `-flambda2-cse-depth 7 -O3` will do the same thing as `-O3 -flambda2-cse-depth 7`. * Minimize the diff a bit * Install optimization-flag handler in native toplevel * Restore the behavior that -Oclassic implies -linscan * Add missing settings to O levels The explicit setting of `fallback_inlining_heuristic` to `false` in the old `o2_flags` was redundant, but it's easier to review this way. * Formatting
* Fix flambda_o3 and flambda_oclassic attributes The implementations of the `flambda_o3` and `flambda_oclassic` attributes were calling into `Clflags`, but the flambda backend implements optimization flags by circumventing `Clflags` entirely. The only obvious solution is to add a hook to `Clflags` so we can simply override what `Clflags.set_o3 ()` does. Fortunately, doing it this way means that the other two ways of setting the flags - the command line and `OCAMLPARAM` - no longer need to do anything special. Tested manually using the included .ml source file. Note that `[@@@flambda_o3]` does nothing if `-Oclassic` is on. This actually reflects existing behavior: `-Oclassic -O3` and `-O3 -Oclassic` are, as far as I can tell, effectively the same as `-Oclassic`. * Make later -O flags completely override earlier ones Also document the test more thoroughly. * Don't have -O* clobber earlier options Now `-flambda2-cse-depth 7 -O3` will do the same thing as `-O3 -flambda2-cse-depth 7`. * Minimize the diff a bit * Install optimization-flag handler in native toplevel * Restore the behavior that -Oclassic implies -linscan * Add missing settings to O levels The explicit setting of `fallback_inlining_heuristic` to `false` in the old `o2_flags` was redundant, but it's easier to review this way. * Formatting
64235a3 flambda-backend: Change Float.nan from sNaN to qNaN (ocaml#466) 14a8e27 flambda-backend: Track GC work for all managed bigarray allocations (upstream 11022) (ocaml#569) c3cda96 flambda-backend: Add two new methods to targetint for dwarf (ocaml#560) e6f1fed flambda-backend: Handle arithmetic overflow in select_addr (ocaml#570) dab7209 flambda-backend: Add Target_system to ocaml/utils (ocaml#542) 82d5044 flambda-backend: Enhance numbers.ml with more primitive types (ocaml#544) 216be99 flambda-backend: Fix flambda_o3 and flambda_oclassic attributes (ocaml#536) 4b56e07 flambda-backend: Test naked pointer root handling (ocaml#550) 40d69ce flambda-backend: Stop local function optimisation from moving code into function bodies; opaque_identity fixes for class compilation (ocaml#537) f08ae58 flambda-backend: Implemented inlining history and use it inside inlining reports (ocaml#365) ac496bf flambda-backend: Disable the local keyword in typing (ocaml#540) 7d46712 flambda-backend: Bugfix for Typedtree generation of arrow types (ocaml#539) 61a7b47 flambda-backend: Insert missing page table check in roots_nat.c (ocaml#541) 323bd36 flambda-backend: Compiler error when -disable-all-extensions and -extension are used (ocaml#534) d8956b0 flambda-backend: Persistent environment and reproducibility (ocaml#533) 4a0c89f flambda-backend: Revert "Revert bswap PRs (480 and 482)" (ocaml#506) 7803705 flambda-backend: Cause a C warning when CAMLreturn is missing in C stubs. (ocaml#376) 6199db5 flambda-backend: Improve unboxing during cmm for Flambda (ocaml#295) 96b9e1b flambda-backend: Print diagnostics at runtime for Invalid (ocaml#530) 42ab88e flambda-backend: Disable bytecode compilers in ocamltest (ocaml#504) 58c72d5 flambda-backend: Backport ocaml#10595 from upstream/trunk (ocaml#471) 1010539 flambda-backend: Use C++ name mangling convention (ocaml#483) 81881bb flambda-backend: Local allocation test no longer relies on lifting (ocaml#525) f5c4719 flambda-backend: Fix an assertion in Closure that breaks probes (ocaml#505) c2cf2b2 flambda-backend: Add some missing command line arguments to ocamlnat (ocaml#499) git-subtree-dir: ocaml git-subtree-split: 64235a3
This adds syntax completion and syntax error highlighting through merlin-js. Merlin is introduced in a separate worker that is initialized asynchronously so as to not delay the first paint. We updated the code-mirror bindings to support the features required by merlin-js. The existing top-level worker is not initialized after the first paint. Co-authored-by: Thibaut Mattio <[email protected]>
6 participants
INSTALL.adocREADME.win32.adoctoINSTALL.adoctools/msvs-promote-pathto simplify the environment set-up for building MSVCSorry that this is so close to the release date - though no code is touched.
Alain hasn't released FlexDLL 0.35 yet - when he does, the git submodule should be moved forwards to that commit both in
trunkand4.03.