Conversation
|
Great, thanks a lot for the fix and sorry for having introduced this
bug!
IMO, it would be more accurate to give this PR its distinct Changes
entry with proper author and reviewer.
And I guess this will have to be cherry-picked to 4.08.
|
|
No problem, @shindere! I don't think parallel builds worked for Cygwin before all the build system changes, either (and they definitely didn't for the native Windows build system!). For the This certainly does need to go in 4.08, yes - |
|
Re. Changes: I think that either approaches are reasonable, but I think @shindere's point was also that debugging this one issue was particularly delicate, and handling it separately better credits the effort evolved. (If I understand correctly, Jérémie was also involved in tracking it down.) I tend to agree with this idea of separating the most work-inducing changes on their own. |
|
OK - Changes entry moved. Those two commits can be squashed onto trunk and then cherry-picked to 4.08. |
damiendoligez
left a comment
There was a problem hiding this comment.
This patch is correct and must be cherry-picked to 4.08.
|
@dra27 CI is okay but there are conflicts in Changes. Would you mind doing the rebase, conflict resolution and squashing? Then I'd be happy to merge, unless you prefer to do it yourself at the same |
d0eeaff to
6e7c43c
Compare
|
Merged into trunk and cherry-picked to 4.08 as commit |
Cygwin should have
EXE=.exeinMakefile.config(it did with the old configure system).This causes a strange chain of events leading to a broken compiler: because of how Cygwin handles
.exe, thecamlheaderprograms end up with a.exeextension (NB this is Cygwin only - the native Windows builds remain correct, with no.exeextension on the camlheader programs). This causes a problem with #2041 sinceSys.file_exists "camlheader"will be true on Cygwin ifcamlheader.exeexists and is execuable butSys.readdirwill not containcamlheaderand so the programs built by the Cygwin builds ofocamlcwill be headerless.This is manifesting itself in testsuite failures - it obviously doesn't affect the build, because bytecode images are never executed directly.
cc @shindere