bug#71916: 31.0.50; build failure on MINGW64

[Andrea Corallo <acorallo@gnu.org>]

Andy Moreton <andrewjmoreton@gmail.com> writes:

> On Wed 03 Jul 2024, Eli Zaretskii wrote:
>
>>> From: Andy Moreton <andrewjmoreton@gmail.com>
>>> Date: Wed, 03 Jul 2024 19:02:06 +0100
>>> 
>>> On Wed 03 Jul 2024, Andrea Corallo wrote:
>>> 
>>> > leuven65 <leuven65@gmail.com> writes:
>>> >
>>> >> I found the latest version of both master and emacs-30 branch (from Monday) failed to be built on MINGW64 / windows10.
>>> >> It failed when starting native compilation, error pops up "Cannot find libgccjit library" (from
>>> >> comp-ensure-native-compiler), maybe the called function "load_gccjit_if_necessary / init_gccjit_functions" return error.
>>> >>
>>> >> But if to revert the change introduced by branch "savannah/scratch/windows-98" (merged on 2024/06/30), the build becomes
>>> >> OK.
>>> >>
>>> >>  git revert 38179f85f8f7236d5de8d9a4333b6a1d4ded63f0
>>> >>
>>> >> So that maybe the issue is caused by the 3 commits in this branch.
>>> >
>>> > Hello leuven65,
>>> >
>>> > could you share the exact error you observe?
>>> >
>>> > Ccing Po Lo
>>> >
>>> > Thanks
>>> >
>>> >   Andrea
>>> 
>>> I've seen the same issue on emacs-30 and master branches. It seems to be
>>> that emacs cannot find the libgccjit DLL after dumping.
>>> 
>>> Running emacs under gdb, and using a breakpoint on `redraw-display':
>>> 
>>> (gdb) r -Q
>>> Starting program: C:\emacs\git\emacs\emacs-30\build\mingw64-x86_64-O2-native\src\emacs.exe -Q
>>> [New Thread 11584.0x27b4]
>>> [New Thread 11584.0x1f54]
>>> [New Thread 11584.0x1ec8]
>>> [New Thread 11584.0x2ad0]
>>> [New Thread 11584.0x1a60]
>>> 
>>> Thread 1 hit Breakpoint 2, Fredraw_display () at C:/emacs/git/emacs/emacs-30/src/dispnew.c:3222
>>> 
>>> (gdb) pp Vlibrary_cache
>>> ((gccjit) (gccjit . t))
>>> 
>>> It seems odd that there are two entries here.
>>
>> Thanks, I think I fixed this now on the emacs-30 branch.
>>
>>> There is code at the end of syms_of_emacs in emacs.c that changes
>>> Vdynamic_library_alist as a workaround for something similar, so have
>>> other changes made that ineffective ?
>>
>> No, the problem was that Vlibrary_cache was not re-initialized after
>> loading the pdumper file.
>
> Thanks Eli, confirmed as working again on emacs-30.
>
>     AndyM

Thanks closing.