Whoo- hoo!! - Can now build perl with optimization level -O2

[sisyphus]

I've just built current blead using gcc version 15.1.0 (MinGW-W64 x86_64-ucrt-mcf-seh, built by Brecht Sanders, r1), specifying -std=c23 in ccflags, and setting optimization level to a simple -O2.

>..\perl -I..\lib -V:ccflags
ccflags='-std=c23 -DWIN32 -DWIN64 -fdiagnostics-color=never -DPERL_TEXTMODE_SCRIPTS -DMULTIPLICITY -DPERL_IMPLICIT_SYS -DUSE_PERLIO -D__USE_MINGW_ANSI_STDIO -fwrapv -fno-strict-aliasing -mms-bitfields';

>..\perl -I..\lib -V:optimize
optimize='-O2';

All tests passed:

All tests successful.

Test Summary Report
-------------------
run/todo.t                                                         (Wstat: 0 Tests: 6 Failed: 0)
  TODO passed:   3-4
../ext/IPC-Open3/t/IPC-Open3.t                                     (Wstat: 0 Tests: 53 Failed: 0)
  TODO passed:   33
Files=2915, Tests=1348953, 1369 wallclock secs (38.27 usr +  6.72 sys = 44.98 CPU)
Result: PASS

I don't know if this success is attributable to the perl source, or to gcc-15.1, or to MCF threads, or to the C23 level - but it pleasantly surprised me, anyway.

I noticed that, in gcc-15.1, the libmcfgthread dll has been given a version boost from libmcfgthread-1.dll (gcc-14) to libmcfgthread-2.dll.
That change calls for a small patch to the win32/GNUmakefile (as presented in Perl/perl5@db172be), which will hopefully be merged before the release of 5.42.0..

[sisyphus]

I get the same good results using gcc version 15.1.0 (MinGW-W64 x86_64-ucrt-posix-seh, built by Brecht Sanders, r1) ... which would suggest that the success is not attributable to the use of MCF threading.

>..\perl -I..\lib -V:ccflags
ccflags='-std=c23 -DWIN32 -DWIN64 -fdiagnostics-color=never -DPERL_TEXTMODE_SCRIPTS -DMULTIPLICITY -DPERL_IMPLICIT_SYS -DUSE_PERLIO -D__USE_MINGW_ANSI_STDIO -fwrapv -fno-strict-aliasing -mms-bitfields';

>..\perl -I..\lib -V:optimize
optimize='-O2';

Next thing to eliminate is the -std=c23 switch.

[sisyphus]

Seems it's not attributable to the -std=c23 switch either.
I think that means that the success is due to gcc-15 ... though I haven't disproven the possibility that the fix is in the perl source.
Anyone have any thoughts or observations ?
(It's getting late over here, and enough fine King Valley red wine has already been consumed ;-)

[shawnlaffan]

That's good news about the optimiser setting.

My first suspicion is that it is due to a gcc-15 change. The change log lists two vectoriser changes for -O2 which may be relevant. https://gcc.gnu.org/gcc-15/changes.html#general

It could also be a change in perl causes it to no longer tickle the underlying gcc issue.

And I'm not sure if it impacts your experiments but gcc-15 also sets -std=c23 by default: https://gcc.gnu.org/gcc-15/changes.html#c

[sisyphus]

My first suspicion is that it is due to a gcc-15 change.

Good digging. I think that's the most likely candidate. (Thanks !!)
I've subsequently found that building perl-5.41.3 using gcc 15 (and with optimization of -O2) also works fine - which suggests that there has been no significant change (in this regard) to the perl source.

And I'm not sure if it impacts your experiments but gcc-15 also sets -std=c23 by default.

Yes - the main objective was to override the -std=c99 setting that GNUmakefile specifies.
You're right in that EXTRACFLAGS= works just as well with gcc-15 as EXTRACFLAGS="-std=c23", and the former is better if you want to ensure that the C level is set to the compiler's default even if gcc version is not 15.

In my case, the script will only ever be used with gcc-15.1.0.
I'll probably stick with EXTRACFLAGS="-std=c23" so that I can readily check the actual C level by running perl -V:ccflags, should I ever forget.