gcc-15 fixes: segfault on fpm install (#1149) and metapackages

[krystophny]

This is a small fix for fpm install segfaulting when compiled with GFortran 15.1.1 on Archlinux. It also adds a workaround for failing compilation on -Wall due to upstream GCC bugs https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120470 / https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119928 that have been fixed in git but exists up to including 15.1 .

[krystophny]

@perazz this is ready to review. Unfortunately due to the upstream bug in GCC 15, we cannot pass bootstrapped tests, so I didn't add it in the CI/CD yet. The bump to GCC default version 15 in Homebrew independently causes the mpich and openmpi to fail on macos, but this is another issue. This one here is ready to merge.

[perazz]

Thank you for your fix @krystophny.

I was a bit hesitant to add the no-interface-checking flag to the debug builds (though I understand it's necessary to build code with the buggy gcc-15 - see #1145 and fortran-lang/test-drive#49).

However, when the option is not recognized (gcc<=14), gcc only errors out for the positive form (-Wexternal-argument-mismatch) but not the negative form (-Wno-...). gcc.gnu.org

So I think it's a good practical fix imho.

[perazz]

I fixed the metapackage CI steps for gcc-15, but there are now more errors (tests not passed by fpm), it seems related to paths. I will investigate as soon as possible.

[krystophny]

Thank you for your fix @krystophny.

I was a bit hesitant to add the no-interface-checking flag to the debug builds (though I understand it's necessary to build code with the buggy gcc-15 - see #1145 and fortran-lang/test-drive#49).

However, when the option is not recognized (gcc<=14), gcc only errors out for the positive form (-Wexternal-argument-mismatch) but not the negative form (-Wno-...). gcc.gnu.org

So I think it's a good practical fix imho.

@perazz these are exactly my thoughts. We are kind of forced to do this, since there is a realistic risk that major distributions or cluster environments will go with GCC 15.1 . Honestly, I am disappointed by GFortran how multiple issues like this could happen. The option -Wexternal-argument-mismatch didn't exist before, so it will do no harm to add the negative one. What we could do is limit the fix to some GCC versions. For this I create a new issue.

[perazz]

@krystophny I've opened a PR (krystophny#1) towards your branch to avoid duplication.

Unfortunately it seems like array reallocation with the array constructor is broken in gcc-15 (a = [a, b]), that is what was causing the failing metapackage tests.

Please look at that one and once that is merged into this one, we can merge this upstream too imho.

[krystophny]

@krystophny I've opened a PR (krystophny#1) towards your branch to avoid duplication.

Unfortunately it seems like array reallocation with the array constructor is broken in gcc-15 (a = [a, b]), that is what was causing the failing metapackage tests.

Please look at that one and once that is merged into this one, we can merge this upstream too imho.

done!

[perazz]

Merging, thank you @krystophny