New CI with Github Actions

[simonlegrand]

Here is a first contribution to the new FreeFem CI.

5 different configurations of FreeFem are tested:

• Minimal version
• Sequential version (with all dependencies)
• Full version OpenMPI
• Full version MPICH
• Full version MS-MPI

Each of these workflows contains jobs for Debug and Release and different combinations of platforms.

Currently, the build step doesn't work for the following configuration:

• Sequential/Windows/Debug
• Full Version MS-MPI/Debug
• Full Version OpenMPI/MacOS-14/*

But most of the tests should fail because of the removal of the continue-on-error: true at the make check step (few tests fail for almost each config). It can be temporarily put back if you prefer.

At first, if you prefer to merge on a different branch, just add the branch name in the

on:
  push:
    branches:
      ..
      - <branch_name>

Finally, due to the particular development workflow (push to test), the commit history is particularly dirty and should probably be squashed before the merge. It can be retrieved here

Cheers,
Simon

[prj-]

Why do you need to install SuiteSparse explicitly when building PETSc? Can't you let PETSc install it? Also, there are a couple of parallel packages in the sequential installs (ParMmg, ParMETIS, to name a few).

[simonlegrand]

Bien vu, I stupidly copy paste former tests. I'm not completely familiar with all the dependencies, can you confirm that I didn't leave anything uselss?

[simonlegrand]

Thanks ;)

[simonlegrand]

Three previous comments fixed in 92027eb

[prj-]

It looks like the job Windows Server 2022 with --enable-debug is constantly failing. Are you trying to debug it? If not, I don't think it's worth to put in the Actions.

[simonlegrand]

Yes, we are waiting for a Windows VM prepared by the Inria IT to debug it interactively without the constraints of the Github Actions Runner (time limit, resources, ...).

[simonlegrand]

I commented the Windows/Debug tests combination in Sequential and Full version MS-MPI. We'll dig into the problem as soon we get access to the virtual machine.

[prj-]

All good on my end (modulo the merge conflicts).

[prj-]

Why do you need to update SuperLU in 3rdparty/getall? The last version is already downloaded by make petsc-slepc, there is no real need to update the legacy getall, is there?

[simonlegrand]

Sorry I pushed the wrong branch. The sequential version of FF, that uses 3rdparty SuperLU, failed to compile with gcc-15 (used by the Windows runner). Out of curiousity, I wanted to try more recent versions of SuperLU but it appears to fail for versions 6.0.1 and 7.0.1 as well, but for different reasons (see here). This is weird since PETSc uses the version 6.0.1and FF links ok against it, but I may have missed a flag or something.

Anyway, I guess the best thing to do for now is to add the -std=gnu17 if gcc-15 is detected. What do you think?

[prj-]

I would just disable SuperLU from the CI.

[simonlegrand]

Sorry for the late reply. I also removed SCOTCH from the sequential version on Windows, it had the same issue as SuperLU. Also, I removed the explicit call to 3rd/getall to download all the dependencies, since make is calling it anyway if the dependencies are not disabled (with --disable-).

[prj-]

If everything pass, I merge, OK for you?

[frederichecht]

yes , Ok pour moi.

…

Le 6 juin 2025 à 16:40, Pierre Jolivet ***@***.***> a écrit : prj- left a comment (FreeFem/FreeFem-sources#329) <#329 (comment)> If everything pass, I merge, OK for you? — Reply to this email directly, view it on GitHub <#329 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABLPNDCTJSPEPYR5NFNSL233CGR7DAVCNFSM6AAAAABT5NPYV2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDSNBZGQ3TKNRYGU>. You are receiving this because you are subscribed to this thread.