build gfortran in macos_arm

[svigerske]

This adds a build of gfortran to the macos_arm job.
The libs build there should require only macOS 11.

I don't see where you redistribute the libs, though. Maybe I didn't really understood the original problem.

The "solution" from this PR is of course not nice. It makes the job run very long. Feel free to just close it.

Here is a log from running the job: https://productionresultssa10.blob.core.windows.net/actions-results/55f4b8b7-2851-435f-813e-2455e24dc47e/workflow-job-run-fcf18aec-d6fd-5643-497a-0979a2df1282/logs/job/job-logs.txt?rsct=text%2Fplain&se=2024-10-03T08%3A44%3A56Z&sig=Y09j4ZyqnR4e%2B%2BFxvi0U9JsKnrw0jFm224z1il4QvuQ%3D&ske=2024-10-03T17%3A39%3A19Z&skoid=ca7593d4-ee42-46cd-af88-8b886a2f84eb&sks=b&skt=2024-10-03T05%3A39%3A19Z&sktid=398a6654-997b-47e9-b12b-9515b896b4de&skv=2024-08-04&sp=r&spr=https&sr=b&st=2024-10-03T08%3A34%3A51Z&sv=2024-08-04
(It looks like the SCIP build failed, but the SCIP tests still succeeded.)

[Opt-Mucca]

A 700k line log file. Christ.

After having a look: I assume the build fail line is some weird GitHub runner magic, given the error is about some thread magic, and the tests and installation both then work afterwards.

I am not sure I'd want this in the build process though. The two hour plus build time is a bit long (assuming all the warnings can be suppressed, otherwise a 700k log is also too long). If this is unavoidable and we want to continue using the builds here for the Python releases then we have to accommodate this. Will think on it.

What we do with the libs: The final job uploads some artifact. The libs of each build inside of the artifact are then put into a release of this repo, with the libs then being downloaded by the build wheel job in PySCIPOpt.

[svigerske]

But I didn't see where you pack the gfortran libs. If they aren't redistributed, then I wonder why there is a problem that they are for macOS 14 (or higher).

[mmghannam]

How about we separate building gfortran to its own pipeline and add the generated binaries to a release and the other pipelines would grab that. Or use a caching mechanism in the pipelines.

We should probably do that for other dependencies too.

[Opt-Mucca]

But I didn't see where you pack the gfortran libs. If they aren't redistributed, then I wonder why there is a problem that they are for macOS 14 (or higher).

Ahhhhhhhhhhhhhhhh. I am truly dumb. You're right. This would need to be done in the wheel packaging (where cibuildwheeldoes try and sneak in libgfortran) For that I need to understand cibuildwheel a bit better.

For separating gfortran: I will look into if there are more options in the parameter settings for now. I still don't understand how this is not affecting everyone else in community with a gfortran dependency.