Issue 1144 new continuous integration

[LourensVeen]

This adds a whole bunch of CI workflows, one for each code, which run whenever that workflow is changed or when the code is updated. This also contains some technical fixes for various codes, where tests that are too resource intensive for the CI are disabled.

Some of the codes remain broken and need attention from an expert, but I think it's better to have a CI that tells you that things are in that state, than to not have one at all, so after I do some more cleanup I think we should merge this anyway and then go look for someone who can fix things.

Addresses #1144

[LourensVeen]

It turns out that getting working MPI and compilers with Homebrew and MacPorts isn't easy. I'd like to get this merged, so I'm going to remove those tests for the time being, so that we at least have per-code tests on Ubuntu and macOS, plus an all-up conda based install and test.

[LourensVeen]

Okay, I think this is ready to go now. Some of the codes are still failing consistently:

• evtwin on macOS with a precision issue (related to SeBa, MESA r2208 and GalactICS fail precision tests on ARM processors #1163?)
• MESA r2208 on macOS (SeBa, MESA r2208 and GalactICS fail precision tests on ARM processors #1163)
• MESA r15140 because Mesa-r15140: ensure makedepf90 configure does not get overwritten #1248 broke its build system (in work for amuse-mesa-r15140 build fails #1259)
• PeTar on macOS because of bad C++ (see PeTar fails to compile on recent Clang #1264)
• SeBa with a precision issue (SeBa, MESA r2208 and GalactICS fail precision tests on ARM processors #1163)

There are also still intermittent timeouts and infrastructure failures, probably because these tests are quite heavy, use MPI, and because this PR runs all of them simultaneously. It's better than it was though due to additional OpenMPI configuration and a switch to MPICH on the macOS side. In normal circumstances code-specific tests will run only for PRs touching that code.

This does not include weekly regression tests for macOS with Homebrew and MacPorts, because I had trouble setting those up correctly. You don't just install a compiler or an MPI library with these things and have a working setup. It would help if someone with more mac expertise could do a nice clean setup for that.

Let's merge this and continue to set up our custom runner, and also to fix the codes so that everything passes all the time and we don't have broken windows anymore.

[rieder]

Ok for me - not sure how #1248 broke MESA but this is not the right place to discuss about that

[LourensVeen]

Make doesn't work if you use spaces instead of tabs 😄

[rieder]

There seem to be a lot of non-CI changes (e.g. to Makefiles) in this PR, is that by design or by accident?

[rieder]

Make doesn't work if you use spaces instead of tabs 😄

🤦
(that's easy to fix though)

[LourensVeen]

By design, it includes some fixes for issues I ran into:

• Make aarsethzare, fractalcluster, galactics and krome use the Fortran compiler detected by autoconf, rather than its default, which doesn't always work
• Use LDFLAGS when linking adaptb's executable, to ensure libs are found correctly
• Skip overly expensive tests that don't run (well) on the CI in gadget2, hermite, mesa-r15140, petar, and the framework tests.
• Specify precision in galactics test, IIRC the default changed after this test was written
• Compile hermite_grx as C++11 explicitly to avoid compiler errors
• Fix C++ compile issue in mi6, register keyword is obsolete
• Use python -m pip install instead of pip install in tupan build for better reliability (and to make it work on the CI)
• Fix resource leaks and other small issue in the framework tests
• Fix automatic installation of mpi4py in a venv on macOS

and somehow a bunch of whitespace fixes in concurrent.py snuck in, sorry.

[rieder]

Looks good