Mac os CI pipeline

[schnellerhase]

Adds a new mac OS CI pipeline, that should be fast enough to run with every commit/PR.

• The CI run both a petsc-free and a build with petsc using the brew package

• Currently we can not verify the python package in the petsc build, as I have no idea how to make petsc4py work properly with the brew package (Note: same holds true for the debian packages and thus restricts the running with debian petsc install)

• There are two failing python tests in the petsc free run, not sure what's going on there, and whats causing this. Needs to be fixed before merge.

Fixes #3312

[jhale]

Looks good. I think this could be an adjustment of the existing macOS pipeline; you could have a variable defining the type of PETSc build, e.g. "none", "brew" or "self".

[schnellerhase]

I couldn't figure out how to maintain both the scheduling of the other macos pipeline and allow the non scheduled new runs in the same file.

[schnellerhase]

The failing tests are fixed now, was caused by cffi_utils failing if petsc4py is not importable before registering numba types.

[garth-wells]

Rather than using Homebrew, how about using conda to install dependencies? The conda PETSc package has more options enabled, e.g. PETSc is configure with parallel LU solver. And petsc4py is available.

See #3319.

[jhale]

I agree with @garth-wells , homebrew's petsc is not really useful for our users so it seems a waste to test against it. Although good to see it's now being built against HDF5 MPI!

I would propose:

• conda petsc (on commits)
• no petsc (on commits)
• homebrew + petsc build (weekly)

[schnellerhase]

I can not get the basix build to pass, pretty sure its something to do with the availability of a C++ standard library, however I'm not really sure how to make that available in the conda setting. See here for the basix standalone failure.

[schnellerhase]

The error I get is

  Undefined symbols for architecture arm64:
    "std::exception::what() const", referenced from:

Never worked on ARM before - do you guys know what else I need to change to make that work?

[schnellerhase]

Okay I guess we have no real fix for this problem for now, so should we return to the case of the brew install and get this merged for now to have a macos pipeline running at all?

Also we could in continuation of #3335 make use of the sub-optimal setting of being able to make petsc run and not petsc4py by compiling the cpp library with petsc support and the python one without support of petsc4py and then run the tests. This is so far not tested in any CI.

[schnellerhase]

The cffi fixes here we should maybe extract and merge independent of the macos pipeline. However just noted they are broken, got lost in some merge conflict.

[jhale]

The cffi fixes here we should maybe extract and merge independent of the macos pipeline. However just noted they are broken, got lost in some merge conflict.

They are at acd43c7 - could you make a new PR with just this?