So far I got it to compile, so I guess the depencies are fine now. But running the tests raises some HDF5 related errors. I've never worked with HDF5 in C++, so maybe @sanathkeshav could you please have a look at the errors?

The tests fail with both hdf5 and hdf5-mpi brew packages. They can't be installed at the same time:

Error: Cannot install hdf5-mpi because conflicting formulae are installed.
  hdf5: because hdf5-mpi is a variant of hdf5, one can only use one or the other

Since we could not come to a consensus on how to proceed with this issue, I would propose to close this PR at this point.
Please let me know what you think @siggmo @IshaanDesai

Since we could not come to a consensus on how to proceed with this issue, I would propose to close this PR at this point. Please let me know what you think @siggmo @IshaanDesai

Fine for me. If you don't need the MacOS support at the moment, I don't think we need to invest more time here.

The macOS test fails with the errors:

/Users/runner/work/FANS/FANS/include/solver.h:364:68: error: use 'template' keyword to treat 'get' as a dependent template name
    const std::string &measure = reader.errorParameters["measure"].get<std::string>();
                                                                   ^
                                                                   template 
/Users/runner/work/FANS/FANS/include/solver.h:390:68: error: use 'template' keyword to treat 'get' as a dependent template name
    const std::string &error_type = reader.errorParameters["type"].get<std::string>();
                                                                   ^
                                                                   template 

which seems to have some solutions, for example as stated in this post. I am no expert in C++ templates, but could this error occur because we now enforce a C++ standard

I would suggest trying to resolve the errors above, because having a macOS test is nice, and @siggmo already invested effort here, so it would be a pity to waste it now.

Okay, now this works natively on my M1 Macbook, but it works because we are not using the clang compilers but gcc-14. This also involved compiling open-mpi itself with gcc-14 via source. This is painfully slow (takes ~11 mins in CI).

• I think we should create a MacOS docker image and use that container for the CI like we do for jammy and noble.
• We should also maybe go back to the conda recipe and enable MacOS support.

But we should put a pin on the Docker image for the macOS CI for later.

I will open an issue.