Patch seems reasonable given the WIP status of GFortran on arm64 and the patch is what upstream recommends at the moment. The diff could be somewhat smaller and easier to follow if you just remove the lines rather than replace with commented out lines though.

Patch seems reasonable given the WIP status of GFortran on arm64 and the patch is what upstream recommends at the moment. The diff could be somewhat smaller and easier to follow if you just remove the lines rather than replace with commented out lines though.

I agree. I left the lines commented out to clearly indicate that this is a temporary fix. I can delete them if that is the preferred approach.

I think the comment given above the patch is sufficient for showing that the patch is temporary. This is where people look first for understanding whether the patch is still necessary.

🤖 A scheduled task has triggered a merge.

Was it a deliberate change to flip from clang + gfortran to gcc + gfortran in the formula? It seems like this goes against the usual policy of using the apple toolchain everywhere possible.

Superenv shims should mean that didn't actually happen (gcc-10 can actually invoke clang in the superenv), but I think the CC/CXX stuff should be unnecessary anyway and can be removed.

Superenv shims should mean that didn't actually happen (gcc-10 can actually invoke clang in the superenv),

I think I don't understand what this means.

What I observed was that mpicc -show used to say clang ... and now it says gcc-10 .... So any flags that were configured in clang mode now didn't work.

Oh right, that part could indeed be broken. I was just talking about the actual build of mpich itself rather than runtime usage of mpicc.

Can we reopen this MR or shall I create a new one?

You can't reopen a merged pull request so it will have to be a new one.

OK see mpich: switched back to clang #73269