Could we yoink the logic from python.mk for this?

Patch looks fine. Approved for commit once you feel like you have tested it sufficiently.

There seems to be a stray gcc in the file.

Maybe both actually...

I recommend that instead of !defined(__cplusplus), maybe we do !defined(unreachable). This should also cover C codebases with their own unreachable macros and matches what we do for offsetof.

In D53975#1233980, @haraldei wrote:

In D53975#1233152, @fuz wrote:

For testing, the gold standard is to build the port and execute the test suite on as many platforms we support as possible, on all currently supported OS versions (that is, 13.5, 14.3, and 15.0).

Do I have access to, and can I use the ref*-machines for that?

I have run the tests on amd64, natively on my dev machine, and can probably get the aarch64 emulation up and running too soon. But it would be nice to be able to at least run the tests on real hardware too.

In D53831#1229909, @fuz wrote:

The point of this test is to check that I got the definitions right, as we do them manually. We could actually use compiler builtins for the various sys/$ARCH/include files to directly grab the types from the C compiler, but we don't.

But didn't you propose making exactly that change?

For testing, the gold standard is to build the port and execute the test suite on as many platforms we support as possible, on all currently supported OS versions (that is, 13.5, 14.3, and 15.0). If you don't have access to some architectures, you can skip testing there, but especially for an infrastructure port like this it is a good idea to be thorough.

LGTM.

Looks reasonable. Should do the same thing as before, literally.

Looks reasonable.

This looks reasonable. I can't say if the conflict is correct, that's for you to decide. What would go wrong with multiple flavours installed at once?