I think this is useful, I have hopes that clang will eventually give us opensource C and Fortran compilers for windows, especially the latter.

Can you compile cpython with clangcl on windows? I am a bit surprised we need to add our own ClangCL class to distutils.

It would indeed be useful to set up a windows CI run under 32-bit and 64-bit for this.

Can you compile cpython with clangcl on windows? I am a bit surprised we need to add our own ClangCL class to distutils.

There is an open ticket with a patch to allow it to build: https://bugs.python.org/issue33351

This said, distutils doesn't support clang-cl any more than distutils supports icc on windows. It is unfortunate that commands like CC={cl|icl|clang-cl} python setup.py develop don't work Windows like that do on Unix.

AKAICT this is why NumPy would need to directly support clang-cl (if it is a good idea).

How would this play out for clang on linux, since the new class is windows-specific?

clang on unix is trivial since you can use cc=clang python setup.py install. Windows doesn't honor cc variables.

What is the magical incantation to get runtests.py or setup.py to use clang on windows?

I used the same trick that is used for icc

python setup.py config --compiler=clang-cl build_clib --compiler=clang-cl build_ext --compiler=clang-cl install

See https://software.intel.com/en-us/articles/numpyscipy-with-intel-mkl

It is likely that I'm not doing it correctly.

build already accepts a compiler option and passes it on to the sub-commands, so it should be sufficient to run setup.py build --compiler=clang-cl. We could add a --compiler option to runtests.py that would get passed on to build

That would probably help. I think clang-cl might still need a few settings since it isn't 100% identical to cl.

@bashtage : I <3 that PR. If you're interested in extra tests, I can launch the full pythran recompilation based on this, we are in great need for such support, and currently rely on a dirty hack for clang-cl support.

I would like to finish this in the next month or so. Needs a good bit of cleaning I think.

Any news on this PR? :)

I hope another ping 4 months later does not constitute bad form. 😉

@mattip I've updated to reflect changes and this is working. I have also improved it. Any idea how I could take this forward?

I think it is there, although ideally it would get a CI run.

python runtests.py --compiler=clang-cl
.....................................................................................
.....................................................................................
.....................................................................................
10394 passed, 454 skipped, 108 deselected, 17 xfailed, 1 xpassed in 73.10s (0:01:13)     

Do we wish to continue working on this PR, given numpy.distutils has been officially deprecated?

The PR is eol. I think the good parts all got merged in other PRs.

Let's close this