cc @zooba, @ethanhs

compared to https://github.com/zufuliu/llvm-utils/blob/master/clang/clang-cl-py3.diff, the change to _msvccompiler.py seems outdated, specifically, PLAT_TO_LLVM_TARGETS is changed to:

PLAT_TO_LLVM_TARGETS = {
    'win32': '-m32',
    'win-amd64': '-m64',
    'win-arm64': '--target=aarch64-pc-windows-msvc',
}

Neat! Thank you for pushing this forward, I've been working on the implementation for PEP 585 recently and haven't had time to complete things.

I believe the CPython CI is going to moving off of Azure pipelines and use Github actions if I am not mistaken, so perhaps using that would be better. (Though I'd wait for Steve to comment before making changes).

@zufuliu, I specifically used --target=x86_64-pc-windows-msvc because that's unambiguous and -m64 might mean msvc or mingw depending on the default target of the compiler.

@isuruf OK. I don't use clang-cl from mingw-w64 LLVM (from msys2 or https://github.com/mstorsjo/llvm-mingw)

there is bug in _msvccompiler.py (it's may be better to fallback to cl.exe when clang-cl.exe isn't found):

        if self.use_clang_cl:
            self.cc = _find_exe("clang-cl.exe")
        self.cc = _find_exe("cl.exe", paths)

and I added _find_llvm in _msvccompiler.py to find LLVM from registry (added by official LLVM installers, with MSVC as default target), so don't require the LLVM\bin folder to be added to %PATH%.

Thanks @zufuliu, fixed it. I didn't add the registry lookup, because I usually use clang-cl from a different source and I wanted to have a way to use the one from PATH even if there's a compiler installed from official sources with the registry key. So, how about checking in PATH and then falling back to the registry? Note that some visual studio installations also have a clang.exe inside.

Codecov Report

Merging #18371 into master will decrease coverage by 0.06%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master   #18371      +/-   ##
==========================================
- Coverage   82.19%   82.12%   -0.07%     
==========================================
  Files        1957     1955       -2     
  Lines      589224   584483    -4741     
  Branches    44428    44477      +49     
==========================================
- Hits       484334   480029    -4305     
+ Misses      95222    94807     -415     
+ Partials     9668     9647      -21     

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3d06953...79fe165. Read the comment docs.

I think PLAT_TO_LLVM_TARGETS can be updated to include 32-bit ARM, at least with LLVM 9.0.1 32-bit ARM target (--target=arm-pc-windows-msvc) is also supported by clang-cl, but there are other bugs (like https://bugs.llvm.org/show_bug.cgi?id=42711) prevents the usage.

There is a mistake in
target_flag = "{}-pc-windows-msvc".format(PLAT_TO_LLVM_TARGETS[plat_name])
it should be
target_flag = "--target={}-pc-windows-msvc".format(PLAT_TO_LLVM_TARGETS[plat_name])

I changed PLAT_TO_LLVM_TARGETS to

PLAT_TO_LLVM_TARGETS = {
    'win32': 'i686-pc-windows-msvc',
    'win-amd64': 'x86_64-pc-windows-msvc',
    'win-arm32' : 'arm-pc-windows-msvc',
    'win-arm64': 'aarch64-pc-windows-msvc',
}

so target flag can be written as
target_flag = '--target=' + PLAT_TO_LLVM_TARGETS[plat_name]

Thanks @zufuliu

Fixed merge conflicts. Ready for a review again.

Any progress on this?

If it is any encouragement, numpy/numpy#13816 builds NumPy and passes all tests using clang-cl.

@isuruf It seems this PR need to be submitted to https://github.com/pypa/distutils, distutils in stdlib is going to be removed.
See pypa/setuptools#2065 and https://bugs.python.org/issue41282

@isuruf Can you rebase your PR on main branch ?

I've already done that multiple times. I'll rebase after I get a review from someone who is interested in merging this.

Thanks for the PR, but I think some of this was done in #101352 and distutils has been removed from CPython main