py-libclang: setup.py renamed

[adamjstewart]

Extracted out of #39

@rbberger @wdconinc @haampie @skosukhin @trws

Next problem is that the package calls spec['llvm'].libs, which calls llvm-config, which is not installed in the CI image because we only installed clang/flang. spack/gitlab-runners#67 is one potential fix, but not very robust. Ideally, there would be some way to say that py-libclang requires llvm+dev and Spack would install it itself. Is there any variant or CMake flag that controls whether llvm-config or the actual library files get installed?

[wdconinc]

Looks good to me. Sorry to have missed this in the recent upgrade of libclang.

[wdconinc]

Next problem is that the package calls spec['llvm'].libs, which calls llvm-config, which is not installed in the CI image because we only installed clang/flang. spack/gitlab-runners#67 is one potential fix, but not very robust. Ideally, there would be some way to say that py-libclang requires llvm+dev and Spack would install it itself. Is there any variant or CMake flag that controls whether llvm-config or the actual library files get installed?

Maybe another approach is to prevent detecting llvm as external when llvm-config is missing. This isn't the only case where llvm-config is used. And it's not the only case of a spack package requiring multiple OS packages to satisfy the assumptions (rust = rustc + cargo).

[adamjstewart]

I think we still have to detect it in order to use clang/flang.