Add is_clang_based compiler property

[alexrichert]

This PR adds a property to compilers called is_clang_based, and sets it to True for clang-based compilers. This will save us copying and pasting lists of clang-based compilers across recipes when clang-specific compiler flags etc. are needed.

[alexrichert]

FYI @bernhardkaindl

[pranav-sivaraman]

I like this! Do the IBM Clang compilers also apply? However, I think Spack doesn't recognize them at this moment?

[alexrichert]

@pranav-sivaraman I added the property with return value True for all the clang compilers present to the best of my ability, so if/when merged, the same should be done for any other clang-based compilers added in the future. I didn't see the clang version of the IBM XL compilers in the configurations, if that's what you're referring to specifically, so yes as far as I know those are not supported yet.

[haampie]

but what does it mean? :) you need to document how this is supposed to be used.

[alexrichert]

@haampie just added it to packaging_guide.rst, let me know if there's an appropriate place to put it in addition or instead.

[haampie]

Documenting its existence in insufficient. The question is what is guaranteed by compiler.is_clang_based.

[alexrichert]

Can you elaborate on what you mean by guaranteed, and where that documentation should go?

[alalazo]

Can you elaborate on what you mean by guaranteed,

Take for instance Apple Clang. It has explicitly disabled direct use of -fopenmp, while upstream clang obviously supports the flag. I guess other clang based compilers might have their own differences too? So, what is guaranteed to be valid for all these compilers when the function returns True?

Can you show where you would like to use this function?

[alalazo]

Having a link to #43357 (review) would have helped getting the context

[alexrichert]

Thanks for clarifying. Yes, that boost issue was the immediate motivation here. Other places where I think it's likely to be useful would be the few dozen packages that have some kind of -Wno-error= setting inside if self.spec.satisfies('%clang/oneapi/apple-clang/etc.') where it's almost certainly the case that the same setting will be needed for all Clang-based compilers. Or in the case of the hdf5 recipe for example, if spec.compiler.name in ["gcc", "clang", "apple-clang", "oneapi"] could be something like if spec.compiler.name=='gcc' or self.compiler.is_clang_based so to avoid keeping a running list of Clang-based compilers.

[sethrj]

Thanks, this will clean up a fair number of packages and prevent further errors.

It looks like rocm doesn't have a compilers test?

[alexrichert]

It appears not.