Roy Jacobson

It seems that the CI doesn't pass because of some cmake/setup.py errors. Will look into it.

Hi @rwgk, thank you for the feedback! If I understand you correctly, then a relatively large refactor is needed in order to cleanly merge the smart holder branch. Is this...

> A bit of a tangent: the smart_holder branch could be merged as-is, it's an add-on, but it would double the number of CI jobs for full testing, which always...

@Skylion007 @henryiii, I'm waiting on an answer from you whether you still support this given @rwgk's objections. There seem to be a problem with the windows tests but I put...

> Slightly random, but the last couple weeks I familiarized myself more with > > * linking all pybind11 extensions statically (with core Python) > > * `-fvisibility=hidden` > >...

> I see, thanks for the explanation. (From past experience: I fear that's exactly what some people will try anyway, but if our stance is "That's your own problem" I'm...

@henryiii I gave up on the CMake installation code and fixed the remaining CI failures, so I think this is (finally) ready for review.

@henryiii @Skylion007 ping

That sounds suspiciously close to https://github.com/llvm/llvm-project/issues/56154 which is about concept normalization 'forgetting' the concept template arguments.

AFAICT the type requirement in the concept is a syntax error, as `template operator()` is not a valid identifier. MSVC and GCC accept, though.