Differing behavior of std::type_index hash and equality from libc++ and libstdc++

[Skylion007]

Full background can be found on the linked discussion pybind/pybind11#4319 (comment) Essentially the behavior of std::type_index is different between the two standard libraries. This behavior seems like a bug in LLVM's implementation and requires us to be careful whenever storing std::type_index in a hashmap when using libc++;

[frederick-vs-ja]

Why is this considered a bug rather than a permissible implemenation divergence?

I think it's intended that libc++'s type_info can be ABI-compatible with other implemenations, but I'm not sure whether it's a goal for libc++'s type_info::hash_code to have the same results as others.

---

Oh, libc++'s type_info doesn't seem to handle the leading '*' in the mangled name, which is probably the bug.

[frederick-vs-ja]

Is this a duplicate of #59790? #34255 is possibly related.

[philnik777]

Closing for now, since we can't fix it until Clang is fixed, and it's a duplicate (of a clang bug).