`std::variant<std::string>{}.index()` is not an integral constant expression

[hewillk]

Clang rejects the following.
Not sure if it's just because a certain language feature isn't implemented yet, correct me if I'm missing something.

#include <variant>
#include <string>

static_assert(std::variant<std::string>{}.index() == 0);

https://godbolt.org/z/8cTnafsMd

[mordante]

I strongly suspect it needs this PR before it works #83335

[philnik777]

Yes, this is simply not implemented yet in libc++.

[huixie90]

Yes, this is simply not implemented yet in libc++.

I could argue that it is a bug in libc++. In #83335, I implemented a paper which added "constexpr" to some operations. However, the default constructor was constexpr since the beginning and we did not make variant<string> default constructor useable in constexpr context. (We only made trivial types usable in constexpr). In any case, I fixed it in #83335 and added a test for this particular example

[philnik777]

@huixie90 The destructor is non-trivial and not constexpr, making it a non-literal type. That's the reason this is rejected currently, and for that to work we have to implement P2231R1.

[ldionne]

Reopening since this is effectively valid AFAICT -- we don't implement that paper.

[philnik777]

@ldionne I'm a bit confused. We don't implement that paper, which we don't currently track as bugs. Why should this stay open?

[ldionne]

We don't track unimplemented papers as Github Issues indeed, however I wasn't under the impression that we'd go ahead and close any issue that tracks the implementation of a paper? Either way seems fine to me, but I thought it would make sense to close this when we merge #83335.