[libc++] Implement `operator<=>` for `unique_ptr`

[avogelsgesang]

Summary

Implements part of:

• P1614R2 The Mothership has Landed

Fixes LWG3426

[avogelsgesang]

Fix tests on older C++ versions

[avogelsgesang]

need to update based on comments in https://reviews.llvm.org/D130852

[avogelsgesang]

Update based on feedback from D130852

[Mordante]

Thanks! Looks quite good overall, but a few remarks.

libcxx/include/__memory/unique_ptr.h

601

712

libcxx/include/memory

509

Interesting these aren't removed, including in the WP.

libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.special/cmp.pass.cpp

73

Can you also test with the other two comparison categories? Not 100% sure it's even possible.

libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.special/cmp_nullptr.pass.cpp

22

Nice catch! It seems we've added _NOEXCEPT in the code. @ldionne should we remove these noexcepts in a separate commit?

55

[avogelsgesang]

address review comments

libcxx/include/memory

509

I agree. The revision history of P1614R2 states

The comparisons between unique_ptr<T, D> and nullptr were originally removed and replaced with a <=>, but this was reverted.

But I don't know where to find the history on why that change was made

libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.special/cmp.pass.cpp

73

I don't understand.

unique_ptr <=> unique_ptr is specified to return std::strong_ordering. If I added, e.g.,

AssertOrderReturn<std::weak_ordering, std::unique_ptr<int>>();

this test would fail because the result of unique_ptr <=> unique_ptr is strong_ordering, not weak_ordering

[Mordante]

LGTM, thanks!
I'll commit this on your behalf.

libcxx/test/std/utilities/smartptr/unique.ptr/unique.ptr.special/cmp.pass.cpp

73

After I wrote the question I wondered whether it would be possible at all. The odd part is that std::shared_ptr uses strong_ordering, but unique_ptr deduces the ordering. But it seems it indeed should always be strong_ordering.