feature/spaceship: Clause 21: Strings

[AnjuDel]

No description provided.

[miscco]

Welcome to the team!

I could only find a readability nit

[AnjuDel]

Welcome to the team!

Thanks!

[CaseyCarter]

Just one comment change, which I'll go ahead and apply.

[StephanTLavavej]

I've pushed a batch of changes:

• Remove outdated test comment (unrelated to other changes).
  • This test has grown far beyond testing only spaceship for containers.
• Fix/test char_traits::comparison_category.
  • We need to modify _WChar_traits and _Narrow_char_traits, because they inherit from private _Char_traits.
• C++20 needs only one _Identity_t overload.
  • This was indicated by an edited example in the paper.
  • I've guarded the extra overload for operator== with #if !_HAS_CXX20, and simply dropped the extra overload for operator<=>.
  • The choice is arbitrary, but to match the Standard's example, I'm keeping the overload that takes _Identity_t on the RHS.
• Spaceship should take basic_string_view by value, like other operators.
  • Also matches the Standardese.
• Simplify syntax with an alias template.
  • It's repeated 8 times, so _Get_comparison_category_t significantly reduces verbosity.
  • The helper now emits type (to match _t) instead of comparison_category.
• Implement LWG-3432 "Missing requirement for comparison_category".
  • There's a "Mandates" that we need to enforce with _Is_any_of_v.
• Add SAL and top-level const for consistency.
  • I am still not a fan of SAL, but _In_z_ is probably the most useful, and we use it consistently in string.
• Avoid negated _HAS_CXX20 condition.
  • We conventionally prefer to avoid negated conditionals (preprocessor, if, etc.) when doing so doesn't complicate the code. This makes the code easier for our human brains to understand.
  • As a bonus, this makes it easier to see the symmetry between the == and <=> used for C++20.
• Test string <=> ptr and ptr <=> string.
  • string <=> ptr is implemented by a separate overload, so we need to test that. We expect the compiler to handle rewriting, but string is sufficiently complicated and important that spending 3 lines to test ptr <=> string is reasonable.
• Test string_view at runtime and compiletime.
  • This could be refactored in the future for less verbosity (using our dual-mode runtime/compiletime pattern), but let's go with this for now.