Extending memcmp optimization to classes

[AdamBucior]

With C++20 it's possible to default comparison operators. With has_unique_object_representations and an intrinsic to detect if type is trivially equality comparable (it has defaulted operator== and all of it's members are trivially equality comparable) we could greatly increase number of cases when equal uses memcmp.

It would be great if we could default operator== for types like tuple and pair, but I don't know if we can do that.

[StephanTLavavej]

I think it might be possible to default equality for pair, but not for tuple which permits different types to be compared.

[CaseyCarter]

I think it might be possible to default equality for pair, but not for tuple which permits different types to be compared.

pair's comparisons can't be generally defaulted since it allows reference members and "A defaulted <=> or == operator function for class C is defined as deleted if any non-static data member of C is of reference type or C has variant members." ([class.compare.default]/2).

That said, I've been thinking about implementing "has no fields of reference type" partial specializations for pair and tuple in ABI broken world so their special member functions could be defaulted. Such a partial specialization of pair could also default its comparisons if those comparisons were specified as members or hidden friends rather than vanilla non-members.

[AlexGuteniev]

This is partially done in #5527.

We may be able to do this also for pair and tuple without ABI break.