On Wed, 3 Sept 2025 at 10:21, Ville Voutilainen <ville.voutilai...@gmail.com> wrote: > > On Wed, 3 Sept 2025 at 12:12, Jonathan Wakely <jwak...@redhat.com> wrote: > > > I want to also see following test being added, that explains why we want > > > to use: > > > std::forward<_Tp>(__p.second); > > > Instead of, seemingly more obvious: > > > std::move(__p).second -> this will produce int& instead of int&& > > > > Good idea. I think I should probably just extend the testcase to cover > > every combination of lvalue/rvalue members, lvalue/rvalue pairs, and > > const/non-const pairs. > > If you could, also take a look at similar coverage for assignment > operators and constructors. There was that recent > misguided suggestion to replace forwards with moves; our > implementation uses forward in assignments and constructors, > but surely doesn't hurt to try to make sure the tests for those cover > reference elements.
We have testsuite/20_util/pair/ref_assign.cc (and also pass all libc++ tests, with the two pending patches I've submitted for std::pair).
