Thanks for the feature! I've exhaustively reviewed the product code but haven't really looked at the test coverage yet; I'll take a look at the tests during the next iteration.

BTW shouldn't _Invoker_ret and _Unforced be moved to <functional> if they're used only by std::function and bind()?

I don't understand why some of the x86 tests are failing.

/azp run

I don't understand why some of the x86 tests are failing.

Two conditional noexcept bugs in Edge (the IntelliSense compiler based on EDG) that are C++17-or-later only. The reducing and filing was horrific, working around not too bad.

BTW shouldn't _Invoker_ret and _Unforced be moved to <functional> if they're used only by function and bind?

Yes, that makes sense to me.

I think it's ready for initial review

Mini-changelog for more tests for conditionally noexcept bind():

• Unwrapping a reference_wrapper is noexcept, but passing the unwrapped reference to the callable object might involve a possibly throwing conversion.
• Forwarding through a placeholder is noexcept, but passing the unbound argument to the callable object might involve a possibly throwing conversion.
• Nested bind() needs to sense whether the nested call might throw.

Mark PossiblyThrowingInt(int) as constexpr so we can construct possibly_throwing_int during constant evaluation; ref(possibly_throwing_int) is simpler than declval<reference_wrapper<PossiblyThrowingInt>>().

Thanks again for implementing this C++20 feature (over 3% of the remaining features, by paper count!). This will ship in VS 2019 16.8 Preview 1. 😺 😸 🎆