`<utility>`: `pair::swap(const pair&)` interacts badly with `__declspec(dllexport)`

[StephanTLavavej]

This is a regression caused by merging #2687 in VS 2022 17.3 Preview 2:

C:\Temp>type meow.cpp

#include <utility>
using namespace std;

struct __declspec(dllexport) MyPair : pair<int, int> {};

C:\Temp>cl /EHsc /nologo /W4 /MD /c /std:c++20 meow.cpp
meow.cpp

C:\Temp>cl /EHsc /nologo /W4 /MD /c /std:c++latest meow.cpp
meow.cpp
C:\Program Files\Microsoft Visual Studio\2022\Preview\VC\Tools\MSVC\14.33.31629\include\utility(107): error C2672: 'swap': no matching overloaded function found
C:\Program Files\Microsoft Visual Studio\2022\Preview\VC\Tools\MSVC\14.33.31629\include\type_traits(1965): note: could be 'void std::swap(_Ty (&)[_Size],_Ty (&)[_Size]) noexcept(<expr>)'
C:\Program Files\Microsoft Visual Studio\2022\Preview\VC\Tools\MSVC\14.33.31629\include\utility(107): note: 'void std::swap(_Ty (&)[_Size],_Ty (&)[_Size]) noexcept(<expr>)': could not deduce template argument for '_Ty (&)[_Size]' from '_Ty'
        with
        [
            _Ty=int
        ]
C:\Program Files\Microsoft Visual Studio\2022\Preview\VC\Tools\MSVC\14.33.31629\include\type_traits(1965): note: see declaration of 'std::swap'
C:\Program Files\Microsoft Visual Studio\2022\Preview\VC\Tools\MSVC\14.33.31629\include\type_traits(1962): note: or       'void std::swap(_Ty &,_Ty &) noexcept(<expr>)'
C:\Program Files\Microsoft Visual Studio\2022\Preview\VC\Tools\MSVC\14.33.31629\include\utility(106): note: 'void std::swap(_Ty &,_Ty &) noexcept(<expr>)': could not deduce template argument for '_Enabled'
C:\Program Files\Microsoft Visual Studio\2022\Preview\VC\Tools\MSVC\14.33.31629\include\type_traits(1962): note: see declaration of 'std::swap'
C:\Program Files\Microsoft Visual Studio\2022\Preview\VC\Tools\MSVC\14.33.31629\include\utility(408): note: see reference to function template instantiation 'void std::_Swap_adl<const _Ty1>(_Ty &,_Ty &) noexcept(false)' being compiled
        with
        [
            _Ty1=int,
            _Ty=int
        ]
C:\Program Files\Microsoft Visual Studio\2022\Preview\VC\Tools\MSVC\14.33.31629\include\utility(406): note: while compiling class template member function 'void std::pair<int,int>::swap(const std::pair<int,int> &) noexcept(false) const'
meow.cpp(4): note: see reference to class template instantiation 'std::pair<int,int>' being compiled

Relevant code:

STL/stl/inc/utility

Lines 331 to 339 in 2263d93

	#if _HAS_CXX23
	constexpr void swap(const pair& _Right) const
	noexcept(is_nothrow_swappable_v<const _Ty1>&& is_nothrow_swappable_v<const _Ty2>) {
	if (this != _STD addressof(_Right)) {
	_Swap_adl(first, _Right.first);
	_Swap_adl(second, _Right.second);
	}
	}
	#endif // _HAS_CXX23

WG21-N4910 [pairs.pair]/45.2 says:

Mandates: For the second overload, is_swappable_v<const T1> is true and is_swappable_v<const T2> is true.

So we aren't supposed to constrain this, but that interacts badly with dllexport which tries to instantiate everything.

I suspect that we need the template <int = 0> workaround, also applied to tuple.

Originally reported as DevCom-10109884 and internal VSO-1586226 / AB#1586226 .

[CaseyCarter]

Mandates: For the second overload, is_swappable_v<const T1> is true and is_swappable_v<const T2> is true.

So we aren't supposed to constrain this, but that interacts badly with dllexport which tries to instantiate everything.

It's permissible but not required to implement a "Mandates" element via constraints as long as an expression that would call this overload won't inadvertently result in a call of some other overload. (So yes we can do this and it's conforming.) The Standard only cares that calls produce a diagnostic when the requirements are not met, it doesn't care if the diagnostic is pretty.

[StephanTLavavej]

I'll fix this with template <int = 0> because adding a requires is more verbose and ran into other issues.

[jwakely]

For libstdc++ we've added requires is_swappable_v<T...> because our tests that explicitly instantiate pair and tuple were failing.

Since you need something similar, maybe it should be an LWG issue.