<mdspan>: Using std::mdspan results in multiple warnings (C5246)

[dasmysh]

Describe the bug

When just initializing an std::mdspan with dynamic extents multiple warnings (C5246) are produced during compilation.

Command-line test case

C:\Temp>type mdspan_testcase.cpp
#include <array>
#include <mdspan>

int main()
{
    std::array test_data{0, 1, 2, 3, 4, 5};
    std::mdspan<int, std::dextents<std::size_t, 2>> test_span(test_data.data(), 2, 3); // mdspan(329,54): warning C5246 + mdspan(124,50): warning C5246
    return 0;
}

C:\Temp>cl /EHsc /Wall /WX /std:c++latest .\mdspan_testcase.cpp
[...]

multiple C5246 warnings occur.

Expected behavior

Compiles without warnings.

STL version

```
Microsoft Visual Studio Community 2022 (64-Bit) - Preview
Version 17.10.0 Preview 1.0
```

[StephanTLavavej]

In general, the STL doesn't attempt to be /Wall clean, only /W4 (and /analyze). We sometimes fix off-by-default warnings on a case-by-case basis though, especially when they appear to be more valuable than noisy.

In this case, I can't immediately see what code is triggering these warnings; the line numbers don't make sense to me.

[frederick-vs-ja]

The warning actually comes form here.

STL/stl/inc/mdspan

Lines 121 to 124 in d6efe94

	template <class _ExtsTuple, size_t... _Indices>
	requires (tuple_size_v<_ExtsTuple> == _Rank_dynamic)
	constexpr explicit extents(_Extents_from_tuple, _ExtsTuple _Tpl, index_sequence<_Indices...>) noexcept
	: _Base{static_cast<index_type>(_STD move(_STD get<_Indices>(_Tpl)))...} {}

The implementation currently depends on brace elision in aggregate initialization, but I think this should be OK. IMO warning C5246 doesn't make much sense.

[dasmysh]

IMO warning C5246 doesn't make much sense.

So you would say this is a compiler issue? I had a feeling it would be, but since the std::mdspan implementation is rather new I wanted to check here first. I would raise it there then.

[frederick-vs-ja]

So you would say this is a compiler issue?

IMHO warning on "the initialization of a subobject should be wrapped in braces" is itself a bad design. Such warning discourages initializing std::array in a standard-guaranteed way ([array.overview]/2).

At least I think we should suppress the warning in <mdspan>.

[StephanTLavavej]

We talked about this at the weekly maintainer meeting. While the STL doesn't attempt to be /Wall clean in general, we think it would be reasonable to add an STL-wide suppression for C5246. Our rationale is that we can't remember ever having incorrect code in the STL caused by brace elision, so the warning seems exceptionally low-value for us.

The place to do so is around here:

STL/stl/inc/yvals_core.h

Line 786 in 243cffc

// warning C4514: unreferenced inline function has been removed (/Wall)

Note that we capture the exact warning text, followed by "(/Wall)" for off-by-default warnings.