<vector>: Avoid a compiler warning with Defect Report P2280R4

[StephanTLavavej]

@Codiferous is implementing WG21-P2280R4 in MSVC. With this Defect Report implemented, MSVC emits its (extremely annoying) warning C4127 "conditional expression is constant" for vector<bool>::max_size(), because we marked a local variable as const and that causes "trial evaluation" to happen. With WG21-P2280R4, _Ints_max is recognized as a compile-time constant, causing the if-statement below to emit the warning.

Unfortunately, the underlying vector's max_size() has to be implemented in terms of the allocator's max_size(), and there's no requirement that it be a compile-time constant. So we can't just boil this whole thing down to an if constexpr.

To avoid this, we can extract the condition into a const bool and add a comment why. C4127 has a special case to avoid warnings when a single variable is being tested.

(I am somewhat tempted to globally suppress C4127 in STL headers, but it hasn't been quite enough of a nuisance to resort to that... yet.)

[AlexGuteniev]

Can the usual way work: extract condition to const bool and then test that bool ? If so, this would avoid bloating debug codegen

[StephanTLavavej]

I believe that extracting the condition to a const bool would avoid the warning, since there's a special case to avoid emitting it for if (variable), but that wouldn't allow us to upgrade the branch to if constexpr, because (AFAICT) there is no actual requirement that a user allocator has a constexpr max_size() (and pre-C++11 user allocators wouldn't).

[AlexGuteniev]

but that wouldn't allow us to upgrade the branch to if constexpr

I think it is still both cleaner way to avoid the warning, at it is almost idiomatic way to do so, and better debug codegen

[StephanTLavavej]

I've pushed that change, since I agree that it's clearer. However, I don't see how this will affect debug codegen.

[davidmrdavid]

LGTM, soft suggestion to improve comment

[AlexGuteniev]

However, I don't see how this will affect debug codegen.

Please see: https://godbolt.org/z/exKhPY5KP

[AlexGuteniev]

However, I don't see how this will affect debug codegen.

Please see: https://godbolt.org/z/exKhPY5KP

Yes I was wrong. The codegen is worse!

[frederick-vs-ja]

Per WG21-P2280R4, IIUC we can determine whether the max_size returns an instance-unrelated constant and use if constexpr in some cases. But such determination is a bit arcane.

[StephanTLavavej]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.