fix treat_as_floating_point_v<_DurationType>

[Igor Zhukov (fsb4000)]

Fixes #2636

[Stephan T. Lavavej (StephanTLavavej)]

Thanks! (I especially appreciate the added test coverage.)

To avoid the need to suppress compiler warnings, I went ahead and pushed what looks like a major rearrangement, but is actually a few simple transformations:

• Use De Morgan's law to flip the condition.
• After an early return, we don't need else.
• Separate if (compile_time_condition && runtime_condition) into nested if constexpr and if.
• Reduce the scope of using _CastedType to where it's needed.

I believe that the resulting code reads naturally: "At compile-time, if we're not treating the rep as floating-point, then we're treating it as integral (which might lose precision). We'll need _CastedType. Perform a runtime check for precision loss - if duration_cast returns a non-equal value, we've lost precision and must immediately return false. Otherwise, we're good to go for updating _Result."

For the treat-as-floating-point case, there is an extremely minor throughput advantage to if constexpr (don't need to instantiate duration_cast<_CastedType>) although I was originally motivated by the compiler warning suppression.

Hope this is okay - it's a bit larger than the changes I usually push without asking.

[Igor Zhukov (fsb4000)]

I was originally motivated by the compiler warning suppression

Good.
Yes, Visual C++ compiler said me the same: "C4127: conditional expression is constant. Consider using 'if constexpr' statement instead"
But I looked at the code and I found that we have 3 other places with:

#pragma warning(push)
#pragma warning(disable : 4127) // conditional expression is constant

So this is why I did "pragma push"...
I should have listened to the compiler :)

[Stephan T. Lavavej (StephanTLavavej)]

Ah, thanks - I had forgotten we still had some of those. I'll go through and clean up (in a separate PR) as many of the 4127 suppressions as possible/convenient (most of them date back to when we couldn't use if constexpr unconditionally). The ones in are probably more trouble to change than they're worth.

[Stephan T. Lavavej (StephanTLavavej)]

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

[Stephan T. Lavavej (StephanTLavavej)]

Thanks again for fixing this bug and testing the affected parsing scenario! 😻 🐞 🎉