Why are C++ version of fpclassify and its friends provided by UCRT?

[frederick-vs-ja]

Such situation seemly blocks many changes, e.g. some relatively easy parts of WG21-P0533...

[amyw-msft]

I believe this is mostly historical, but that decision is now locked to maintain compatibility. The fpclassify function is a C function, so when the C Runtime was split from the VC++ Redistributable and added to the Windows OS as the Universal C Runtime (for VS 2015), fpclassify went with it as a member of math.h. The STL header cmath has historically simply imported it into the std namespace, like many other C math.h functions.

The Universal C Runtime ships with the Windows OS and Windows SDK instead of the VC++ Redistributable, so there are build and runtime complications involved in updating it consistently to match a C++ standard, let alone 'transfer' the function implementation from the UCRT to the STL. That said, header-only changes in the UCRT should be able to be done and would simply require up-to-date Windows SDKs to access once the changes are shipped. Being the C runtime does not preclude it from taking changes for the purpose of conforming to the C++ standard where possible.

[frederick-vs-ja]

I think I've found how to implement the easy parts of WG21-P0533 without intrinsics (FP overloads of fpclassify and signbit are unfortunately not yet possible), but only in <cmath>. It's still required to make the changes also available in <math.h>. Thanks again!