Thiago Macieira via Std-Proposals:
> We wanted to add a specialisation of std::numeric_limits for qfloat16 as well
> as other cmath functions like inf(), fpclassify(), etc. Since doing that was
> UB, we had to add them with a q prefix (qInf, qFpClassify, etc.).
>
> In order to write generic code that deals with qfloat16, float and double, we
> needed to add the same overloads and specialisations for float and double.
>
> Conclusion: now Qt code has to use qFpClassify() instead of std::fpclassify().
>

This is because <cmath> is fundamentally broken because it doesn't use customization point objects. <cmath> needs a proper C++
replacement.