John Maddock wrote:
I'm open to the idea of some sort of "Math core" lib, even though it's likely to suffer feature creep, ...
It seems to me that the fpclassify.hpp/sign.hpp functions, that correspond to the "Classification macros" section of C99, are "core" enough to stand apart from the rest of the Math library. We could have a "float" module, the analog of the "integer" module, but for the built-in floating point types, which would be the natural place for those functions.
but provided - and this is going to be the controversial bit - that tests and docs remain in the "one true Math lib". In other words it's just a small module containing some headers, with links to the Math lib for docs and tests.
No opinion on that one. It's just a handful of functions anyway. Pretty much anything will work.