constexpr signbit

[mborland]

No description provided.

[mborland]

@NAThompson This is good for review. I deviate slightly from the C++ standard (same as copysign) in that using UB to find the sign of a NAN is not allowable in constexpr contexts.

[jzmaddock]

That all looks fine to me, I wonder how useful it will be if we can't detect signed zero though?

Hmm... just a thought, we can detect signed infinities right? So how about checking the sign of 1/x when x is zero? Might need to verify that the type has infinities first, and I worry it might trigger compiler traps, but would make the facility more generally useful.

[mborland]

That all looks fine to me, I wonder how useful it will be if we can't detect signed zero though?

Hmm... just a thought, we can detect signed infinities right? So how about checking the sign of 1/x when x is zero? Might need to verify that the type has infinities first, and I worry it might trigger compiler traps, but would make the facility more generally useful.

We can detect signed infinities, but division by 0 is not allowed in constexpr contexts because it is UB.

[jzmaddock]

We can detect signed infinities, but division by 0 is not allowed in constexpr contexts because it is UB.

Oh :( I knew there had to be a catch somewhere. Oh well, never mind, carry on as you were ;)