Specialize NaNMath.pow to allow for ::Int ^ ::Int -> ::Int

[ChrisRackauckas]

NaNMath does not handle integer dispatches. This fixes it so that a wrapper function handles the integer dispatch, restoring the behavior of ^

[devmotion]

I just noticed the Int problem as well. Wouldn't it be easier to just never rewrite ^ to NaNMath.pow if args[2] isa Int? There shouldn't be any numerical issues anyway in this case.

[ChrisRackauckas]

In theory everyone's symbolic types should match the types they end up passing to the function. In practice I think it's more robust to not assume that, and since it's practically zero overhead to not assume that (since this dispatch should compile out in almost all cases of symbolic generated functions), I would take the more robust route.

[devmotion]

On a second thought, I guess we should just add pow(x::Real, y::Int) = x^y to NaNMath?

[devmotion]

I opened JuliaMath/NaNMath.jl#83.

[ChrisRackauckas]

Yeah that's a more general solution.