Support for extended precision (BigFloats)?

[PerezHz]

Hi! I was trying to use Elliptic.jl in order to evaluate K(k) to high accuracy using BigFloats, but noticed that if given a BigFloat as an input, Elliptic.K returns a Float64:

julia> using Elliptic

julia> Elliptic.K(big"0.512341234")
1.8646739600308775

@oxinabox explained to me that, by construction, Elliptic.jl functions explicitly cast all Reals down to Float64, and was wondering if the code may be adapted for other types such as BigFloat, perhaps adapting the constants such as D1MACH1 with something like const D1MACH1 = realmin(T) with T<:AbstractFloat, etc., for these types?

[ettersi]

See JuliaMath/SpecialFunctions.jl#79.

[dlfivefifty]

I think this fallback should be removed:

Elliptic.jl/src/Elliptic.jl

Line 63 in 7ed7dfb

E(x::Real) = E(Float64(x))

It would be better for the code to error on a BigFloat value than to silently convert to Float64>