Julia: Base.TwicePrecision{Float64} <: Real == false

[eford]

https://github.com/JuliaLang/julia/blob/a43c46b68796228838041a140356f96e6f76184b/base/twiceprecision.jl#L184-L187

Is there a good reason why TwicePrecision{Float64} shouldn't also be a subtype of Real?
If so, how should one specify that why want a StepRangeLen where R and S might be regular reals or a TwicePrecision version?

[timholy]

If you want "regular reals," use StepRangeLen(ref, step, len) directly. If you want TwicePrecision, use range(start, step=step, length=len).

TwicePrecision is an internal type designed to enable hitting the endpoints of Float64 ranges accurately. If we start turning it into more of a regular number, keep in mind we might eventually need QuadPrecision to enable precise TwicePrecision range operations. I'd rather not go down that road, personally.

[eford]

My problem is that I'm trying to create struct's and functions that have types based on the types in a parameter which is a range. If the range happens to be using a TwicePrecission, I still want it to call the same version of the function as would normally be called if it were a real. I guess I could do something like
T1 although I was hoping there was a clearner way.

[timholy]

That doesn't look terribly messy to me. But if you don't want to have to deal with even acknowledging the existence of TwicePrecision, you may not need to:

julia> r = 1.2:2.8:16.4
1.2:2.8:15.2

julia> eltype(r)
Float64

julia> typeof(r)
StepRangeLen{Float64,Base.TwicePrecision{Float64},Base.TwicePrecision{Float64}}

julia> supertype(typeof(r))
AbstractRange{Float64}

You might be able to write all your methods to dispatch on AbstractRange{T} where T and just use the generic interface functions first, step, last.