weird dispatch/inference/arithmetic bug

[StefanKarpinski]

julia> f(x::Number, y::Number) = x^2 - 2y + 1

julia> f(2, 3.0)
-4611686018427387903

[JeffBezanson]

Hate to say it but works for me. Interesting though:

julia> f(x::Number, y::Number) = x^2 - 2y + 1

julia> f(2, 3.0)
-1.0

julia> hex(4611686018427387903)
"3fffffffffffffff"

julia> num2hex(1.0)
"3ff0000000000000"

The bits seem to be related to the correct answer in some weird way.

Is this your normal build? Passing all other tests?

[StefanKarpinski]

Yeah, I checked it multiple times because it seemed so out of character. It looks almost like a case of arithmetic extension where zero extension should be used. Strange.

[JeffBezanson]

Turns out to be a type inference problem. Working on it.
I get this in promote_type:

promote_rule::F(T::T<:Union(Int64,Complex128),S::Type{Float64})::None

The first argument should be of type Type{T<:Union(...)}.

[ViralBShah]

BTW, it does work for me. Not sure if this is because it is already fixed, or my setup is unaffected.

[JeffBezanson]

fixed in commit b483b55