[julia-users] Re: Easy way to declare function works on array of floats

[Steven G. Johnson]

On Sunday, May 25, 2014 3:39:33 AM UTC-4, gael....@gmail.com wrote:

> But,
> function foo(a::Array{FloatingPoint}, b::Array{FloatingPoint})
>     a + b
> end 
> is unfortunately not working. I don't know whether this is a bug or a 
> choice.
>
 
This is not a bug, it just doesn't mean what you think.    (This is such a 
common question, it really should be in the FAQ.)

An Array{Float64} is not an instance of Array{FloatingPoint}.   
a::Array{FloatingPoint} is an array of generic "boxes" which can contain 
any floating-point type, potentially a different type for *each element*; 
e.g. a[1] could be a Float64, a[2] could be a Float32, a[3] could be a 
BigFloat, etcetera.   An Array{Float64} is an array where *every element* 
is a Float64, and hence this can be stored and computed with much more 
efficiently than an Array{FloatingPoint}.

What you want is

function foo{T1<:FloatingPoint, T2<:FloatingPoint)(a::Array{T2}, 
b::Array{T2})
    a + b
end

This says that the arrays can each be of any floating-point type, but the 
array elements are all of the same type.

You could also make the code a bit more generic with:

function foo{T1<:FloatingPoint, T2<:FloatingPoint)(a::AbstractArray{T2}, 
b::AbstractArray{T2})
    a + b
end

This allows the arguments to be any subtype of AbstractArray (which 
includes Array but also things like ranges).