Change array type in mutating function?

General Usage
1

I am trying to define some simple functions for 2-D and 3-D rotations, and I would like to define a mutating version as well. This function

function rotate!(r, θ)
    r[:] = exp(θ*[0 1 ; -1 0])*r
end

works for arrays of floats, but fails for arrays of ints because the type of r is Array{Int64}, and I’m trying to fill it with floats.

Is there a way to have this function also mutate the array type? I tried a few things and couldn’t find a solution.

2

No, the type of a value can never be mutated in Julia, so there’s no way to mutate an Array{Int} into an Array{Float64}. The easiest way around this is to just not pass an Array{Int} into your function.

If you want a convenient way to operate on Int array inputs, you could consider making both a mutating and non-mutating version of the function:

function rotate!(r)
  r .= 1.5
end

function rotate(r)
  out = similar(r, promote_type(eltype(r), Float64))
  rotate!(out)
  out
end

This is a common pattern in Julia, where the non-mutating function is just a convenient wrapper to allocate some output, call the mutating version on that output, and then return the result.

3

No, the element type of an array is fixed.

In this specific case, the mutating function is not written properly.
It allocates a number of temporary arrays, does the multiplication and writes the result back from a temporary array to the original. Whenever you have that pattern, it’s best to just return the temporary array instead of doing extra work of writing its contents into the original.

4

Okay, I had a suspicion that was the case. I did write up non-mutating versions, so I might try something like you suggest.

hmmm… okay, I’ll have to keep that in mind.

5

Well, you can use reinterpret, and change the eltype while keeping the data. But that won’t do the right thing in this case.