julia resizing to n-1 dimension array in singleton dimensions

[silgon]

Hello, I would like to know why julia doesn't treat an array of N dimensions with a singleton in one of the dimensions as an array of (N-1) dimensions.
To explain that a little bit let's say I have the following code.

julia> a=ones(3,3,3);
julia> a[:,:,1]
3x3 Array{Float64,2}:
 1.0  1.0  1.0
 1.0  1.0  1.0
 1.0  1.0  1.0
julia> a[1,:,:]
1x3x3 Array{Float64,3}:
[:, :, 1] =
 1.0  1.0  1.0
[:, :, 2] =
 1.0  1.0  1.0
[:, :, 3] =
 1.0  1.0  1.0

In the second line you can see that the singleton dimension a[:,:,1] is converted to Array{Float64,2} reducing the dimensions. Nonetheless this effect doesn't happen with a[1,:,:] because it keeps the Array{Float64,3} dimension. So in the first case I'm able directly to do matrix operations because I have Array{Float64,2} but it's not the same in the second case because I have to transform Array{Float64,3} to Array{Float64,2} by reducing my singleton dimension.

I have the following options to resize it:

julia> reshape(a[1,:,:],size(a,2),size(a,3))
3x3 Array{Float64,2}:
 1.0  1.0  1.0
 1.0  1.0  1.0
 1.0  1.0  1.0
julia> slice(a,1,:,:)
3x3 SubArray{Float64,2,Array{Float64,3},Tuple{Int64,Colon,Colon},3}:
 1.0  1.0  1.0
 1.0  1.0  1.0
 1.0  1.0  1.0

My problem now is that this conversion I have to make it a lot of times in my case, because I use it in an iterative process. I fear that this can lead to performance problems just because of this conversion will slow down the process, any thoughts or recommendations?

[Jutho]

This has already been changed in julia v0.5, see PR #13612 . Note however that a[1,:,:] still copies the data, so this could still be costly in an iterative process if you don't actually need a copy and are satisfied with just a view. Then slice(a,1,:,:) is the way to go.

[silgon]

do you mean v0.45? I don't see any v0.5 in the tags on github. I'm asking in order to change to that version, because now I have the version provided by the ppa:staticfloat/juliareleases repositories on ubuntu 14.04

[Jutho]

v0.5 is still in development, so I guess I should have said, this is changed on the master, which you can get e.g. as nightly built ( http://julialang.org/downloads/ ) or by cloning the github repository and building it yourself. But unless you explicitly need a copy, I think you want to use slice which works just fine on the current release version of julia.