Push single values to a multidimensional array?

[robsmith11]

Thanks for sharing this package. It's a very useful data structure to use when I write simulations and need moving windows of multidimensional data.

I was surprised by the behavior of push! at first:

julia> a = CircularArrayBuffer(zeros(4,3))
4×3 CircularArrayBuffer{Float64, 2}:
 0.0  0.0  0.0
 0.0  0.0  0.0
 0.0  0.0  0.0
 0.0  0.0  0.0

julia> push!(a, 0.5)
4×3 CircularArrayBuffer{Float64, 2}:
 0.0  0.0  0.5
 0.0  0.0  0.5
 0.0  0.0  0.5
 0.0  0.0  0.5

On second thought, it of course makes sense that this behavior is useful in some situations, but would it also be possible and make sense to add another function or add an option to push! that would allow one to efficiently push single values?

Something like:

julia> push!(a, 0.5, broadcast=false)
4×3 CircularArrayBuffer{Float64, 2}:
 0.0  0.0  0.0
 0.0  0.0  0.0
 0.0  0.0  0.0
 0.0  0.0  0.5

[findmyway]

Hello @robsmith11 ,

Actually the first case you shown above is an unexpected error.

I should have added a check at:

CircularArrayBuffers.jl/src/CircularArrayBuffers.jl

Line 68 in 0885223

function Base.push!(cb::CircularArrayBuffer{T, N}, data) where {T,N}

to make sure that the length of data should match buffer.step_size.

What you want in the second example is kind of unclear to me.

Let's say we have a buffer like this:

julia> b = CircularArrayBuffer{Int}(3,2)
3×0 CircularArrayBuffer{Int64, 2}

julia> push!(b, [1,1,1])
3×1 CircularArrayBuffer{Int64, 2}:
 1
 1
 1

julia> push!(b, [2,2,2])
3×2 CircularArrayBuffer{Int64, 2}:
 1  2
 1  2
 1  2

Then after executing push!(b, 3), do you wish to have

julia> [1 2; 1 2; 2 3]
3×2 Matrix{Int64}:
 1  2
 1  2
 2  3

?

This somewhat violates my assumption above （the length of each element should be the same with the step_size).
And I think in this case, the buffer is just a CircularVectorBuffer. So you can achieve the same goal with it like this:

julia> b = CircularVectorBuffer{Int}(6)
0-element CircularVectorBuffer{Int64}

julia> append!(b, 1:3)
3-element CircularVectorBuffer{Int64}:
 1
 2
 3

julia> append!(b, 4:6)
6-element CircularVectorBuffer{Int64}:
 1
 2
 3
 4
 5
 6

julia> append!(b, 7)
6-element CircularVectorBuffer{Int64}:
 2
 3
 4
 5
 6
 7

julia> reshape(b, 3, 2)
3×2 reshape(::CircularVectorBuffer{Int64}, 3, 2) with eltype Int64:
 2  5
 3  6
 4  7

[robsmith11]

Ah thanks for the tip. I had tried using reshape with a standard vector, but wasn't able to pop! and push! after reshape because that would mean resizing and shared array. But pushing to a circular vector after reshaping does indeed still work.