Should append! accept any iterable?

[ivarne]

append! is currently limited to instances of AbstractArray. I can't really see why this restriction is necessary, as lots of similar functions gladly take any iterable in similar situations.

Somewhat related to #7642

[JeffBezanson]

For append! it helps to know the length of the iterable. We could have it fall back to append!(a, collect(itr)) or repeatedly calling push!, but those aren't particularly fast so there's not much value over writing that code yourself. But we could add these in the interest of being more generic.

[quinnj]

I don't think this is particularly useful. I think in general, I would prefer append! to be strict in its usage to help me, as a user, be smart about performance.
Is there an example of an iterable where this would be useful and is non-trivial (and is not fishy on performance?)
I guess I could kind of see an application for DataFrames where you want to take a row iterator from one DataFrame and append each row to another DataFrame. In that case, you'd probably want the push! behavior.

[mcg1969]

For any collection that implements length, overloading append! avoids length(B) calls to :jl_array_row_end. The implementation of collect would be a good model here, which tests applicable(length,itr) to preallocate the storage:

function append!{T}(a::Array{T,1}, items::itr)
if applicable(length,itr)
    n = length(items)
    ccall(:jl_array_grow_end, Void, (Any, Uint), a, n)
    i = length(a)
    for x in itr
        a[i+=1] = x
    end
else
    for x in itr
        push!(a,x)
    end
end

prepend! wouldn't be as elegant if length is not defined:

function prepend!{T}(a::Array{T,1}, items::itr)
if applicable(length,itr)
    n = length(items)
    ccall(:jl_array_grow_beg, Void, (Any, Uint), a, n)
    i = 0
    for x in itr
        a[i+=1] = x
    end
else
    prepend!(a,collect(T,items))
end

[mcg1969]

This dovetails with @JeffBezanson's suggestion of a Collection abstract class. That way, only collections with the same eltype would be accepted, if such a restriction were desirable.

[mcg1969]

An alternative: provide a collect! parallel to collect that appends to an existing array.

[AnnaDana]

Is there another function that takes a variable array and add it to a higher dimension one? This can be used in implementing algorithms such as column and constraint generation.

[StefanKarpinski]

Is there another function that takes a variable array and add it to a higher dimension one?

Not quite sure I understand. Can you elaborate a bit on what you mean?

[IainNZ]

@AnnaDana nice to see another OR person using Julia! Do you you mean something like

A = rand(2,2)  # 2x2 matrix
x = rand(2)  # length 2 vector
hcat(A,x)  # 2x3 matrix

[AnnaDana]

Briefly, after each solve of sub problem, when I want to add a new column which is basicaly new variable, I do the following:

iteration = iteration +1
@defVar(MasterProblem, x[iteration, 1:5] >=0)

And this works fine! I was just wondering if there is a function so that I define my variable x with dynamic size once, and at each iteration add a new row to it.

[IainNZ]

@AnnaDana since this is a JuMP specific question, can you open an issue at https://github.com/JuliaOpt/JuMP.jl ?

[vtjnash]

Seems like this should be revisited now that general iterables have a length computable attribute?

[JeffBezanson]

It looks like this was added by e909c1c

[JeffBezanson]

We're missing append!(::BitArray, iterable). Should add that method.

[StefanKarpinski]

Adding methods is non-breaking, so this does not belong on the 1.0 milestone.

[Keno]

The missing method was implemented by @JeffBezanson in 4844a4b