map with different length iterators

[boathit]

Is it reasonable that map can deal with infinite length iterators well but fails to work with two different finite length iterators?

map(+, [1,2,3], repeated(5))

3-element Array{Any,1}:
 6
 7
 8


map(+, [1,2,3], repeated(5, 10))

3-element Array{Any,1}:
 6
 7
 8


map(+, 1:3, repeated(5, 10))

3-element Array{Any,1}:
 6
 7
 8

works well

 map(+, [1,2,3], 1:10)

ERROR: DimensionMismatch("dimensions must match")
 in map at abstractarray.jl:1310

map(+, 1:3, 1:10)

ERROR: DimensionMismatch("dimensions must match")
 in map at abstractarray.jl:1310

error occurs

[ihnorton]

Suggested alternative:

map(+, [1,2,3], 1:10...)

I'm unsure if it would be a good idea to auto-expand the range 1:10 object (a.k.a. colon(1,10)) in cases like this.

(there are a some similar cases where such auto-expansion might be convenient -- but possibly undesirable globally -- such as indexing a KeyIterator: keys(Dict(1=>"a"))[1] -> MethodError. instead requires: [keys(Dict(1=>"a"))...][1])

[boathit]

map(+, [1,2,3], 1:10...) will be expanded into map(+, [1,2,3], 1,2,3,4,5,6,7,8,9,10) which departs from the original intention that expects to perform map(+, [1,2,3], [1,2,3,4,5,6,7,8,9,10])

[JeffBezanson]

Maybe this should be allowed for 1-d arrays of different lengths?

[raghav9-97]

@JeffBezanson Is it logically correct if map(f, x, y) work with different lengths of x and y and if length(x) < length(y) the resulting array size should be length(x)?

[raghav9-97]

I also think that issue #29523 is related to this problem.

[timholy]

An orthogonal solution is to define new iterators like repeatlast(x) which turns a container x into an infinite-length iterator, "padding" with the final element; one could also define Iterators.truncate(y, x) so that a longer iterator y matches a shorter iterator x. I think this would be much more explicit about intent, and also more flexible.

[raghav9-97]

The promote_shape(::Indices, ::Indices) function works just opposite to what has been suggested for Iterators.truncate(y, x).So instead of promote_shape map should make a call to Iterators.truncate(y, x) function?

[timholy]

No, promote_shape has a very different purpose:

julia> promote_shape((3, 5), (2, 6))
ERROR: DimensionMismatch("dimensions must match")
Stacktrace:
 [1] promote_shape(::Tuple{Int64,Int64}, ::Tuple{Int64,Int64}) at ./indices.jl:103
 [2] top-level scope at none:0

[jonas-schulze]

IIUC this should be fixed by #29927:

map(f, iters...) = collect(Generator(f, iters...))
Generator(f, I1, I2, Is...) = Generator(a->f(a...), zip(I1, I2, Is...))

[mschauer]

Yes, the reported issue was fixed by that:

julia> map(+, [1,2,3], 1:10)
3-element Array{Int64,1}:
 2
 4
 6