Make map! correctly handle differently-sized containers.

[maleadt]

Closes #36235.

The map! docstring specifies:

Like map, but stores the result in destination rather than a new collection. destination must be at least as large as the smallest collection.

However, we're neither boundschecking the destination container, nor using the smallest input collection:

# should throw
julia> map!(identity, [0], [1,2])
1-element Vector{Int64}:
 1

# uses undefined values
julia> map!(+, [0,0], [1,2], [3], [4])
2-element Vector{Int64}:
           8
 10187977346

This PR corrects that.

Will need some careful benchmarking, maybe add some @inbounds to map! calls and/or outlining the throwing.

FWIW, the performance of the example in #30624 (comment) is identical.

[maleadt]

@nanosoldier runbenchmarks(ALL, vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here.

[maleadt]

Looks like this exposes some other existing issues, e.g. with OffsetArrays:

julia> map!(+, [0, 0, 0, 0], fill(0, -4:-1), fill(0, -4:-1), fill(0, -4:-1))
Int64[]

julia> map!(+, [0, 0, 0, 0], fill(0, -4:-1), fill(0, -4:-1))
4-element Vector{Int64}:
 0
 0
 0
 0

[maleadt]

@nanosoldier runbenchmarks(ALL, vs=":master")

[maleadt]

The changes in allocations here can be ignored, I think. They come from the additional bounds checks, but only when running with --check-bounds=yes (that the boundscheck is elided for that specific test). The failing test:

include("../../test/testhelpers/OffsetArrays.jl")
using Main.OffsetArrays
R = fill(0, -4:-1, 7:7, -6:-6)
B = OffsetArray(reshape(1:24, 4, 3, 2), -5, 6, -7)
@allocated maximum!(R, B)

1264 bytes on master, 1248 on this PR, but 1408 bytes when forcing bounds checks.

[maleadt]

@nanosoldier runtests()

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here.

[nanosoldier]

The package evaluation job you requested has completed - possible new issues were detected.
The full report is available.

[adienes]

was the attempt here successful? is there anything I can do to assist? e.g. particular benchmark regressions

[maleadt]

Sadly not, the performance regressions are real, but I didn't have the time to look into them. Feel free to take this up and create your own PR, if you want.

[LilithHafner]

Closing as abandoned, feel free to re-open if you'd like to pick it back up