fix #39379, use first∘LinearIndices instead of 1

[mbauman]

to resolve a zero-index getindex to a linear index.

fix #39379

[mbauman]

Note that this is a zero-cost abstraction:

julia> @code_llvm Base._to_linear_index([])
;  @ abstractarray.jl:1196 within `_to_linear_index'
define i64 @julia__to_linear_index_133({}* nonnull align 16 dereferenceable(40) %0) {
top:
  ret i64 1
}

[simeonschaub]

Is it a problem that _to_linear_index will now throw for empty arrays? It might at least make some error messages potentially more confusing.

[simeonschaub]

Ah, never mind. We do actually support first for empty ranges.

[mbauman]

Is it a problem that _to_linear_index will now throw for empty arrays?

It doesn't (see the unconditional ret 1 above).

julia> LinearIndices([])
0-element LinearIndices{1, Tuple{Base.OneTo{Int64}}}

julia> first(LinearIndices([]))
1

[KristofferC]

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

[nanosoldier]

Something went wrong when running your job:

NanosoldierError: error when preparing/pushing to report repo: failed process: Process(`git pull -X ours`, ProcessExited(1)) [1]

Unfortunately, the logs could not be uploaded.
cc @christopher-dG

[christopher-dG]

I upgraded nanosoldier.jl on one of the nodes, I'll have a look at this in the morning

[mbauman]

I’d be surprised if we even have any benchmarks for A[] where A is not zero-dimensional.

[KristofferC]

I just wanted to check that there wasn't any overhead from this in normal benchmarks (things starting to fail to inline etc).