argmin / findmin incorrect behaviour on SparseVector on 1.8

[matbesancon]

This behaviour is incorrect, I'm not sure there is documentation on which case would produce nothing for findmin

On 1.8.0-beta3

julia> v = zeros(3)

julia> argmin(v)
1

julia> v[1] = 2

julia> argmin(v) 
2

# incorrect on sparse
julia> v = spzeros(3)
3-element SparseVector{Float64, Int64} with 0 stored entries

julia> v[1] = 2
2

julia> argmin(v) === nothing
true

[StefanKarpinski]

This seems like a fairly significant bug. Adding to the milestone.

[matbesancon]

Yeah I tripled-checked the doc before, thought I had missed an expected behaviour

[matbesancon]

given the date of merge (Apr 26), I thought this was integrated but the behaviour is still present in 1.8-rc1 (May 27).
Will the next RC contain the fix?

[matbesancon]

the issue is still there on 1.8 released

julia> using SparseArrays
julia> direction = spzeros(3); direction[1] = 2

julia> argmin(direction) === nothing
true

[KristofferC]

SparseArrays.jl (https://github.com/JuliaSparse/SparseArrays.jl) needs to create a backport 1.8 branch where they backport the fix for this and it can be put into 1.8.

[matbesancon]

Linking the corresponding issue:
JuliaSparse/SparseArrays.jl#239

[matbesancon]

@KristofferC is the backport planned for 1.8.1? It is a pretty serious regression to have lying around IMO and can break a lot of user code (as it broke at least one registered package)