Inconsistency between findmax and maximum with respect to NaN

[wsshin]

I read this discussion about max and maximum returning NaN when the collection includes NaN. For example,

julia> VERSION
v"0.7.0-DEV.82"

julia> v = [1.0, NaN, 2.0]
3-element Array{Float64,1}:
   1.0
 NaN
   2.0

julia> maximum(v)
NaN

However, findmax still returns the maximum among non-NaN values and its index:

julia> findmax(v)
(2.0, 3)

This seems inconsistent. Is this something intended or a bug?

[StefanKarpinski]

Seems like findmax needs to be updated to match maximum. There should also be tests to check that they agree.

[ararslan]

Note that the behavior of findmax is documented to skip NaNs: https://docs.julialang.org/en/latest/stdlib/collections/#Base.findmax-Tuple{Any}

NaN values are ignored, unless all elements are NaN.

[StefanKarpinski]

Yes, it's documented to do this. The question is whether it should be changed to match maximum or not. It seems to me that these functions should have matching behavior.

[ararslan]

Right, I'm just addressing "is this something intended or a bug." I agree that the behaviors should match.

[cafaxo]

In 1.1, this works correctly:

julia> v = [1.0, NaN, 2.0]
3-element Array{Float64,1}:
   1.0
 NaN  
   2.0

julia> findmax(v)
(NaN, 2)

julia> maximum(v)
NaN

I think this issue can be closed. (Might be a duplicate of #23209?)

[StefanKarpinski]

Seems like it on both counts. Thanks!