Allow empty reductions for maximum(Unsigned) and compose

[eschnett]

No description provided.

[rfourquet]

A few words explaining the rationale would be useful. I understand the reasoning for compose, but having maximum(UInt[]) === UInt(0) would need some discussion. E.g. why would it work only for unsigned, e.g. why not also maximum(Int[]) == typemin(Int) and same for floats?

[eschnett]

I thought about Int etc., and I thought this would require more discussion. For example, typemin(BigInt) doesn't exist. typemin(Int) is probably rarely the value that people want – one often stores non-negative values as Int, and then the neutral element is 0. If there is no clear-cut rule, then the safe way out is to do nothing and ask people to use the init argument instead.

Using -Inf for Float64 etc. makes more sense.

If it was easily possible to deduce automatically whether typemin exists for a particular type, then we could always use that as default. I think this is possible with a @generated function, but I'm not sure that is worth the trouble without discussion.

Unsigned seems a safe choice since the neutral element for maximum there is unambiguous.

[eschnett]

The current error is in the documentation for another package (SparseArrays). How should I address this problem? Is there a way to link two pull requests?

[nlw0]

Seems related to #44819

[giordano]

This is changing an error tested in SparseArrays (https://buildkite.com/julialang/julia-master/builds/29424#018b81a2-76cf-4a83-9c0f-bba97e31c621/768-1631) which is now in a different repository:

Error in testset SparseArrays/sparsematrix_ops:
Test Failed at /cache/build/default-amdci5-1/julialang/julia-master/julia-7469597a4f/share/julia/stdlib/v1.11/SparseArrays/test/sparsematrix_ops.jl:246
  Expression: maximum(sparse(Int[]))
    Expected: "reducing over an empty"
     Message: "ArgumentError: reducing with maximum over an empty collection of element type Int64 is not allowed.\nYou may be able to prevent this error by supplying an `init` value to the reducer."

Edit: JuliaSparse/SparseArrays.jl#466 was just opened.

[adienes]

updated by #55628