Deprecate countnz in favor of using count(predicate, x)

[andreasnoack]

As discussed in #23005, it is ambiguous which kind of non-zero to test with (!iszero or t -> t != 0) so with this PR the user is required to be explicit. To summarize the difference

julia> map(t -> t == 0, [rand(2) for i in 1:3])
3-element Array{Bool,1}:
 false
 false
 false

julia> map(!iszero, [rand(2) for i in 1:3])
3-element Array{Bool,1}:
 true
 true
 true

julia> map(t -> t != 0, ["Julia" for i in 1:3])
3-element Array{Bool,1}:
 true
 true
 true

julia> map(!iszero, ["Julia" for i in 1:3])
ERROR: MethodError: no method matching zero(::String)
Closest candidates are:
  zero(::Type{Base.LibGit2.GitHash}) at libgit2/oid.jl:166
  zero(::Type{Base.Pkg.Resolve.VersionWeights.VWPreBuildItem}) at pkg/resolve/versionweight.jl:80
  zero(::Type{Base.Pkg.Resolve.VersionWeights.VWPreBuild}) at pkg/resolve/versionweight.jl:122
  ...
Stacktrace:
 [1] iszero(::String) at ./number.jl:27
 [2] (::getfield(Base, Symbol("##61#62")){typeof(iszero)})(::String, ::Vararg{String,N} where N) at ./operators.jl:974
 [3] _collect at ./array.jl:678 [inlined]
 [4] collect_similar(::Array{String,1}, ::Base.Generator{Array{String,1},getfield(Base, Symbol("##61#62")){typeof(iszero)}}) at ./array.jl:625
 [5] map(::Function, ::Array{String,1}) at ./abstractarray.jl:1969

The deprecation suggests !iszero as well as t -> t != 0 but still calls t -> t != 0 to avoid breaking current behavior. We might want to revisit sparse string arrays which is the only thing I'm aware of that relies on this behavior but that is for a different PR.

[nalimilan]

This is inconsistent with the findn definition above and with its docstring, which states "determined by A[i]!=0".

[nalimilan]

Note that the deprecation is different from the current definition, which (as you remarked) is technically breaking. I think the best approach would be to use Base.depwarn to print a custom deprecation warning proposing both forms, but continue calling x -> x != 0 under the hood.

[andreasnoack]

I've changed the deprecation to suggest both versions following your suggestion.

[nalimilan]

What's this new method? Should go in a different commit if that's really intended.

[andreasnoack]

This method was missing. There is already a (dense) generic fallback here

julia/base/reduce.jl

Lines 698 to 704 in 5645fb2

	function count(pred, a::AbstractArray)
	n = 0
	for i in eachindex(a)
	@inbounds n += pred(a[i])::Bool
	end
	return n
	end

but not a sparse version. The count(pred, x) method is suggested by the deprecation warning so I think it makes sense to add it in this PR.

[nalimilan]

Got it. Might be worth a word in the commit message.

[nalimilan]

Unfortunately, using x -> x != 0 in findn triggers this error: generated function body is not pure. this likely means it contains a closure or comprehension. Just create an equivalent function?