iterators like pemutations do not support @threads

[Marco-Congedo]

iterators like pemutations do not support @threads

julia> P=permutations(1:3)
Combinatorics.Permutations{UnitRange{Int64}}(1:3, 3)

julia> for p∈P
       println(p)
       end
[1, 2, 3]
[1, 3, 2]
[2, 1, 3]
[2, 3, 1]
[3, 1, 2]
[3, 2, 1]

julia> @threads for p∈P
       println(p)
       end

ERROR: TaskFailedException

nested task error: MethodError: no method matching firstindex(::Combinatorics.Permutations{UnitRange{Int64}})

Closest candidates are:
  firstindex(::Any, ::Any)
   @ Base abstractarray.jl:447
  firstindex(::Base.JuliaSyntax.SourceFile)
   @ Base C:\workdir\base\JuliaSyntax\src\source_files.jl:131
  firstindex(::Markdown.MD)
   @ Markdown C:\Users\congedom\AppData\Local\Programs\Julia-1.10.0\share\julia\stdlib\v1.10\Markdown\src\parse\parse.jl:27
  ...

Stacktrace:
 [1] (::var"#539#threadsfor_fun#148"{var"#539#threadsfor_fun#147#149"{…}})(tid::Int64; onethread::Bool)
   @ Main .\threadingconstructs.jl:199
 [2] #539#threadsfor_fun
   @ Main .\threadingconstructs.jl:181 [inlined]
 [3] (::Base.Threads.var"#1#2"{var"#539#threadsfor_fun#148"{var"#539#threadsfor_fun#147#149"{…}}, Int64})()
   @ Base.Threads .\threadingconstructs.jl:153

...and 5 more exceptions.

[adknudson]

I think in general, threaded for-loops don't work on non-indexible collections. This happens in Base as well:

# This works because you can index into a Range object
Base.Threads.@threads for p ∈ 1:4
    println(p)
end

# This throws the same TaskFailedException as in your example
Base.Threads.@threads for p ∈ Iterators.product(1:2, 1:2)
    println(p)
end

Combinatorics.permutations creates an iterator that tracks state through a vector of Ints, and so it doesn't use linear indexing, which is probably what the @threads macro expects:

Combinatorics.jl/src/permutations.jl

Lines 47 to 53 in 912d506

	function Base.iterate(p::Permutations, state::Vector{Int}=fill(firstindex(p.data), p.length))
	next_permutation!(state, firstindex(p.data), lastindex(p.data))
	if first(state) > lastindex(p.data)
	return nothing
	end
	[p.data[i] for i in state], state
	end

[adknudson]

If your state is small enough to fit in memory, then you can collect the permutations and call @threads on the returned collection:

P = permutations(1:3)

@threads for p in collect(P)
    println(p)
end

Just be sure that the time saved by threading can offset the overhead of collecting the permutations.