Support parallelizing over collections that have non-Int lengths

[sbromberger]

Changes splitrange to use Integer instead of Int so that mixed-width ints can be used. This fixes the following LightGraphs issue on 32-bit machines:

julia> g = Graph(10,20)
{10, 20} undirected simple Int32 graph

julia> Parallel.stress_centrality(g);

julia> g = Graph{Int64}(g)
{10, 20} undirected simple Int64 graph

julia> Parallel.stress_centrality(g);
ERROR: MethodError: no method matching splitrange(::Int64, ::Int32)
Closest candidates are:
  splitrange(::Int32, ::Int32) at /buildworker/worker/package_linuxarmv7l/build/usr/share/julia/stdlib/v1.0/Distributed/src/macros.jl:235

[sbromberger]

@pfitzseb on Slack (I agree wholeheartedly):

imho all functions there should use Integer or T where T<:Integer

[sbromberger]

x-ref: sbromberger/LightGraphs.jl#985

[JeffBezanson]

A slightly better fix is to change the calls to splitrange to splitrange(Int(N), ...). Otherwise splitrange will be unnecessarily specialized. At least, the np argument is always an Int and doesn't need to be changed.

[sbromberger]

@JeffBezanson - ref #28652. Thank you.

[sbromberger]

I'll leave it to someone else to decide, then. My #28652 alternative was closed (I disagree with the closure, but whatever). In any rate, I've fixed it locally.

[sbromberger]

Gah. Didn't mean to close this.

What path do we want to take?

[sbromberger]

Ping on this - any chance it could be merged soon? This fixes LightGraphs on 32-bit machines.

[vchuravy]

LGTM, would be great to have a test though.

[KristofferC]

Bump for test and this can be merged.

[sbromberger]

Could someone else please write the test? I edited this in GitHub's editor and can't figure out how to add a different file to the change set.

[KristofferC]

Check the branch out locally or go to https://github.com/sbromberger/julia/tree/patch-1 and edit with GitHub.

[KristofferC]

This test doesn't test what was implemented (and thus fails).

[andreasnoack]

Maybe

for T in (Int64, Int128)
  n = @distributed (+) for i in Base.OneTo(T(10))
    i
  end
  @test n == 55
end

This only happens whenever length returns something that is not an Int which is relatively rare.

[sbromberger]

Yep. I had something written out last night but yours is simpler. Thanks.

[sbromberger]

This test doesn't test what was implemented (and thus fails).

Yup. The test was for the "old" fix. Sorry 'bout that. I've just pushed new tests (hat tip: @andreasnoack) that hopefully will work.

[vchuravy]

Thanks Seth! Note to whoever merges this, please do a squash merge.

[sbromberger]

please do a squash merge.

It'd be better for my reputation if you did a fixup instead, but I'll take a squash :)

[KristofferC]

The commit messages can be modified when squashing so all traces of Github editing can be removed ;)

[sbromberger]

Travis failure is OSX-only and seems to be unrelated:

The log length has exceeded the limit of 4 MB (this usually means that the test suite is raising the same exception over and over).

The job has been terminated