@distributed fails silently when lacking @sync or reduction

[grahamas]

For a simple example:

@distributed for (i, j) in Base.Iterators.product([1, 2], [3, 4])
    @show i + j; i + j
end

This should error, as in #57, but no error is reported. The code silently never runs the contents of the distributed loop. If you tack on a @sync or even put in a reduction, i.e. @distributed (*) then the error is reported properly.

This is relevant to my use-case where I set up a RemoteChannel and listen to it after running the @distributed loop asynchronously.

julia> versioninfo()
Julia Version 1.3.0
Commit 46ce4d7933 (2019-11-26 06:09 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i5-7200U CPU @ 2.50GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.1 (ORCJIT, skylake)

[mbauman]

When you do a t = @distributed for, you get back a task. The task's value is always nothing after all the workers come back. You can explicitly make sure no task errored with a fetch(t) — this is how you can check for both iteration space errors as well as errors inside the distributed work. In this case (#30343), I think we could probably detect it ahead of time and give a nicer error. But there may be interior errors that you'd probably want to check, too — and I'm not sure we can do anything there.