Load module on remote workers from master

[GregPlowman]

Fix #19960

[amitmurthy]

@vtjnash is the right person to review this.

[GregPlowman]

Testing is quite tricky as it requires workers on remote hosts.

I have something that might work (see below), but I don't know much about the test environment.
I've looked at test/distributed_exec.jl but I can't see where there are remote workers.
The section for ssh simulates different hosts with ["localhost", string(getipaddr()), "127.0.0.1"], but these hosts are all the local machine aren't they?

addprocs(hosts ...) # add workers on remote hosts

begin
    load_dir = tempdir()
    push!(LOAD_PATH, load_dir)

    function create_module(module_name, module_body, precompile::Bool)
        module_file = joinpath(load_dir, module_name, "src", string(module_name, ".jl"))
        isdir(dirname(module_file)) || mkpath(dirname(module_file))
        module_code = """
            __precompile__($precompile)
            module $module_name
                $module_body
            end
        """
        write(module_file, module_code)
    end

    create_module("Issue19960A", "export f; f() = myid()", true)
    create_module("Issue19960B", "export g; using Issue19960A; g() = f()", true)

    using Issue19960B

    @test g() == 1
    for wid in workers()
        @test remotecall_fetch(g, wid) == wid
    end
end

[GregPlowman]

Is it possible this could be backported to v0.5 (v0.5.2)?

[GregPlowman]

Rerun tests

[tkelman]

why is the catch here at all?

[GregPlowman]

Given your comment, it probably doesn't need to be there.
(As you've probably guessed I don't really understand testing)
I thought that there had to be an explicit @test failure, rather than just a thrown error from the try block.
I can remove it.

[tkelman]

something went wrong with the rebase as well, when you do git rebase -i origin/master you should remove all the commits that aren't yours from the list that shows up, and when you're happy with the local state of the branch (just your commits, on top of latest master) you can push to the glp/remote_loading branch of your fork with --force

[GregPlowman]

Yes I can see that it's screwed up.
But also test/compile.jl is showing extra changes (changes made to master after my original PR but before the update). I don't think those changes should be showing here.
I don't know how to get out of this mess, except closing this and making a new PR.

[tkelman]

You can amend a pull request in place with an interactive rebase, usually git rebase -i origin/master assuming origin is JuliaLang/julia and you've done a git fetch origin recently. Once the local copy is cleaned up, do git push --force GregPlowman branchname assuming GregPlowman is a remote for your fork

[GregPlowman]

Thanks Tony!
I've already created a replacement PR, but I'll certainly try to follow your advice next time.

[GregPlowman]

Replaced by #21695