[Distributed] don't sidestep require logic

[vchuravy]

We were trying to be clever, but there is logic in require that we shouldn't sidestep.

fixes: #26805

[ararslan]

Can you add a test to ensure that the behavior noted in the linked issue doesn't regress in the future?

[vchuravy]

@StefanKarpinski is there an easy way to generate a fresh environment with a stub package for testing, without replicating most of test/loading.jl?

[StefanKarpinski]

You could just reuse test/project if you want. It doesn't currently have any top-level source but you could turn it into a package that you can load for this test, which would have the side effect of making sure that all the package resolution stuff works in the distributed case as well.

[vchuravy]

I don't have time right now to do that kind of work, so I propose either waiting for another week or merging as is and I will come back to add tests.

[StefanKarpinski]

I'm fine with either. This is a bug-fix so doesn't block the alpha release.

[JeffBezanson]

Merge?

[ararslan]

Would be good to have a test.

[StefanKarpinski]

Insisting on a test has blocked merging this for three months already.

[ararslan]

Why would we merge it now without tests if we weren't going to merge it before without tests?

[StefanKarpinski]

So let me get this clear: you'd rather leave this broken than have it fixed but untested? That's really putting the process cart before the horse of quality. Please, feel free to write a test by all means, but I mainly want the bug fixed.

[KristofferC]

If the choice is between no test but works and no test and doesn't work, seems like the former is better?

[ararslan]

I'm not saying we absolutely need to add a test here. I'm saying we should, since a bugfix should have verification that it fixes a bug. I'm just confused as to why there seems to be little interest in testing this.

[StefanKarpinski]

There's not a lack of interest in testing, but this seems like a serious pain to test, no one has written tests and we're running out of time.

[vchuravy]

I have a test incoming, apologies for the delay, JuliaCon took up a lot of time in the last months.

[vchuravy]

Added a test that was developed against master where it fails with:

make -C test Distributed
make: Entering directory '/home/vchuravy/src/julia/test'
    JULIA test/Distributed
Test    (Worker) | Time (s) | GC (s) | GC % | Alloc (MB) | RSS (MB)
┌ Warning: Replacing module `TestPkg`
└ @ Base loading.jl:872
┌ Warning: Replacing module `TestPkg`
└ @ Base loading.jl:872
┌ Warning: Replacing module `TestPkg`
└ @ Base loading.jl:872
      From worker 3:	WARNING: using TestPkg.TestPkg in module Main conflicts with an existing identifier.
      From worker 4:	WARNING: using TestPkg.TestPkg in module Main conflicts with an existing identifier.
      From worker 2:	WARNING: using TestPkg.TestPkg in module Main conflicts with an existing identifier.
ERROR: LoadError: On worker 2:
importing TestPkg into Main conflicts with an existing identifier