hatch-1.3.1/hatchling-1.6.0

[jcgruenhage]

Testing the changes

• I tested the changes in this PR: briefly

I've tried building and testing all packages that depend on hatchling, and except for hatch everything went smoothly. The test failures in hatch don't look very problematic to me though. Should we just deselect those?

Log: https://pastebin.com/raw/uywdGiZY

[icp1994]

Shouldn't python3-editables be a dependency for hatchling? That one is not packaged yet it seems.

[jcgruenhage]

yeah, should be, although it seems unrelated to the problem

[jcgruenhage]

(As in: I tested it locally again with the update to the newer hatch/hatchling versions and there the tests passed again, and only after that I also added the editables dep)

[icp1994]

Ah I see the editable tests were already disabled via make_check_args

[jcgruenhage]

yeah... ._.

I've tried re-enabling them after I had added the editables dep, but they still seem to fail. I'll try to figure out why they still fail

[jcgruenhage]

So, I've updated this again, and there's been new tests failing that weren't among the failed tests yet. I haven't had the time to investigate this, but I don't want to have a longer and longer list of failures here, so in case anyone else has the time to help investigate the test failures building up for hatch, I'd appreciate it.

[icp1994]

After some tinkering around, I think I have figured out what's causing the new errors. I'll list the reasoning sequentially for easier reading.

1. The tests for hatchling comes bundled with the hatch source in tests/backend folder.
2. At the time of hatch-1.3.1 release, the latest release of hatchling was 1.5.0. The tests/backend inside hatch-1.3.1.tar.gz were therefore meant to run alongside hatchling-1.5.0.
3. But with the current template, building hatch-1.3.1 pulls hatchling-1.6.0 as dependency so some of the old tests which aren't compatible anymore with the newer hatchling version fails.

We should be able to merge hatchling-1.5.0 and hatch-1.3.1 first and then hatchling-1.6.0 in a separate commit since the hatchling package runs no tests. Or we could decouple hatchling tests from hatch so this headache doesn't occur in the future whenever we skip a couple minor updates and the tests become incompatible.

[jcgruenhage]

Thanks for doing the detective work, totally makes sense! I'm currently trying what happens if I just ignore backend/tests and tests/backend in hatch, then it should work just fine. I prefer this over carefully updating the versions in a specific order, because that'll cause headaches in the future as well.

I'd still like to get the tests in hatchling to work though, so that these tests are only ignored when building hatch, but actually are being run when building hatchling.

[jcgruenhage]

Let's see what the CI says, but it builds successfully locally, tests pass correctly!