[CI] use actions/{upload,download}-artifact@v4

[NilashishC]

SUMMARY

As per the failure in https://github.com/ansible/pylibssh/actions/runs/13234496470/job/36936955478?pr=675, v3 of actions/upload-artifact and actions/download-artifact are deprecated and cannot be used. This PR updates it to v4.

ISSUE TYPE

• Maintenance Pull Request

[NilashishC]

@webknjaz Hi. Could you please have a look at this PR? This needs to be merged before #675.

Thank you.

[webknjaz]

@NilashishC this is a move in the right direction. It is, however, incomplete.

You have to upgrade re-actors/checkout-python-sdist@release/v1 to re-actors/checkout-python-sdist@release/v2 (because v1 uses an older version of the download action).

And additionally, the wheel uploads must change their artifact names to be unique (example: https://github.com/aio-libs/yarl/blob/07c1b4f/.github/workflows/reusable-build-wheel.yml#L92C15-L95C28).

Finally, the download step in the publishing jobs must change to merge+pattern (example: https://github.com/aio-libs/yarl/blob/07c1b4f/.github/workflows/ci-cd.yml#L532C9-L533C29). The last one will not be testable in the PR CI runs and will only fail later if you don't get it right.

[NilashishC]

@webknjaz thanks for the review.

The variables required to make the wheel name unique in reusable-build-wheel.yml seems to be commented out in inputs - https://github.com/ansible/pylibssh/blob/devel/.github/workflows/reusable-build-wheel.yml#L17-L26. Should I uncomment those and add it to the wheel name? Or should I choose other inputs.

[NilashishC]

@webknjaz thanks for the review.

The variables required to make the wheel name unique in reusable-build-wheel.yml seems to be commented out in inputs - https://github.com/ansible/pylibssh/blob/devel/.github/workflows/reusable-build-wheel.yml#L17-L26. Should I uncomment those and add it to the wheel name? Or should I choose other inputs.

Okay, I did try something to make it unique - https://github.com/ansible/pylibssh/pull/676/files#diff-3826dffee70012c1e07ee3d63a4cf5ba7ed82c17bbac68145432a97e71dbade7R187-R192.

However, now some of the build jobs in the matrix fail.

[webknjaz]

@NilashishC your commit message says

use merge+pattern for download artifact

But that commit only makes use of the merge input and doesn't replace name with pattern. So these jobs only download the sdist while they expect the wheels to be present.

There might be other problems (I saw some weirdness in the logs) but I would imagine that those are unrelated. Fix this first and we'll see how it goes then.

[NilashishC]

@webknjaz Apologies for my lack of understanding here. But I'm unsure what the pattern would be . The patterns differ in reusable-cibuildwheel vs reusable-build-wheel (because the inputs are different). How do you recommend we handle this in the download steps?

[webknjaz]

Wouldn't just adding an asterisk at the end do it? This should work for as long as the prefix remains the same.

[NilashishC]

So, all x86 ones are passing, but not the other architectures.

[webknjaz]

@NilashishC will you be cleaning up the commits?

[webknjaz]

So, all x86 ones are passing, but not the other architectures.

As for https://github.com/ansible/pylibssh/actions/runs/13268100337/job/37040748168?pr=676#step:9:50, it's probably a separate issue that I'd look into separately from this PR.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[NilashishC]

@webknjaz I think I have resolved all your comments now. Here's the PR for actions/cache - #677.

[NilashishC]

@NilashishC will you be cleaning up the commits?

Do you want me to squash them into a single one?

[webknjaz]

Do you want me to squash them into a single one?

It's up to you now. I like a series of atomic commits. Not a blind squash. Currently, the repo still allows the fake merge methods like “squash” or “rebase” but I'm eventually going to enable the merge queues feature which will take those away.
So if you want this to be accepted via the squash method, it's fine for now. Just make sure there's nothing unrelated to the title of the PR in the diff.

[webknjaz]

Looks like one more factor is missing here:

Suggested change

	${{ inputs.manylinux-image-target-arch }}-
	${{ inputs.manylinux-image-target-arch }}-
	${{ inputs.manylinux-image-target-qemu-arch }}-

It's not breaking the CI in PR builds because a portion of jobs are being skipped. However, it'll likely explode on release.

[webknjaz]

Urgh.. https://github.com/ansible/pylibssh/actions/runs/13306998592/job/37160303882?pr=676#step:6:15 fails due to asterisks. Looks like that inputs hashing idea will probably have to make its way into this PR's scope. I'll think about it and maybe make a direct commit into your branch.