Add sphinx-builder linkcheck

[LecrisUT]

Depends on: #2483

There are a bunch of linkcheck failures, many of which come from the auto-generated pages. Not sure how these should be handled

[martinhoyer]

@happz looks like we have a few more broken links being generated.
No idea about this one ( spec/core: line 131) broken None -
Test docs, like /tests/execute/results-custom.fmf -> tests/execute/result/custom

Looking into how it's being generated.

[LecrisUT]

Oh wow, this is what it expanded for me:

* Verified by `/tests/core/adjust <https://github.com/LecrisUT/tmt/tree/feat/linkcheck/tests/core/adjust/main.fmf>`_ 

It seems the generation automatically puts in the git repo and branch, but then on a PR, the git repo can be either target or source depending on the checkout. I think that is where it trips up 🤔

[happz]

@happz looks like we have a few more broken links being generated.

Yes, my #2940 did not try to address everything. There's #2868 for the links to tmt internals and tests as created by the template, which seems to be the vast majority of the remaining broken links.stories

No idea about this one ( spec/core: line 131) broken None - Test docs, like /tests/execute/results-custom.fmf -> tests/execute/result/custom

Looking into how it's being generated.

Check out #2868 (comment). I'd be happy to look into fixing this, since it's the code I wrote and left in less than ideal state, but if you feel like diving in, be my guest :)

Today I'd go with the idea mentioned in the last paragraph, i.e. moving the macro away from the template, making it a full-fledged Python code & expose it back to the template, mostly because it can be tested easily. I'd add a couple of unit tests to verify it's rendering links correctly - should be mostly Link on the input, str on the output, therefore easy to parameterize with a pile of test cases. Feel free to ping me if you hit some dumb piece of code.

Oh wow, this is what it expanded for me:

* Verified by `/tests/core/adjust <https://github.com/LecrisUT/tmt/tree/feat/linkcheck/tests/core/adjust/main.fmf>`_ 

It seems the generation automatically puts in the git repo and branch, but then on a PR, the git repo can be either target or source depending on the checkout. I think that is where it trips up 🤔

That's one issue for sure, there are more though, e.g. some links lead to main.fmf when the test is not defined in main.fmf and the file has a different name, and so on. Basically the macro tries to guess, and in some cases it's wrong.

[martinhoyer]

Thanks @happz!

if you feel like diving in, be my guest :)

I'll be afk for the rest of this week unfortunately. Also, I've familiarized myself with the templates a bit and think I get the general idea, but there is still plenty of things I'm seeing for the first time to catch-up with.

[happz]

With #3001 applied, I get no broken links reported from the linkcheck builder.

[LecrisUT]

With #3001 applied, I get no broken links reported from the linkcheck builder.

Let me rebase on your commit first just to make sure it works for forks as well

[LecrisUT]

The failure is because it is run on a fork, these are resolved in #3108

[martinhoyer]

The failure is because it is run on a fork, these are resolved in #3108

If it's because of fork, wouldn't it break more links than just one?

[LecrisUT]

I don't quite remember, but there were a bunch of broken links inside the rst generated file, but it so happened that linkcheck aggregated the report or something, don't quite remember.

On RTD it showed fine because it uses a different git checkout approach and it doesn't fail here:

tmt/tmt/utils/__init__.py

Lines 4318 to 4325 in 5fbbd89

	ref = run(Command("git", "rev-parse", "--abbrev-ref", "HEAD"))
	if ref != fmf_id.default_branch or always_get_ref:
	fmf_id.ref = ref
	else:
	# Note that it is a valid configuration without having a default
	# branch here. Consumers of returned fmf_id object should check
	# the fmf_id contains everything they need.
	fmf_id.ref = None

But git checkout in GH actions checksout the merge commitand that's where it breaks. Check the log of the checkout action to replicate locally.

[thrix]

LGTM

[happz]

/packit build

[happz]

/packit build

[happz]

Hmm, blocked by #3108 :/

[LecrisUT]

Hmm, blocked by #3108 :/

Or maybe superseded-by since that one contains the commit from here

[psss]

Or maybe superseded-by since that one contains the commit from here

Let's keep the linkcheck change here and remove it from #3108.

[psss]

/packit build

[psss]

The extended unit tests failure will be fixed by #3227. Remaining failure is irrelevant.