py domain: Ignore aliases for resolving :any: cross-references

[ewjoachim]

Subject: closes #10088 (EDIT: already closed as duplicate) and also closes #9577 I believe, since they are duplicates.
With this PR, we can use :any: again in Sphinx v4.x with autodoc & python aliases.
I believe that everything that used to work will still work

Feature or Bugfix

• Bugfix

Purpose

See #10088
The issue is that that autodoc adds canonical (which is a good thing), and any doesn't play well with aliased objects (which is the bad thing we're solving here)

Detail

What I did, it's all in the python domain code:

• If there are more than one matches when resolving any, we remove aliased nodes from matches.
  • either this brings the number of matches to 1 and it's all fine
  • or there's still more than one match and it means the issue is real.
• In the tests with the canonical option, I've added multiple any lookups, and the test checks that there is no warning.

Just to be sure, I've run the test before the fix and it's red, and after the fix, it's green.

I don't know if this qualifies for a "hotfix" (not sure).

Relates

• Sphinx 4.x + :any: + autodoc + aliasing objects = warning #10088 "More than one target found" when all targets actually resolve to the same point #9577
• Sphinx 4.0 finds more than one target for cross-reference #9205 was the same issue for normal xref resolutions, while this solves any xref resolutions

[ewjoachim]

Hm, hey, anyone ?

[ewjoachim]

@tk0miya if you have any lead as to what I can do to help this PR go forward ?

[AA-Turner]

Skips a list comprehension step

Suggested change

	if len(matches) > 1:
	# Remove all duplicated matches
	matches = [(name, obj) for name, obj in matches if not obj.aliased]
	for name, obj in matches:
	if obj[2] == 'module':
	for name, obj in matches:
	if len(matches) > 1 and obj.aliased:
	# Skip duplicated matches
	continue
	if obj[2] == 'module':

A

[ewjoachim]

Done (I've stored len(matches) > 1 in a variable)

[AA-Turner]

Please update CHANGES for 5.x

A

[ewjoachim]

We should be all good now :)

[ewjoachim]

Rebased after 5.0 release :)

[ewjoachim]

(is it expected that the bug label has not been set on the PR ?)

[ewjoachim]

Rebased after 5.0.1. Also, see #10515

[ewjoachim]

Oh, I see you added this to the 5.1.0 milestone, I'll update CHANGES so that it appears in the 5.1.0 section :)

[AA-Turner]

Thanks @ewjoachim!

A