Fixed bug in extras handling for link requirements #12392
Conversation
| return base | ||
| return self._make_extras_candidate(base, extras, comes_from=template) | ||
|
|
||
| def _make_base_candidate_from_link( |
There was a problem hiding this comment.
_make_candidate_from_link actually always returned a BaseCandidate when no extras where specified. The new extras handling requires a base candidate so I extracted this part.
|
@uranusjr if you are available to review this that would be super helpful. |
|
Recently created a workflow which involves installing local extras, tested this branch and everything works fine. |
|
I'm preparing to do the last 23.3 patch release and I'm trying to determine if this PR fixes a regression of 23.3. To do so, I've tried to run the test added in this PR (test_new_resolver_constraint_on_link_with_extra) and it looks like it fails equally in 23.1.2 and 23.2.1. @sanderr Could it be this PR addresses something new, or that the test you added does not actually reveal the regression but something else? |
|
@sbidoul This is definitely a fix for a regression in 23.3. But you're right that the test case doesn't cover the regression directly. I have to admit I can't exactly recall all details, but I believe the test case covers a case that is strictly broader than the regression. Either way, I've just added a second test case that more closely resembles the scenario reported in #12373, and I've confirmed that it succeeds on both this branch and 23.2.1, while it fails on 23.3.1. |
|
Thank you for your work on this @sanderr. |
Fixed bug in extras handling where constraints on a package are not compatible with an explicit link candidate that also has an extra.
Closes #12372