Improve the message for "Resolution Conflict" errors

[pfmoore]

Fixes #8377

@nlhkabu @ei8fdb comments appreciated 🙂

[pfmoore]

Here's a nasty edge case that the current code does not handle at all well (based on the issue in #8131):

If we have a local directory "pkg" containing the source for "pkg 0.1.0" and the user runs pip install ./pkg "pkg != 0.1.0" they will get the following error:

ERROR: Cannot install oops because these package versions have conflicting dependencies.
The conflict is caused by:
    The user requested ExplicitRequirement(LinkCandidate('file:///C:/Users/Gustav/AppData/Local/Temp/pytest-of-Gustav/pytest-114/test_new_resolver_local_and_re0/workspace/scratch/pkg'))
    The user requested pkg!=0.1.0

There are a number of possible solutions. For instructions on how to do these steps visit: https://pypa.io/SomeLink
ERROR: No matching distribution found for pkg

Problems with this:

1. That "Cannot install oops" is a placeholder, because none of the causes have "parent" values for us to report.
2. That ExplicitRequirement(...) is hopeless. We need to trace back from that to the original directory name the user supplied. We may have that info (it gets passed around in came_from attributes, but we may have lost them when we build candidates from requirements).

More work to do here...

Interestingly, we didn't have any sort of test for this case. I've added one as the first stage of fixing the issues 🙂

[pfmoore]

Converting this to a draft so we don't accidentally merge it.

[uranusjr]

Regarding the ExplicitRequirement problem, IMO it’s good enough if we return the absolute path, if that’s as far as we can get.

[pfmoore]

IMO it’s good enough if we return the absolute path

I'm pretty sure getting any closer to "what the user typed" than an absolute path is going to be far more work than the improvement in the message would justify.

If the UX people find, in their research, that it is worth that effort, we can iterate on it later. I'm going to stick with absolute path for now. (Even that likely needs a new method on the Candidate, to abstract away the difference between linked candidates and other types...)

[uranusjr]

If we’re adding a method, would it feel more comfortable to have a format_for_error() or similar to handle all the “show a human-readable representation for this requirement” logic?

[pfmoore]

Yeah, that sounds like a nice idea :-)

[pfmoore]

OK, I think I've fixed that case.

[uranusjr]

Does this work for remote links?

[pfmoore]

OK, I guess that's another test to add :-)
Thanks.

[pfmoore]

I added a fix, but couldn't work out a reasonable way to add a test. Suggestions welcome :-)

[BrownTruck]

Hello!

I am an automated bot and I have noticed that this pull request is not currently able to be merged. If you are able to either merge the master branch into this pull request or rebase this pull request against master then it will be eligible for code review and hopefully merging!

[pradyunsg]

One thought about how the cached packages/links look, rest seems OK to me.

[pradyunsg]

How does this look for cached paths?

[pfmoore]

I'll try it tomorrow, but from my recollection "annoyingly long" ;-)

[pradyunsg]

Can we use the original link w/ (cached) in the name in that case?

OK to do this in a follow up, if you file an issue to aggregate all the call sites that print req.link.

[pfmoore]

OK, I've opened #8463 for the follow-up work requested by @pradyunsg. Merging this now, as it's clearly better than what we currently have. We can easily make any further improvements in follow-up PRs.