PackagingWG/2020-11-02-pip-teamwide-meeting

Legacy Wiki Page

This page was migrated from the old MoinMoin-based wiki. Information may be outdated or no longer applicable. For current documentation, see python.org.

Monday, 2 Nov 2020

Participants:

  • Pradyun

  • Sumana

  • Georgia

Agenda:

  • Yay, beta is out!

    • Feels like we checked the checkbox! :-)

  • Status & blockers

    • UX team: no blockers to report. status: analyzing surveys, still doing some interviews. Will check further re blockers.

    • Pradyun: beta is out! issues: mostly as listed below.

    • Sumana: did some triage + replaying to issue. some announcements (not widespread, likely today). making a “things that I do after a pip release” list.

  • Note: Ernest has a week off.

  • Invoices!

  • issue followup on things that we would like to finish before stable 20.3 release (from https://github.com/pypa/pip/projects/5 , https://github.com/pypa/pip/projects/6 , and https://github.com/pypa/pip/milestone/38 )

    • New resolver: Failure despite correct version numbers, when extras and already-installed package are encountered together #8785 https://github.com/pypa/pip/issues/8785

      • last TODO was that Pradyun would investigate - any outcomes?

        • Not yet – haven’t looked into this since the call w/ TP.

        • Upon re-look: hairy problem. Bunch of edge case. Surfacing more often than we had expected. Needs to be fixed before 20.3.

        • TODO: Will need 3-4 hrs of Pradyun diving into it, connecting in currently unresolved issues.

    • https://github.com/pypa/pip/issues/8380 ResolutionTooDeep{.backtick} error message

      • Have we resolved this with some other PR that maybe didn’t get attached?

      • We basically removed the ResolutionTooDeep{.backtick} error in https://github.com/pypa/pip/pull/8275, because we figured “how many package names have we looked at” is a bad metric. :)

        • We should probably to reintroduce this, as a hard-stop error at 1000 backtracks on the same package name.

        • should this block the 20.3 release? It’s nice-to-have for when resolver behaves badly, as in #9011…. current threshold of 2 million is A LOT.

        • ResolutionTooDeep{.backtick} – we changed it to 2 million package names, so no one ever sees it

          • but users will get backtracking error message

        • PG: should not block 20.3 release

          • SH: ok with not blocking the release on re-introducing this error. the backtracking provides context to the users.

          • GA: sounds good.

          • TODO: Sumana to explain this on the bug and move it out of blockers

    • https://github.com/pypa/pip/issues/9011 Pip 20.2.4 goes into infinite resolution of dependencies #9011

      • What’s next here? Investigation or implementation?

        • Investigation needed, seems a resolvelib bug.

        • Definitely a blocker for 20.3.

        • Not sure how likely it is that users hit this, since “when does this happen” is not clear.

          • But users hit this already, so, likely definitely gonna be a problem if not solved.

        • TODO: Pradyun to investigate [probably to talk with TP as well]

    • printing warnings about pip potentially introducing conflicts involving existing dependencies: https://github.com/pypa/pip/issues/7744#issuecomment-717773481

      • What needs to happen next here? Implementation, or more UX work to polish the warnings/information messages?

      • Implementation, and then *maybe* another round of reviews if I was wrong about what’s do-able.

      • TODO: Pradyun to implement

    • Including backtracking in the user guide https://github.com/pypa/pip/pull/9040

      • incorporate Sumana’s commit and merge?

      • TODO: Pradyun to ^

    • https://github.com/pypa/pip/issues/8495 New resolver error message is confusing if a package has inconsistent dependencies #8495

      • Is this post-release work or do we need it for 20.3?

      • Should be doable for 20.3, but I wouldn’t block the release for it.

      • TODO: Sumana to note in issue: not a 20.3 blocker

    • https://github.com/pypa/pip/issues/9083 numpy, PEP 440, comparison, version normalization

      • How hard is this to fix?

        • Has an approved PR, can be merged as is (waiting on writing a test tho): https://github.com/pypa/pip/pull/9085

        • Pradyun resigns himself to making the 5-line test

        • TODO: implement test, merge