• PackagingWG
• 2020-10-06-pip-teamwidemeeting

Tuesday 6 Oct 2020

Participants:

• Pradyun
• Georgia [left after first 30 min]
• Sumana

Agenda: what do we now know and what improvements we've made + what we can change -- make a decision tree, and figure things out -- and figure out if we're putting out 20.3 in October.

• Performance

  • https://github.com/pypa/pip/issues/8664

    • Pradyun needs to review https://github.com/pypa/pip/pull/8932 and probably merge it

    • 8932:

      • to add infra for 8932 on the resolvlib side https://github.com/sarugaku/resolvelib/pull/57 - is in progress being reviewed -- needs to happen on resolvlib side ... this logic is reusable and should be in resolvlib. Status quo *was* "put the logic in pip". Pradyun to talk with TP to discuss path forward.

      • would need to cut a resolvlib release, update vendored version in pip, update pip implementation -- this would take at least a day, maybe longer. ORRRRR merge #8932 in pip [first or instead], which would be fastest.

    • https://github.com/pypa/pip/issues/8664#issuecomment-702345229 worst slowdown we saw is about 3x

  • on informing user when resolver is doing a lot of backtracking
    • Pradyun and Bernard spoke today, and have a rough idea -- came up with a few ideas on how to present this, and Pradyun thought through how to implement it, what's doable, and now we're there in terms of spec, need implementation work + exact wording. We figured out where we want to print message, what info to present to user.
    • Question: is Bernard developing exact wording? [we believe so]
      • Georgia will confirm in the UX meeting. Relevant Etherpad: [removed, not sure whether it's a public link]

• releases & 20.3

  • how does Pradyun feel about new resolver as default in 20.3 in october?
    • we should be able to make release in second half of the month. Assuming we don't find a showstopper in a bug report in next few days

  • reviewing https://github.com/pypa/pip/projects/5 and https://github.com/pypa/pip/projects/6

  • Q: want to spend a minute to talk about 20.2.4 later.

    • SH: I think there's been some fairly clear message about what should go into 20.2.4 point releases. https://github.com/pypa/pip/issues/8511#issuecomment-701325392

      • about what should go in -- exact PRs available (merged or ez-to-merge)
        • option 1: cut a release off 20.2.3
        • option 2: cut a release off master

      • There are docs PRs we really want to get out, and there are performance & general improvements we really want to get out. 20.2.4 including performance improvements: we'd be able to talk to users who had perf problems and ask "does this help?"

      • BUT: this would be a point release -- we have merged a few things into master that cannot go into a point release. So, what would work well: bundling docs & performance improvements into 20.2.4 with them. Doable. Lets us have all the positives we talked about. BUT, tiny concern: we may have to change/update the documentation about how to use the new resolver by default.

        • https://github.com/pypa/pip/pull/8942

        • TODO: Sumana: write a ticket to write a PR changing docs about resolver migration/default/etc., and then we shall merge it just before cutting pip 20.3
  • More general bug triage from project TODO boards:

    • TODO: Sumana to ask Nicole about https://github.com/pypa/pip/issues/7744

    • TODO: Sumana to note on https://github.com/pypa/pip/issues/8683 that this in progress, move card to correct column -- also #8346 -- DONE

    • TODO: Sumana to close https://github.com/pypa/pip/issues/7871 and explain with link to these notes -- DONE

    • TODO: Sumana mark #8346 as not a blocker, explain on issue -- DONE

    • https://github.com/pypa/pip/issues/8076 -- Pradyun agrees that the "escape hatch" should be "the old resolver exists" but there is KIIIIIND of a legit use case where a dependency has declared [certain requirements] and I want to override. And there is precedent where other package managers do it. but in any case, this is NOT a blocker for 20.3 with new resolver being default.

      • TODO: Sumana to move out of blockers (DONE), and the bug reports in late Oct will tell us more about the use cases. Here we have 3 people saying "we want this"; need broader info for how much this is really needed and in which cases.

    • TODO: move https://github.com/pypa/pip/issues/8307 out of blockers and explain - remove from the board entirely -- DONE

    • https://github.com/pypa/pip/issues/8495 will be very rare. Really 2 issues: 1 if there is a diamond dependency with a conflict at the end. We print the package's name twice. Not ideal but not a big problem. The other issue, which Dustin has flagged, is that advice is not actionable for certain kinds of users. Valid point. There is a relatively straightforward approach to fixing this which Dustin has suggested.

      • TODO: develop concrete wording for error message change.
      • TODO: Sumana to move out of blockers column -- DONE. Pradyun says: nice to have this in 20.3 but I don't want to delay release for it. Nice to have in 20.3.

    • TODO: ask Nicole about progress and what's necessary regarding https://github.com/pypa/pip/issues/8492 -- "no blockers. status: need to make a PR for feature flag docs. Need to look into research question we will discuss later. If I have time this week, look into https://github.com/pypa/pip/issues/8492 to give user more info when they get ResolutionImpossible error. Would be nice to have that in by 20.2. But most importantly, don't make the new resolver default before this is sorted." -- quote from Nicole in July

    • TODO: Pradyun to talk with TP about https://github.com/pypa/pip/issues/8785 and figure out what we need to do next

    • TODO: Pradyun to review https://github.com/pypa/pip/issues/8827 , a bugfix PR that needs to be merged.