Packaging/2020-05-11-pip¶

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 Information may be outdated or no longer applicable.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.

11 May

Participants:

Paul
Pradyun
TP

Agenda:

- Pradyun is happy that someone else closed the meeting notes like he’s been doing for so long! (with [syntax])

- UX + Pradyun syncups!

Circular dependency figured out
UX team + Ilan will collect test cases
Main problem now:
- What metrics to use for the “too complicated” error?
- TP: Maximum versions tried per package, sounds perfect!
  - Everyone agrees. Paul: Doesn’t sound like it’d be too difficult in resolvelib.
- Paul: When we do the finder logic in the provider, there’s never any reason we’d return 2 compatible ones.
  - will ask on Zulip, since this is a bigger topic that might not need much discussion or need lots of discussion. “)

- PubGrub:

TP: Was reading PubGrub before this call. Current takeaway is that Python specifiers is missing – negation: Django > 3.0{.backtick} to Django <= 3.0{.backtick}. We’d need to implement negation (non trivial!)
TP: Implementing incompatibility tracking would be better; since it’s significantly faster.
TP: Does not make sense to switch to PubGrub now (Python packaging logic lacks negation, which is essential for PubGrub)

- What are we working on next?

Paul: Couple/Three of things
- Tidying up anything that comes out of the constraints issue (2 failing tests that TP found)
  - As long as it works for OpenStack, what we should do is remove those tests, and declare that constraints aren’t meant to do that anyway.
  - Will add / improve documentation.
    - Only valid as "project <= version"{.backtick} and similar forms.
    - Pradyun: Validation?
      - Paul: Don’t want to get side-tracked by that entire discussion – will do if it’s easy. Will come back with “let’s not bother” if it’s hard.
- Documentation of update mode (reminded by an issue mention/notification)
  - Will have another pass.
  - Wanna get some writeup that talks about what we do. The longer I take, the more annoyed at this situation.
  - Wanna be less annoyed at how complicated it is!
- Refactoring Requirements-bit, so that they’re on the provider instead.
  - It also makes sense to move upgrade_strategy{.backtick}.
  - Have the factory be responsible for get the candidates from Finder.
  - Provider puts them in the correct order.
  - All ordering logic in Provider. All internal-interaction with Factory.
    - Pradyun: we should document this structure is being taken.
    - TP: since we need to rewrite find_matches{.backtick}, I was planning on adding a new class for it – your plan sounds equally good.
    - Paul: This is where the question about more-than-one-candidate-per-question.
  - Particularly, when it’s doing the matching by versions etc in the sorting.
  - It might be easier if the factory to return 1 candidate per version.
  - [Pradyun (note taker) got distracted/zoned out]
  - Paul: That sounds like what we wanna do.
TP:
- Yanked
  - depends on https://github.com/pypa/pip/pull/8066
- find_matches(){.backtick} rewrite
  - Might conflict with Paul’s planned refactor.
  - Paul: Tell me to stop, if there’s conflicts.
  - TP: I will likely be introducing a new class.
- TP: need to look at the tests.
- Pradyun: we should look at all the issues, and write down something about this.
  - Paul: would be really nice.
  - TP: I tried doing that – I got stuck on the constraints
  - [we talk about how we get stuck while doing this – it’s horrible or I can try this (few minutes later) this is horrible]
  - while we’re looking at the tests, it would be nice to finally thrash out how “old resolver test” vs “new resolver” tests should work.
    - might be check the environment.
    - Add a --new-resolver{.backtick} option in pytest and check that throughout the suite
      - interaction with YAML tests
Pradyun
- Actually triage issues (lol)
  - Did a bit last week, but had to walk away after getting frustrated with the in-place builds responses (and landing on #5599).
- Get 20.1.1 out.
  - Reverting in-place-builds.
    - PR
    - Paul: we’ve got to do it, because of the issues.
      - We’ve had a couple of people who’ve come in and say “this is cool!”
      - Should give a heads up to those people.
      - Reopen issues, telling people why we had to do the revert.
      - Paul to do this on one issue. Pradyun will copy-paste on others, after reverting.
- Wrap up visualization stuff
  - Put it in a library / tool, instead of a bunch of file in my local resolvelib.
- Look at the failures
  - Note down what’s happening here.
  - Need to put a clock or something, to keep me from digging too deep.
- Look into the .[extras]{.backtick} (stretch goal)
- TP: I’d be very impressed if you make it past the first 2 bullets this week.
- Paul: Yea, that’s a lot of stuff.

Constraints – covered above