OSS 26.07

Source: sichard.ca/ghstats/pip

A maintainer shared this chart of open issues in the pip repo. (It’s been dropping since June.)

Most of the open issues are more than a year old. My guess is that pip has limited maintainer bandwidth, so recent issues and problems users hit directly have to rank higher than re-checking old ones. That’s probably why old issues stayed open long after they were already fixed.

I started contributing to pip recently. The motivation is simple. I wanted to contribute to a project with a long history and a complex structure. pip carries a lot of backward-compat code for legacy behavior, and plenty of code can’t be changed lightly because users already rely on bugs as features. (Even plain text output may be parsed by someone.)¹

These days AI agents write a lot of code in a day. Spending a long time discussing a one- or two-line change can feel slow, but I think it’s also what made the work interesting to me as a developer.

Work (July 12 - August 13)

Code

• pip
  • Fixed a bug where the self version check under a zipapp read the version of the pip installed in the environment instead of the running zipapp. #13084, #14217
  • Fixed a traceback when entry_points.txt contains a malformed entry. #8573, #14220
  • The (latest) marker in pip index versions was never printed. Changing the output could break tools that depend on its format, so I left the output alone and went with removing only the dead code. #14177, #14178
  • Corrected the macOS configuration docs to describe XDG_DATA_DIRS and XDG_DATA_HOME, which actually work, instead of XDG_CONFIG_*. #14200, #14207
  • Added a regression test that Requires-Python skip reasons appear without debug logging. #14170
  • The Git partial clone opt-out environment variable from last month got merged. #14163

Triage

• pip
  • Confirmed that already-fixed issues no longer reproduce, then closed them. #10620, #11572, #12061, #8949, #9512, #8582, #10476, #14196
  • Closed issues that were external problems, not pip’s. #14214, #14224
  • Triaged duplicate issues. #14216, #14244
  • Checked whether these issues still happen today. #9034, #13147, #14183

Reviews

• pip
  • A bug where pip <subcommand> --version just ran without an error. (By design only pip --version is supported.) #14231
  • Reproduced the issue where pip list --outdated ignored --uploaded-prior-to. #14189, #14190
  • A refactoring PR splitting exceptions.py into separate modules. (Seems like a good improvement.) #14176
  • A PR removing shell=True from pip config edit. The behavior it was changing looked intentional, so I dug through the history before reviewing. #14180
• django-auditlog
  • Checked and approved the PR making the LogEntry model swappable. #805
  • Approved a docstring fix for the Action class, outdated since Action.ACCESS was added. #817

Next goals

pip

• Improve pip download and pip install --target, which leave incomplete files behind when the disk fills up. The same pattern appears elsewhere in pip, so I’m aiming for a reusable fix. #9964
• A bug where the “no matching distributions” hint appeared for packages that were only excluded by version conflicts. The fix is waiting for review. #14193, #14194
• A PR that warns about ~...dist-info directories left by incomplete removals. Still working on it. #14047
• Imports break when a submodule is split into a namespace package and the new version is installed over an existing install. A maintainer suggested this one. The history is fairly involved, so I’ll set aside time for it later. #12211

django-auditlog

• Working on change detection comparing raw values and serialized values separately. #835
• A workflow that checks in CI whether a PR updated the CHANGELOG. (It seems to get missed often…) #836, #837

Footnotes

1. xkcd 1172: Workflow ↩