Comparing changes

Closes #2021 The session-scoped `utils_temp_dir` fixture symlinks `git` (and `uv`) into a temp directory that is then put on `PATH`. The symlink call was guarded only by `suppress(FileExistsError)`, so on Windows without elevation or Developer Mode it raises `OSError: [WinError 1314]` and errors every test depending on `pipx_temp_env` — effectively the whole suite. CI does not see this because the `windows-2025` runner can create symlinks, so the suite is green there and unrunnable on an ordinary Windows machine. This falls back to putting the utility's real directory on `PATH` when the symlink cannot be created. Copying or hard linking is not an alternative: Windows resolves an executable's DLLs relative to its image path, and `git.exe` needs the siblings in its own directory. ### Testing `tests/test_uninstall_all.py` on Windows 11, Python 3.13: **5 errors → 5 passed** (verified both before and after the change). A `changelog.d` fragment is included. --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Bernát Gábor <gaborjbernat@gmail.com>

`pipx environment --value PIPX_MAX_LOGS` exits 2 with `argument --value/-V: invalid choice: 'PIPX_MAX_LOGS'`, and a plain `pipx environment` never prints the variable, though `main.py` reads it to prune the log directory and `docs/reference/environment-variables.rst` lists it as settable. Of the 26 names in that document it is the only one `ENVIRONMENT_VARIABLES` omits; the derived table matches 9 for 9. Why the suite is quiet about it: `test_cli` and `test_cli_global` close with `for env_var in ENVIRONMENT_VARIABLES: assert env_var in captured.out`, so the oracle is the same list the command prints from and a missing name cannot fail it. `test_cli_with_args` checks a hand-written subset, and its rejection case passes `"SSS"`, which is not a pipx variable at all. The new parity test reads the documentation table as its oracle instead, in the shape of `test_docs_urls.py`. Reverting the one-line source change fails that test's settable case and the new `PIPX_MAX_LOGS` parameter of `test_cli_with_user_environment_value`; the derived case passes either way and is there as a pin. `PIPX_MAX_LOGS` stays out of `_get_derived_values()`, so it prints raw and empty when unset, like every other settable-only variable, matching where the documentation files it. Same shape as #1493, which added `PIPX_GLOBAL_*` to this list after #1492 reported them missing from the documentation and from `pipx environment`. `pre-commit run --all-files` passes; `ty check` matches the clean checkout. ______________________________________________________________________ - [x] ran the linter to address style issues (`pre-commit run --all-files`) - [x] wrote descriptive pull request text - [x] ensured there are test(s) validating the fix - [x] added news fragment in `changelog.d/` folder - [ ] updated/extended the documentation — no change needed, `docs/reference/environment-variables.rst` documents `PIPX_MAX_LOGS` today and the code is what disagreed with it. --------- Co-authored-by: Dylan Pulver <dylanpulver@users.noreply.github.com> Co-authored-by: Bernát Gábor <gaborjbernat@gmail.com>

`main` has been red for days, and every open PR inherits it. Three tests fail in rotation across runners, and all three trace back to the suite reaching the real internet. `test_install_fetch_missing_python_invalid` asks pipx for a standalone Python 3.0 without mocking the release index, and `test_run_script_from_internet` fetches a gist from githubusercontent. When GitHub answers `403: rate limit exceeded` urllib leaves the `SSLSocket` unclosed, and since `filterwarnings = ["error"]` the resulting `ResourceWarning` reaches pytest's unraisable collector and fails whichever test the collector happens to interrupt. That is why `test_install_run_in_separate_directory`, which opens no connection of its own, has been failing as often as the two that do. The fetch test now takes the `mocked_github_api` fixture the interpreter tests already use, and the script test now serves its payload over loopback so the URL parsing, size cap and decode still run against the real code path. On top of that an autouse fixture refuses any connection to a non-loopback address, so the next test that reaches out fails on its own line with the address it dialled rather than surfacing minutes later as somebody else's `ResourceWarning`. `--net-pypiserver` and `--all-packages` want real indexes and opt out. Separately, `test_uninject_running_app` opened three pipes on its `Popen` and never closed them, which is the Windows half of the same red: the GC finalizes the pipes whenever it gets there and the `ResourceWarning` errors out. `Popen` as a context manager closes all three. I ran the full suite with the guard armed, which is how the gist fetch surfaced.

Navigation Menu

Repository navigation

Comparing changes

Open a pull request

Commits on Aug 30, 2026

This comparison is taking too long to generate.

Uh oh!