settings: parse UV_SYSTEM_PYTHON in EnvironmentOptions - #21175
Closed
VedantMadane wants to merge 80 commits into
Closed
VedantMadane wants to merge 80 commits into
VedantMadane wants to merge 80 commits into
Conversation
Member
|
As it stands, this change alone doesn't really address the issue? You've added the parsing bit, but you're not using it anywhere, and we're still using clap to parse this variable in 11 places... |
VedantMadane
force-pushed
the
env-options-system-python
branch
from
August 22, 2026 13:21
73cb939 to
9007ac5
Compare
…1075) `PYTHONEXECUTABLE` and `__PYVENV_LAUNCHER__` can change the executable and virtual environment reported by a Python interpreter, but interpreter metadata was cached using only the absolute and canonical paths of the requested executable. Include both launcher overrides in the interpreter cache key so metadata from one launcher context cannot be reused in another. See astral-sh#21062. --------- Co-authored-by: Zanie Blue <contact@zanie.dev>
…ace (astral-sh#20944) Co-authored-by: William Woodruff <william@yossarian.net> Signed-off-by: William Woodruff <william@yossarian.net>
…1079) This extends our automations for issue triage to take reproducible bugs with a regression-test and attempt to generate a PR fixing the issue. - Context from the issue triage and reproduction are carried over into the job to attempt to fix the bug - The issue branch is updated with details on the fix attempt and result - If a fix is proposed, a draft PR is opened stacked on the regression test PR This also introduces re-usable actions for loading the issue context and codex thread from another workflow. --------- Co-authored-by: Ee Durbin <ee@astral.sh> Co-authored-by: Zanie Blue <contact@zanie.dev>
## Summary We have `from_registry_requirement` that does the equivalent of all of this code. (Cleanup in preparation for PR to fix astral-sh#20477.) ## Test Plan Existing coverage.
Reverts astral-sh#21055, CodSpeed team reports that their backend can now support uv's samply results
Projects with checked-in type stubs may not require building their native extensions for type checking, but `uv check` always installs the project while synchronizing its dependencies. Support `--no-install-project` and `UV_NO_INSTALL_PROJECT` so those projects can install their dependencies and run checks without building or installing themselves. Fixes astral-sh#21083. Co-authored-by: Zanie Blue <contact@zanie.dev>
missed policy for sts in astral-sh#21079
Invalid `uv version --bump` values currently produce unstyled errors without a trailing newline. The custom parser introduced in astral-sh#16555 returned raw, unformatted `clap` errors; construct command-aware errors instead so malformed bump values use the standard styling, usage guidance, and line termination. Fixes astral-sh#21063. Co-authored-by: Zanie Blue <contact@zanie.dev>
Reverts astral-sh#21087, issue_comment is not supported 🤦🏼
## Summary Previously, uv tracked Git metadata incorrectly in several cases: - Linked worktrees watched their entire worktree metadata directory, so staging an unrelated file invalidated the build even when the commit had not changed. - Branch references updated outside the worktree were not watched, potentially leaving embedded commit hashes, dates, tags, and tag distances stale. - Packed branch references and relative worktree paths registered nonexistent Cargo inputs, forcing a rebuild on every invocation. We now track each worktree's actual `HEAD`, resolve branch references through the shared Git directory, and handle relative paths and packed references. When the current branch is packed, we also watch the nearest existing reference directory so a newly created loose reference becomes visible. This can invalidate a packed-reference worktree when another branch in the same repository changes.
Add a minimal integration test that reproduces astral-sh#21089. Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
third time is a charm, (now that [ost-simple-sts supports issue_comments](https://github.com/open-security-tools/ost-simple-sts/releases/tag/v0.0.5)) authorize automation for shuffling issue-context around on issue comment events
## Summary `uv check` runs `ty check` in a separate process, but previously applied terminal settings only to uv itself. As a result, ty could still emit colored output and draw progress bars despite `--color never`, `--no-progress`, or the `--quiet` used by `ty-pre-commit`; its cursor-control sequences could also erase output from earlier hooks. - Forward uv's resolved color choice to the ty subprocess using `ColorChoice::as_str()`. - Centralize uv's progress policy in `Printer::suppresses_progress()` and disable ty progress whenever uv suppresses it, including quiet mode. - Add a hidden `uv check --show-command` diagnostic option that prints the actual ty invocation with shell-safe argument quoting. - Exercise a real ty binary when covering explicit flags, quiet mode, automatic and forced color, environment-derived terminal settings, and script paths containing spaces. Fixes astral-sh/ty-pre-commit#20.
Empty PEP 508 requirements currently produce an orphaned caret after the error message, including for inline-script dependencies. Skip span formatting when the parser input is empty and update the existing parser and integration snapshots. Fixes astral-sh#21089. Co-authored-by: astral-automations-bot[bot] <305554984+astral-automations-bot[bot]@users.noreply.github.com>
…h#21093) When a regression-test pull request is promoted or merged while an automated fix is being prepared, the original `uv-dev` pull request closes and the fix workflow rejects it as changed. Continue only when the promoted upstream pull request preserves the verified regression branch and commit, while still rejecting unrelated closures or changed heads. Co-authored-by: Zanie Blue <contact@zanie.dev>
Cache cleanup tests manually probed copy-on-write filesystem configuration and could execute on unconfigured local filesystems, while cache tests duplicated the same snapshot size-filter override in several places. Use `TestContext::with_cache_on_cow_fs` to configure or skip filesystem-dependent tests, require configured filesystems to produce actual clones, and share `TestContext::with_cache_size_filters` across cache cleanup, pruning, and sizing tests. --------- Co-authored-by: Zanie Blue <contact@zanie.dev>
…onment indicates corruption (astral-sh#21073) Interpreter metadata can remain stale when a Python launcher loads a different `libpython` without its executable timestamp changing. For a project without an existing environment, the first `uv run` can select the stale cached base interpreter, create `.venv` with incorrect `pyvenv.cfg` metadata, and evaluate `requires-python` against the wrong Python version. On a subsequent `uv run`, the existing environment exposes the inconsistency: its interpreter reports the actual version while `pyvenv.cfg` and the cached base interpreter report the stale version. When both paths refer to the same underlying interpreter, clear only the base interpreter cache entry before continuing normal interpreter discovery. Discovery then queries the executable, saves the actual version, validates `requires-python` against current metadata, and recreates the environment once with the correct `pyvenv.cfg`; later runs reuse it. Existing environments with genuine incompatibilities retain their normal behavior. This change does not detect a stale base interpreter cache before the first environment is created, so the initial `uv run` can still use incorrect metadata. Recovery begins only when a subsequent invocation observes the inconsistent environment. Fixes astral-sh#21066. --------- Co-authored-by: Zanie Blue <contact@zanie.dev>
we copied the reproduce-bug's deterministic titles, but really we should have something of substance (while remaining concise) in the PR title and summary.
…#21101) Generated regression tests can stop at the reported example and miss other demonstrated manifestations of the same bug, including misleading existing coverage. Teach the regression-test agent to inspect related settings and producer/consumer paths, add the smallest worthwhile set of same-root-cause regressions, and correct directly related tests without expanding into speculative or unrelated cases. Co-authored-by: Zanie Blue <contact@zanie.dev>
Cache cleanup tests assemble custom snapshot filters or mutate the test context inside snapshot assertions, and the existing count helper also hides package counts that these tests should retain. Add a focused removed-file-count context builder and apply cache-specific filters when constructing each test context so cleanup output stays stable without weakening package-count assertions. Co-authored-by: Zanie Blue <contact@zanie.dev>
Co-authored-by: astral-automations-bot[bot] <305554984+astral-automations-bot[bot]@users.noreply.github.com> Co-authored-by: Zanie Blue <contact@zanie.dev>
) Derivation-chain regression tests in lock, sync, pip install, and pip compile each rebuild the same temporary source-path filter outside their test context. Apply the existing filter directly during context construction and use context filters in the snapshots, keeping the build-failure diagnostics unchanged while removing repeated setup. --------- Co-authored-by: Zanie Blue <contact@zanie.dev>
…#21114) Several integration tests create a test context and immediately shadow it solely to install static snapshot filters, which obscures setup and scatters context configuration through the test. Attach those filters when constructing the original context while preserving intentional rebinding for dynamic paths, staged assertions, and optional filesystem setup. Co-authored-by: Zanie Blue <contact@zanie.dev>
Cross-origin redirect tests repeatedly build standalone localhost snapshot-filter vectors after creating their test contexts. Register the existing localhost normalization directly on each context and read its filters at the assertion site, preserving redirect and authentication coverage while removing duplicated filter plumbing. Co-authored-by: Zanie Blue <contact@zanie.dev>
…21113) The `uv self version --output-format json` snapshot filtered `commit_hash` inside `short_commit_hash`, so both fields appeared as `[LONGHASH]` and the short hash was not independently asserted. Match complete hash field names and configure the static JSON filters on `TestContext`, preserving distinct `[HASH]` and `[LONGHASH]` placeholders. Co-authored-by: Zanie Blue <contact@zanie.dev>
…sh#21111) The version cache-independence test attempted to make the cache unusable with `UV_CACHE_DIR`, but `TestContext::version()` supplies an explicit `--cache-dir` that takes precedence, so its success snapshots never exercised an invalid cache. Point the effective context cache directory at an existing file and contrast successful version reads with a version write that fails to initialize the cache. Co-authored-by: Zanie Blue <contact@zanie.dev>
…21148) This is another optimization inspired by nab, I'm not using my codex credits for anything else at the moment, so I have it working on whether those optimizations translate to uv, I know y'all need those spare tokens ;o). `Installable::to_resolution` ends by walking the locked graph and evaluating a conflict marker on every dependency edge. Each call re-encodes the activated projects, extras, and groups into a fresh `Vec<ExtraName>`, a `format!` per item, but the set is final before the walk starts. `ActivatedConflictItems` encodes it once. The earlier passes can still activate an extra, so they keep encoding per edge. A lock with no declared conflicts has an empty activated set and nothing to reuse. The star example is `uv sync --frozen --dry-run --all-extras --group alpha` on a project pulling `apache-airflow[all]`. Profiling builds, median of 60 runs, and the sync plan is byte-identical: | extras | activated items | wall | instructions | | ---: | ---: | ---: | ---: | | 40 | 61 | 53 → 42 ms | 428.2M → 264.0M | | 2 | 23 | 44 → 37 ms | 248.4M → 173.4M | <details> <summary>Reproduction</summary> pyproject.toml: ```toml [project] name = "mre" version = "0.1.0" requires-python = ">=3.12,<3.13" dependencies = ["apache-airflow[all]==2.10.5"] [project.optional-dependencies] e00 = ["idna"] e01 = ["idna"] [dependency-groups] alpha = ["idna"] beta = ["idna"] [tool.uv] conflicts = [[{ group = "alpha" }, { group = "beta" }]] [build-system] requires = ["hatchling"] build-backend = "hatchling.build" ``` Commands: ```console $ uv lock --exclude-newer 2026-07-01 $ uv venv $ uv sync --frozen --dry-run --all-extras --group alpha ``` The first row adds `e02` through `e39`. </details>
astral-sh#21146) The `venv_included_in_sdist` snapshot fails in the Gentoo package build. The error line is identical; only the hint is missing: ``` 6 │- Caused by: at byte [OFFSET]: unsafe symbolic-link target "[PYTHON-3.12]": is absolute 7 │- 8 │-hint: The source distribution includes a virtual environment. ... 6 │+ Caused by: at byte [OFFSET]: unsafe symbolic-link target "[PYTHON-3.12]": is absolute ``` `ExtractError::UnsafePath` carries `position`, `context`, `value` and `reason` — the link target, never the entry that declared it. The target was being matched with `is_virtual_environment_python`, which requires the parent directory to be `bin`. That shape is right for an archive entry like `.venv/bin/python`, but a base interpreter is not required to live in `bin`. Gentoo's build lays them out at `<root>/python/3.12/python3`, where the parent is the version, so the predicate returns false and the hint is dropped. Reproduced outside the test suite by pointing a venv's `bin/python` at an interpreter whose parent is not `bin`: ```console $ uv build --preview-features tar-codec error: Failed to build ... Caused by: at byte 1024: unsafe symbolic-link target ".../pydir/python/3.12/python3": is absolute ``` No hint. With the same project and a target at `.../cpython-3.12/bin/python3`, the hint appears. That A/B is the whole bug. This matches the link target on its file name alone and keeps the stricter entry-shaped check for the entry path in `InvalidLink`. Added `venv_included_in_sdist_interpreter_outside_bin`, which builds that layout explicitly. It fails without the change and the existing `venv_included_in_sdist` is unaffected. Full `build` suite: 191 passed. `cargo clippy --all-targets` and `cargo fmt --check` clean. Closes astral-sh#21128 --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-authored-by: Charlie Marsh <charlie.r.marsh@gmail.com>
`uvx git+https://github.com/uPesy/easyeda2kicad.py` fails with "Not a valid package or extra name". `has_python_script_ext` calls `Path::extension()` on the raw argument, so any URL whose last segment ends in `.py` is read as a script path. uv then parses the whole URL as a package name, which is where the message comes from. This skips the script check when the argument carries a scheme `Scheme::parse` recognises. Pairing `split_scheme` with `Scheme::parse` is deliberate and matches the existing use in `verbatim_url.rs`: a bare `split_scheme` would read the `C:` in `C:\script.py` as a scheme and break script detection on Windows. Before: ```console $ uv tool run --offline git+https://github.com/uPesy/easyeda2kicad.py error: Not a valid package or extra name: "git+https://github.com/uPesy/easyeda2kicad.py" ``` After, it reaches the git fetch and stops only on `--offline`: ```console Updating https://github.com/uPesy/easyeda2kicad.py (HEAD) error: ... Remote Git fetches are not allowed because network connectivity is disabled ``` Two tests added beside the existing script tests. They use `--offline`, so they exercise the parse without touching the network. Both fail without the change with the exact error above. The existing `script.py`, missing-`script.py` and `--from script.py` behaviours are unchanged. Closes astral-sh#21141 --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-authored-by: Charlie Marsh <charlie.r.marsh@gmail.com>
…ardless of libc (astral-sh#21137) ## Summary When using musl we would fall back to using the inode, which would cause churn when going between musl backed and glibc backed uv versions on the same host. Using rustix, this remains consistent as long as the underlying OS supports the call for the specific filesystem. This will cause a small bit of churn on musl systems, but that churn will just be in the form of a cache invalidation when updating uv. ## Test Plan Existing coverage for the functionality. Manual tests for checking musl and linux now work the same (but we bypass glibc/musl so it didn't feel worth it to try to come up with an automated regression test here).
Allow `uv workspace metadata --sync --exact` to remove packages outside the selected resolution while preserving the existing inexact behavior of `--sync`. Match `uv run --exact` semantics and ensure exact synchronization also cleans environments for scripts without dependencies. Co-authored-by: Zanie Blue <contact@zanie.dev>
…h#21164) Promotion currently accepts the latest head of a ready `uv-dev` pull request, so commits pushed after it was marked ready can be promoted without another approval. Require the `head_sha` from `ready_for_review` and stop dispatching promotion on `synchronize`. If the source or promoted head no longer matches, leave the source open, return it to draft, explain the mismatch, and assign the human who marked it ready. Co-authored-by: zaniebot <242828183+zaniebot@users.noreply.github.com>
Signed-off-by: William Woodruff <william@yossarian.net>
`pyx` has ended service. disable publication tests targeting pyx.
CI fails to check out the repository in private copies of uv because the workflows explicitly remove `GITHUB_TOKEN` repository access. Public checkouts work without that access, which hides the problem upstream. Grant `contents: read` to the CI entry point and affected reusable workflows, including jobs that override their inherited permissions, so private CI can read its own source without granting repository write access. Co-authored-by: zaniebot <242828183+zaniebot@users.noreply.github.com>
<!-- Thank you for contributing to uv! To help us out with reviewing, please consider the following: - Does this pull request include a summary of the change? (See below.) - Does this pull request include a descriptive title? - Does this pull request include references to any relevant issues? --> ## Summary Adds a 7 day cooldown to all of our PEP 723 style scripts. <!-- What's the purpose of the change? What does it do, and why? --> ## Test Plan NFC. <!-- How was it tested? --> --- <sub>Stack created with <a href="https://github.com/github/gh-stack">GitHub Stacks CLI</a> • <a href="https://gh.io/stacks-feedback">Give Feedback 💬</a></sub>
<!-- Thank you for contributing to uv! To help us out with reviewing, please consider the following: - Does this pull request include a summary of the change? (See below.) - Does this pull request include a descriptive title? - Does this pull request include references to any relevant issues? --> ## Summary Each of the commits in this PR addresses a different group of `uv audit` findings, in different lockfiles. <!-- What's the purpose of the change? What does it do, and why? --> ## Test Plan NFC. <!-- How was it tested? --> --- <sub>Stack created with <a href="https://github.com/github/gh-stack">GitHub Stacks CLI</a> • <a href="https://gh.io/stacks-feedback">Give Feedback 💬</a></sub>
This PR enables PGO for uv releases, starting with Linux x86-64. The release pipeline is modified as follows: - We build instrumented `uv` and `uvx` release binaries inside the existing manylinux container. - We train directly on eleven shared `test/ecosystem` fixtures: cibuildwheel, Cookiecutter, Flask, HTTPX, LLM, the OpenAI Python SDK, Poetry, pytest-cov, Sentry, Zulip, and the 38-project pyx workspace. - Every project exercises cold and warm `uv pip compile` resolution, cold and warm `uv lock` resolution, `uv export`, `uv pip` installation, and `uv sync` installation on every release platform. - Per-project `exclude-dependencies` settings omit packages without compatible release-platform wheels while preserving the rest of each real dependency graph. - Raw profiles are merged online by workload family, then merged with `llvm-profdata` and fed back into the existing `maturin` build. - JupyterLab, Saleor, Semantic Kernel, Transformers, and Warehouse remain separate, held-out evaluation projects. We benchmarked the builds on Linux x86-64 using the expanded corpus for training and five separate held-out ecosystem projects for evaluation. | Held-out project | Dependency resolution | Wheel installation | Project locking | Lockfile export | | ---------------- | --------------------: | -----------------: | ---------------: | --------------: | | JupyterLab | 11.2% faster | 0.0% | 10.1% faster | 10.6% faster | | Semantic Kernel | 9.3% faster | 7.1% faster | 12.5% faster | 8.7% faster | | Transformers | 11.7% faster | 4.1% faster | 9.5% faster | 11.0% faster | | Saleor | — | — | 9.7% faster | 10.0% faster | | Warehouse | — | — | 9.7% faster | 7.6% faster | | Geometric mean | **10.7% faster** | **3.7% faster** | **10.3% faster** | **9.6% faster** | Across all sixteen held-out workloads, wall time decreased by **8.9%** (95% CI: 8.4–9.8%) and CPU time decreased by **8.4%** (95% CI: 7.9–9.0%). The matched Linux executable was **13.3% smaller** (58.91 MB to 51.05 MB), or **12.3% smaller** after gzip compression (22.68 MB to 19.90 MB). (Saleor and Warehouse include source-only dependencies, so they are benchmarked on locking and export rather than wheel-only installation.) Additional platforms are covered in subsequent PRs in the stack: macOS ARM64, Windows x86-64, and Linux ARM64. Ruff and ty follow the same approach; see the stacks here: - astral-sh/ruff#27570 - astral-sh/ty#4213
## Summary This PR enables PGO for uv's macOS ARM64 releases, following the approach outlined in astral-sh#21001.
## Summary This PR enables PGO for uv's Windows x86-64 releases, following the approach outlined in astral-sh#21001. The existing static-CRT configuration is preserved.
## Summary This PR enables PGO for uv's Linux ARM64 releases, following the approach outlined in astral-sh#21001. (To enable PGO, we also move to a native ARM64 runner.)
## Summary The PGO build script was introduced after we added seven-day dependency cooldowns to our PEP 723 scripts, so its inline metadata and lockfile did not include the repository-wide `exclude-newer` setting. As a result, script lockfile validation and locked script type checking both fail on `main`. Declare the same `P7D` cooldown as the other scripts and regenerate the script lockfile so both CI checks accept it. Closes astral-sh#777.
`zizmor-action` uploads findings to GitHub code scanning by default, which is unavailable in private copies without Code Security enabled. Keep code-scanning uploads for `astral-sh/uv` and report findings as workflow annotations elsewhere. Check the repository name instead of GitHub's `fork` flag so independently created private copies are covered too. --------- Co-authored-by: zaniebot <242828183+zaniebot@users.noreply.github.com> Co-authored-by: Zanie Blue <contact@zanie.dev> Co-authored-by: William Woodruff <william@astral.sh>
Enable the HTTPX SOCKS extra in the publish integration script and update its lockfile.
VedantMadane
force-pushed
the
env-options-system-python
branch
from
September 21, 2026 11:15
9007ac5 to
087bd15
Compare
|
We support using AI (i.e., LLMs) while coding. However, AI should not be used to communicate with maintainers. Please read and follow our AI policy when submitting issues and pull requests. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Parse \UV_SYSTEM_PYTHON\ in \EnvironmentOptions\ as part of #14720.