pip.parse ignores uv.lock dependency-edge platform markers
维护者通常 1 天内回复
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 20/100
- Issue 类型
- 缺陷
- 描述清晰度
- 描述清楚
- 活跃度
- 停滞
- 技术栈
- python
- 领域
- backend, build-system
调研方向
Read _parse_uv_lock_json in python/private/pypi/parse_requirements.bzl and inspect PR #4286, which already proposes a fix and includes regression coverage. The reproduction in this issue shows the failure for uvloop across Linux x86_64 and macOS arm64; done means only the Linux artifact is selected and the related parser and uv-lock conversion tests pass.
由索引模型根据 Issue 内容生成。
描述
Affected Rule
pip.parse(uv_lock = ...), specifically _parse_uv_lock_json in python/private/pypi/parse_requirements.bzl.
Is this a regression?
Unknown. Confirmed against upstream main at 9f3ab0630537ce54356d78a28f4618d926ec31ce.
Description
When importing a universal uv.lock, packages with platform-specific dependency edges are assigned to platforms on which they are unreachable. The parser checks package-level resolution-markers but ignores the marker on dependency edges. uv does not necessarily write package-level markers for these dependencies.
For a Linux-only uvloop dependency, the parser selects a Linux wheel for Linux and its source archive for macOS. If a package has a compatible macOS wheel, that artifact can also be selected even though the package should be absent on macOS.
Minimal reproduction
Create this pyproject.toml:
[project]
name = "rules-python-platform-marker-repro"
version = "0.1.0"
requires-python = ">=3.9"
dependencies = ["uvloop==0.22.1; sys_platform == 'linux'"]
[tool.uv]
environments = [
"sys_platform == 'linux' and platform_machine == 'x86_64'",
"sys_platform == 'darwin' and platform_machine == 'arm64'",
]
Generate uv.lock with uv lock. The lock generated by uv 0.11.21 contains:
[[package]]
name = "rules-python-platform-marker-repro"
version = "0.1.0"
source = { virtual = "." }
dependencies = [
{ name = "uvloop", marker = "platform_machine == 'x86_64' and sys_platform == 'linux'" },
]
The uvloop package entry has Linux wheels and an sdist, but no package-level resolution-markers. The lock's top-level resolution markers are separate from package-level markers.
Importing this lock with pip.parse for target_platforms = ["linux_x86_64", "osx_aarch64"] should resolve uvloop only for Linux. To confirm the failure independently of package installation, I invoked the unmodified public parse_requirements function through an isolated Bazel module against a JSON serialization of this actual lock, with Python 3.9 Linux x86_64 and macOS arm64 platform definitions. The parser returned:
uvloop-0.22.1-cp39-cp39-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl: ["linux_x86_64"]
uvloop-0.22.1.tar.gz: ["osx_aarch64"]
Expected: Only the Linux wheel; no macOS artifact.
Actual: The parser additionally selects uvloop's sdist for macOS. No macOS package build or installation was attempted in this reproduction; the incorrect artifact selection itself was observed directly.
Environment
- Host: macOS arm64.
- Target marker environments: Linux x86_64 and macOS arm64, Python 3.9.6.
- Bazel: 9.1.1.
- uv: 0.11.21.
- rules_python: upstream commit
9f3ab0630537ce54356d78a28f4618d926ec31ce.
Proposed fix
PR #4286 walks the dependency graph from virtual/editable workspace roots per target platform, applying edge markers and retaining packages only on platforms where they are reachable. It also preserves package resolution markers and version/source identities and follows requested dependency extras.
The same real-lock reproduction selects only the Linux artifact with that patch. The PR includes regression coverage for this case and related graph/extras cases; all 66 selected parser and uv-lock conversion tests pass locally.
- 主要语言
- Starlark
- 星标
- 688
- 派生
- 728
- 平均合并
- 1 天 1 小时
- 30 天内合并 PR
- 94
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
bazel-contrib/rules_python 的其他 Issue
-
难度 2/5 1-3 小时 新手友好度 68/100
bazel-contrib/rules_python#4251 · 1 个 reaction ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 78/100
bazel-contrib/rules_python#4164 · 1 条评论 ·
维护者通常 1 天内回复
-
go:embed stdlib_list.txt file in gazelle/python/std_modules.go is missing可能已有人在做 @udaya2899 于 22 天前认领。 未关闭
难度 2/5 1-3 小时 新手友好度 72/100
bazel-contrib/rules_python#3821 ·
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 38/100
bazel-contrib/rules_python#4278 ·
维护者通常 1 天内回复
-
难度 3/5 1-2 天 新手友好度 45/100
bazel-contrib/rules_python#4269 ·
维护者通常 1 天内回复
查看 bazel-contrib/rules_python 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 77/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 67/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 76/100
-
bug
难度 2/5 1-3 小时 新手友好度 87/100
huggingface/smolagents#3045 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 70/100
router-for-me/CLIProxyAPI#6851 ·
维护者通常 1 天内回复