Enable LTO and PGO for native Linux AArch64 builds on Python 3.10-3.12
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 25/100
- Issue 类型
- 功能
- 描述清晰度
- 描述清楚
- 活跃度
- 停滞
- 领域
- build-system, ci-cd, performance
调研方向
阅读 3.10、3.11 和 3.12 维护分支中现有的优化块以及 Linux AArch64 配方设置。验证原生 GitHub Actions 作业是否运行 PGO 训练和优化重建；完成的标准是移除 AArch64 覆盖、递增配方构建编号、保持 PPC64LE 覆盖不变，并评估 #880 的兼容性风险。
由索引模型根据 Issue 内容生成。
描述
Summary
Linux AArch64 release builds on the Python 3.10, 3.11, and 3.12
maintenance branches still set _OPTIMIZED=no, disabling the existing
LTO and PGO build path.
That override was introduced in 2019 because optimized builds took too
long in emulated CI. These branches now run their Linux AArch64 jobs on
native ubuntu-24.04-arm runners with AArch64 container images.
Current state
| Branch | Linux AArch64 release result |
|---|---|
| 3.10 | LTO disabled, PGO disabled |
| 3.11 | LTO disabled, PGO disabled |
| 3.12 | LTO disabled, PGO disabled |
The optimized block already present in each branch enables LTO and, when
the build is not cross-compiling, adds --enable-optimizations, runs
profile-opt, and uses PROFILE_TASK="-m test --pgo".
Proposal
Remove only the linux-aarch64 _OPTIMIZED=no override from the 3.10,
3.11, and 3.12 maintenance branches and bump each recipe build number.
Keep the PPC64LE override unchanged because that target is still mapped
to an x86_64 build platform.
This reuses each branch's existing optimization path:
- 3.10 and 3.11: existing
--with-ltoplus PGO - 3.12: existing
--with-lto=fullplus PGO
Evidence
- Historical disable commit:
https://github.com/conda-forge/python-feedstock/commit/b2374b9934dde07caa374cda2d4ffb7c518da07c - Previous optimization re-enable PR for the then-current branch:
https://github.com/conda-forge/python-feedstock/pull/739 - Native ARM runner transition:
https://github.com/conda-forge/python-feedstock/commit/babe1dd57b80d10394868f16e84918779abde9fe
Validation
I will open one draft PR per maintenance branch. Native Linux AArch64
GitHub Actions jobs will be used to validate the PGO training and optimized
rebuild. The PRs will be linked here after creation.
Known compatibility concern
Native AArch64 Python rebuilds are currently associated with the downstream
cross-AArch64 C++ extension linking regression tracked in #880. This proposal
does not change the affected sysconfig/linker behavior; the optimization PRs
will remain drafts so that this rebuild risk can be assessed explicitly.
Pull requests
- 主要语言
- Shell
- 星标
- 47
- 派生
- 112
- 平均合并
- 1 天 2 小时
- 30 天内合并 PR
- 15
环境准备
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。
conda-forge/python-feedstock 的其他 Issue
-
难度 2/5 1-3 小时 新手友好度 88/100
conda-forge/python-feedstock#934 · 1 条评论 ·
维护者通常 1 天内回复
-
难度 3/5 1-2 天 新手友好度 48/100
conda-forge/python-feedstock#929 ·
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 48/100
conda-forge/python-feedstock#880 ·
维护者通常 1 天内回复
-
难度 3/5 1-2 天 新手友好度 58/100
conda-forge/python-feedstock#879 ·
维护者通常 1 天内回复
-
bug
难度 3/5 1-2 天 新手友好度 38/100
conda-forge/python-feedstock#867 · 4 条评论 ·
维护者通常 1 天内回复
查看 conda-forge/python-feedstock 的全部 Issue
相似的 Issue
-
enhancement
难度 2/5 1-3 小时 新手友好度 68/100
alunduil/alunduil-infrastructure#629 ·
维护者通常 1 天内回复
-
难度 1/5 1 小时以内 新手友好度 92/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
duckdb/duckdb-skills#19 ·
-
难度 2/5 1-3 小时 新手友好度 78/100
-
难度 2/5 1-3 小时 新手友好度 78/100
YosysHQ/oss-cad-suite-build#216 ·