[PGO] Provide Pre-Generated Profiles
还没有人认领这个 Issue。
- 主要语言
- Python
- 星标
- 77.2k
- 派生
- 35.9k
- PR 合并指标
- PR 指标待抓取
描述
Feature or enhancement
PGO (Profile-Guided Optimization) offers significant performance benefits for Python. However, the current approach has a drawback: it makes the builds non-reproducible.
This issue could be resolved if upstream provided pre-generated profiles (perhaps via CI pipelines). These profiles could then be included in the distributed tarball, similar to how Chromium handles it. By doing so, distributions wouldn't need to:
- Compile the source code twice.
- Perform profiling themselves.
In addition to streamlining the process for maintainers, this would also help maintain package reproducibility—a key consideration for many distributions.
I am not sure, about the integration with GCC, since in my testing the compilation failed with a preset profile (but I didnt dig deeper into this yet.
Has this already been discussed elsewhere?
This is a minor feature, which does not need previous discussion elsewhere
Links to previous discussion of this feature:
No response
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。
调研方向
尚未确定任何文件、测试或入口点。首先调查 CPython 的 PGO 构建和 CI 流水线,尤其是 GCC 如何使用预设 profile。完成的标准是在 CI 中生成可复现的 profile,并将其包含在分发的 tarball 中,同时无需下游进行 profiling 或再次编译。
由索引模型根据 Issue 内容生成。
评估
- 技术栈
- python
- 领域
- build-system, performance
- Issue 类型
- 功能
- 难度
- 5/5
- 预计耗时
- 一周以上
- 活跃度
- 停滞
- 描述清晰度
- 需要澄清
- 新手友好度
- 30/100