python / python/cpython

Assertion errors/stale cache in typeobject.c when tp_version_tag is not cleared.

未关闭
#99,293 7 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

还没有人认领这个 Issue。

interpreter-core type-crash
主要语言
Python
星标
77.2k
派生
36k
PR 合并指标
PR 指标待抓取

描述

The simple method cache in typeobject.c uses the tp_version_tag field and the Py_TPFLAGS_VALID_VERSION_TAG bit in the tp_flags field to track changes on the type that might invalidate the cache. Unfortunately the code currently assumes that any code clearing the tp_flag also clears the tp_version_tag, and checks this assumption with an assert. Unfortunately, many SWIG-generated extension modules violate this assertion.

SWIG prior to 4.1.0 cleared the tp_flags bit without touching the tp_version_tag field (https://github.com/swig/swig/blob/v4.0.2/Lib/python/pyrun.swg#L1230). This was unintentionally fixed in 4.1.0 by using PyType_Modified() instead, but unfortunately there are a great many SWIG-generated extension modules checked in all over the place (e.g. https://github.com/OSGeo/gdal/blob/6bd07b20b3e55c2fc94da611244a615a4fd2991f/swig/python/extensions/osr_wrap.cpp#L2296) so we cannot assume tp_version_tag is changed when the tp_flag bit is cleared.

The end result is an assertion error when assertions are enabled (I would implore everyone to run tests with assertions enabled...), or potentially a stale cache entry when they are not. It's hard to gauge the likelihood of the stale cache entries. In the case of SWIG it's extremely unlikely, as SWIG merely sets the this attribute on the type, which is not likely to be cached or looked up with _PyType_Lookup. For other cases where extension modules clear the tp_flag bit without changing the tp_version_tag it might be more likely.

(I have a PR to fix this already.)

  • PR: gh-99294
  • PR: gh-101736

贡献指南

打开贡献指南

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

调研方向

从 typeobject.c 开始,检查 tp_version_tag 和 Py_TPFLAGS_VALID_VERSION_TAG 周围的缓存失效逻辑,然后查看 PR gh-99294 和 gh-101736,了解已经在进行的工作。完成的标准是 assertion 和缓存过期问题得到解决,并且启用 assertion 的测试通过。

由索引模型根据 Issue 内容生成。

评估

技术栈
c, python
领域
backend
Issue 类型
缺陷
难度
4/5
预计耗时
3-5 天
活跃度
停滞
描述清晰度
描述清楚
新手友好度
20/100

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。