Uses a higher granularity than seconds for `.pyc` file invalidation
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
Bug report
Bug description:
reproduction:
#!/bin/sh
mkdir -p src
rm -fr src/__pycache__
rm -fr __pycache__
echo 'from src.test import f' > main.py
echo 'print(f())' >> main.py
echo 'def f():' > src/test.py
echo ' return "Hello!"' >> src/test.py
# prints "Hello!"
python3 main.py
echo 'def f():' > src/test.py
echo ' return "Goodbye!"' >> src/test.py
# also prints "Hello!", but note the file does not say that anymore
python3 main.py
(this is probably sensitive to how fast your computer is)
this seems to stem from (i guess) that the timestamps are 32-bit, and write the current second (unix time, as 32-bit). and then, when the file is edited and the python script runs again, it gets a cache hit (same timestamp as file 32-bit mtime) and yields the old content.
is this known/intended? is there a mitigation that isn't just "delete the pycache"? i tried looking if anyone else ran into this and could not find anything.
i ran into this as part of a workflow that first edited a python script via shell (edits mtime), then invoked python on it (updates pyc to that mtime), then that python invocation updates the same script a second time and those changes were weirdly lost and confused me for a while. after having to delete the pycache a few times, i looked into it and ended up at (seemingly) this issue.
also (not relevant), having not looked at the code, what happens in 2038 when 64-bit file timestamps go over the 32-bit value? will the pyc just wrap around or is there some other planned handling?
maybe this format should be updated to 64-bit milliseconds (this seems.. difficult, maybe a new magic?) (and technically milliseconds can also race, though python is usually not fast enough to hit that), or perhaps the pycache should be invalidated when the mtime matches (but this would loop a bunch of times when the script is ran multiple times on the same second until current time is 1 second later than the filetime)
CPython versions tested on:
3.12
Operating systems tested on:
Linux
Linked PRs
- gh-121502
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
在 CPython 3.12 上运行 shell 复现,并检查 import cache 涉及的 .pyc 失效处理和时间戳处理。将行为与链接的 PR gh-121502 进行比较;应以重复的同秒编辑不再执行过时源代码且相关边界情况已覆盖为完成标准。
由索引模型根据 Issue 内容生成。
评估
- 技术栈
- python
- 领域
- tooling
- 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
- 需要澄清
- 新手友好度
- 25/100