Parameter loader turns __pycache__ directories into parameter nodes

未关闭
  • 哪里不对？
#638 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 2 天内回复

@anth-volk 已经在做这个了。

开始于 2026年10月29日。

  • #637 来自 @anth-volk —— 未关闭

评估

难度
2/5
预计耗时
1-3 小时
新手友好度
35/100
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

调研方向

Read policyengine_core/parameters/parameter_node.py, focusing on the subdirectory branch and the existing EXCLUDED_PARAMETER_CHILD_NAMES checks. Reproduce the issue with a parameter directory containing __pycache__ or a hidden subdirectory, then check the parameter tree excludes both. A linked pull request (#637) is already open.

由索引模型根据 Issue 内容生成。

描述

bug

Problem

ParameterNode(directory_path=...) makes every subdirectory of a parameter directory a child node. EXCLUDED_PARAMETER_CHILD_NAMES in policyengine_core/parameters/parameter_node.py already lists __pycache__, but only the YAML-file branch and the inline-data branch check it. The subdirectory branch does not, so a __pycache__ directory becomes an empty <directory>.__pycache__ node.

Python writes __pycache__ into a parameter directory whenever that directory is also a Python package that gets imported. The loaded parameter tree therefore depends on whether bytecode has been written to disk.

Effect

policyengine-us 2.2.1 has six parameter directories that hold only an __init__.py and CSV files, and US variable modules import them:

  • gov.hhs.medicaid.geography
  • gov.hud.fmr
  • gov.hud.income_limits
  • gov.hud.payment_standards
  • gov.states.fl.doe.sr.rates.reimbursement
  • gov.states.tx.twc.ccs.payment.rates_expanded_age_groups

Without bytecode caches, each of these nodes has no children. With caches, each has one empty __pycache__ child. policyengine.py skips parameter nodes with no children, so its US parameter-node count changes by 6 depending on whether the caches exist: 28,559 without them and 28,565 with them, measured with policyengine-us 2.2.1 and policyengine-core 3.33.3.

The Public API's v2 catalog takes its parameter nodes from policyengine.py. Its exact-count integration test (tests/integration/test_v2_catalog_installed.py) reports 29,361 parameter nodes on macOS with PYTHONDONTWRITEBYTECODE=1 and 29,367 elsewhere. The count follows the files on disk, not the environment variable: once caches have been written, for example by pip install's bytecode compilation or an earlier run, PYTHONDONTWRITEBYTECODE=1 still gives the higher count.

Hidden directories have the same problem. A .ipynb_checkpoints directory inside a parameter directory loads copies of the YAML files next to it as extra parameters, under names that are not valid dotted paths.

Expected

The loaded parameter tree should not depend on bytecode caches or on hidden tool directories. Subdirectories named __pycache__, and hidden subdirectories, should not become parameter nodes.

主要语言
Python
星标
22
派生
30
平均合并
4 天 20 小时
30 天内合并 PR
26

环境准备

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。

PolicyEngine/policyengine-core 的其他 Issue

查看 PolicyEngine/policyengine-core 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。