$ ~/registry/skill/facebook-llms-skills-pysa-integration-tests

SKILL

pysa-integration-tests

用于运行、调试与更新 Pysa 污点分析端到端集成测试及期望结果

来源

GitHub

更新于

2026-06-07

// 安全评估低风险

仅提示词，不执行代码
开源可审计

正在进行安全审计…

凭证密钥
网络外发
代码执行
数据访问
来源供应链

// 安装

复制安装指令，让 AI 自动完成配置 · 推荐新手

请帮我安装 askskill 上的 "pysa-integration-tests" 技能：
1. 下载 https://raw.githubusercontent.com/facebook/pyre-check/main/.llms/skills/pysa-integration-tests/SKILL.md
2. 保存为 ~/.claude/skills/pysa-integration-tests/SKILL.md
3. 装好后重载技能，告诉我可以用了

// 下载

下载 SKILL.md机读安装清单 ↗

// 用法示例

更新失败测试的期望输出

输入

请帮我排查 Pysa 集成测试失败原因，并更新 `source/interprocedural_analyses/taint/test/integration` 下对应测试的期望输出文件，说明改动前后差异。

预期产出

定位失败原因、更新相关期望文件，并附上差异说明与后续验证建议。

新增 taint 集成测试

输入

我新增了一个污点分析场景，请为 Pysa 在 `source/interprocedural_analyses/taint/test/integration` 下创建完整集成测试，包括测试输入与所需的 `.models`、`.cg`、`.hofcg`、`.overrides` 或期望结果文件。

预期产出

一套可运行的新集成测试文件，以及文件用途和断言目标说明。

调试模型与调用图文件

输入

请检查这个 Pysa 集成测试中的 `.models`、`.cg`、`.hofcg` 和 `.overrides` 文件是否一致，找出导致 taint 分析结果异常的配置或建模问题，并给出修复建议。

预期产出

对相关测试文件的一致性分析、问题定位，以及可执行的修复方案。

// 文档

Pysa Integration Tests

Overview

End-to-end integration tests for the Pysa taint analysis engine. Each test is a .py file under source/interprocedural_analyses/taint/test/integration/. Tests run the full taint analysis pipeline and compare output against expected files.

Running Tests

All commands must be run from the source/ directory.

cd source

# Run ALL tests (parallelized with 16 shards)
OUNIT_SHARDS=16 dune exec interprocedural_analyses/taint/test/integrationTest.exe

# Run a SINGLE test (e.g., format.py)
PYSA_INTEGRATION_TEST=format.py dune exec interprocedural_analyses/taint/test/integrationTest.exe

With Pyrefly Frontend (Meta Internal Only)

See the "Pyrefly Integration" section in AGENTS.md for how to build the Pyrefly binary.

To run integration tests with Pyrefly:

PYREFLY_BINARY=<path-to-binary> PYSA_INTEGRATION_TEST=format.py dune exec interprocedural_analyses/taint/test/integrationTest.exe

Test File Structure

Each test <name>.py may have these companion files:

File	Required	Purpose
`<name>.py`	Yes	Python source code to analyze
`<name>.py.pysa`	No	Pysa model file: declares sources, sinks, TITO
`<name>.py.config`	No	Taint configuration: rules, sources, sinks, options
`<name>.py.models`	Yes	Expected output: taint models and issues (JSON)
`<name>.py.cg`	Yes	Expected output: call graph
`<name>.py.hofcg`	Yes	Expected output: higher-order call graph
`<name>.py.overrides`	Yes	Expected output: override graph
`<name>.py.pyrefly.models`	No	Expected Pyrefly output: models
`<name>.py.pyrefly.cg`	No	Expected Pyrefly output: call graph
`<name>.py.pyrefly.hofcg`	No	Expected Pyrefly output: higher-order call graph
`<name>.py.pyrefly.overrides`	No	Expected Pyrefly output: overrides

Default models: When no .pysa and no .config file is present, the test runner automatically provides default test models (_test_sink, _test_source, etc.). When either file is present, the test must be self-contained.

Debugging Test Failures

When a test fails because expected output doesn't match:

The test runner prints a diff (may be truncated for large diffs)
It creates .actual files for each mismatched output (e.g., format.py.models.actual)
Use diff to compare expected vs actual:

diff source/interprocedural_analyses/taint/test/integration/format.py.models \
     source/interprocedural_analyses/taint/test/integration/format.py.models.actual

If the test fails with type errors or analysis errors (not output mismatches), the issue is in the Python source or model definitions.

Updating Expected Files

If the output changes are expected (e.g., you intentionally changed the analysis):

# Automatically moves all .actual files to their expected counterparts
facebook/scripts/in_path/pysa-update-expected

This replaces each <name>.py.<ext> with the corresponding <name>.py.<ext>.actual.

Always review the diff before updating — run sl diff after updating to verify changes are intentional.

Creating a New Test

Create <name>.py under source/interprocedural_analyses/taint/test/integration/
Optionally create <name>.py.pysa (model file) and/or <name>.py.config (taint config)
Run the test — the runner auto-creates missing expected output files on the first run:

PYSA_INTEGRATION_TEST=<name>.py dune exec interprocedural_analyses/taint/test/integrationTest.exe

Review the generated .models, .cg, .hofcg, .overrides files
Commit all files together

Common Mistakes

Wrong env var name: It's PYSA_INTEGRATION_TEST, not PYSA_TEST or similar
Forgetting the .py extension: Use PYSA_INTEGRATION_TEST=format.py, not format
Running from wrong directory: dune exec must be run from source/