ci: document and test path-based unit test patterns in pr bot

[amd-chrissosa]

Summary

The test_file_patterns config already supports full-path glob matching (patterns containing / are matched against the full file path via _is_test_file), but this was underdocumented and had no unit test coverage.

This matters because some repos (notably rocm-systems) have test directories whose files use no test_* naming convention (e.g. projects/hip-tests/catch/unit/memory/atomicAdd.cc).

• policy.yml: expand the test_file_patterns comment to clearly document both matching modes (basename vs full-path) with concrete examples for how downstream repos configure path-based patterns for their test directories
• test_policy_check_ut.py: add two new tests covering path-based patterns (one pass, one fail), and fix the make_policy() fixture which referenced Policy fields that no longer exist (max_files_changed, max_total_changes, max_single_file_changes), causing 20 pre-existing test failures on main

Closes #6654

Test plan

• python -m pytest skills/therock_pr_bot/test_policy_check_ut.py — 32 pass (up from 3), 2 pre-existing failures unrelated to this change
• test_path_based_pattern_satisfies_requirement: file under projects/hip-tests/** satisfies the unit test requirement alongside a CLR source change
• test_path_based_pattern_code_only_still_fails: source-only change with no matching test path still fails correctly

Follow-up

A separate PR to rocm-systems (ROCm/rocm-systems#8782) syncs policy_check.py and adds projects/hip-tests/** to its test_file_patterns.

🤖 Generated with Claude Code

[therock-pr-bot]

✅ All Checks Passed — Ready for Review

Check	Status	Details
🌿 Branch Name	✅ Pass	—
📝 PR Title/Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	✅ Pass	PR does not contain code files — Unit Test auto-passed
🔎 pre-commit	✅ Pass	—
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—
🤖 therock-pr-bot	✅ Pass	—

🎉 All checks passed! This PR is ready for review.

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

[therock-pr-bot]

🎉 All checks passed! This PR is ready for review.

[ScottTodd]

The policy.yml in this repository currently only includes

    test_file_patterns:
      - 'test_*'
      - 'testing_*'
      - '*_test.*'
      - '*_tests.*'
      - 'Test*'
      # Path-based: any file under a 'test/gtest/' directory is a unit test
      # (e.g. projects/miopen/test/gtest/unit_conv_solver_ConvWinoRageRxS.cpp).
      - '**/test/gtest/**'

so projects/hip-tests/catch/unit/memory/atomicAdd.cpp would not match without also adding this projects/hip-tests/** pattern. We can apply that pattern in rocm-systems, but what about always adding a simpler regex pattern like *test*? There would be false positives if we don't include a separator character like _, -, /, .

---

I ran a few audits and found only five source files that would match *test* as part of the file name without being an actual test file:

• generateStats.cpp (genera teSt ats.cpp)
• generateStats.hpp (genera teSt ats.hpp)
• filewritestream.h (filewri test ream.h
• Two copies of ComputeStoreVgprs.py (Compu teSt oreVgprs.py)

The larger concern with using an overly broad filter pattern like *test* for the entire file path is that it would match any file under a test directory, including "non-source" files like like .md, .txt

Here's my full conversation with Codex running some analysis: https://gist.github.com/ScottTodd/49e4990c3bb05092f6d51690713acd77

---

Based on that analysis, I think standardizing test file names and encoding into enough policy patterns is probably better than a broad case-insensitive *test* regex. It shouldn't be too difficult to run a more focused audit on each subproject, find the style used, and encode that style into the policy while watching for false positives / false negatives.

[amd-chrissosa]

Good analysis — the \ false positive rate is low but the broader path concern is real. Agree that the right long-term approach is a per-subproject audit rather than a broad catch-all, with each team encoding their actual conventions into the policy (as we are doing for rocm-systems with the path pattern for hip-tests). Will track that as follow-up work. Thanks for running the audit.