Focus Testing on High-Churn Files | Generated by AI

Home 2026.05

Here’s the data and the strategy:

Churn data (top 15 files by commit count):

62  ww/main.py                     ← CLI entry point, by far #1
12  ww/note/create_normal_log.py
10  ww/note/create_note_utils.py
10  ww/note/create_log.py
 9  ww/image/screenshot.py
 8  ww/sync/remote.py
 8  ww/llm/openrouter_client.py
 8  ww/github/gitmessageai.py
 6  ww/note/note_workflow.py
 6  ww/note/create_note.py
 6  ww/note/create_note_from_clipboard.py
 6  tests/test_main.py
 5  ww/note/screenshot_log.py
 5  ww/git/git_update.py
 4  ww/search/search_*.py (4 files)
 4  ww/git/git_squash.py, git_show_command.py, etc.

The distribution is extreme:

Why this strategy works:

  1. Churn = bug injection surface. Files that change frequently are where regressions happen. A file touched 62 times (main.py) has 62x more opportunities for bugs than a file touched once.

  2. The 80/20 is even more extreme here. The top ~45 files (6%) account for the vast majority of commits. Testing those covers where 80%+ of future changes will land.

  3. Current coverage is backwards. You have 60-100% on note/git/image/utils — but some high-churn files like main.py (62 commits), create_normal_log.py (12), screenshot.py (9), remote.py (8), openrouter_client.py (8) may have zero or thin coverage.

  4. The 0%-coverage untested modules (agent/, audio/, cloud/, etc.) are mostly low-churn. They were built once and left alone. Testing them has low ROI — they rarely change, so bugs there are rare.

The playbook:

Step 1: git log --format= --name-only ww/ | sort | uniq -c | sort -rn > /tmp/churn.txt
Step 2: For each file in the top 30% (~20-25 files):
   - Does a test file exist? (tests/*/<module>/test_<name>.py)
   - If yes, what's the coverage? (pytest --cov=<module>)
   - If no, that's a gap — write a test
Step 3: Priority order = churn count × (1 - existing_coverage)

The formula: priority = commits × (1 - coverage%) — this gives you a ranked list where a file with 62 commits and 0% coverage (main.py) scores highest, and a file with 1 commit and 80% coverage scores near zero.

Bottom line: Yes, this is the right way to think about it. Testing high-churn files first is the single highest-leverage move for improving real-world test quality. The current 12% overall number is misleading — if you cover the top 30 files by churn well, you’re protecting where it actually matters.

Generated by AI. Curating and sharing still takes effort. If you find it useful, feel free to donate. WeChat: @lzwjavaWeChat QR · X: @lzwjava · Say hi 👋

Back Donate