majiayu000 PyPTO Git Commit Workflow
Prerequisites
Check what changed to determine which agents to run:
git diff --name-only
git diff --cached --name-only
Determine testing needs based on changed files:
| File Types Changed | Run Code Review | Run Testing | Run Clang-Tidy |
|---|---|---|---|
C++ (.cpp, .h) |
✅ Yes | ✅ Yes | ✅ Yes |
Python (.py, bindings, tests) |
✅ Yes | ✅ Yes | ❌ Skip |
Build system (.cmake, CMakeLists.txt) |
✅ Yes | ✅ Yes | ✅ Yes |
Docs only (.md, .rst, docs/) |
✅ Yes | ❌ Skip | ❌ Skip |
Config only (.json, .yaml, .toml, .github/) |
✅ Yes | ❌ Skip | ❌ Skip |
| Mixed (code + docs/config) | ✅ Yes | ✅ Yes | If C++ changed |
Launch appropriate agents IN PARALLEL:
code-reviewer- ALWAYS run for all changestesting- ONLY run if code files changedclang-tidy- Runpython tests/lint/clang_tidy.pyif C++ files changed (via Bash agent)
Workflow
- Analyze changed files to determine testing needs
- Launch in parallel (single message with multiple Task tool calls):
- code-reviewer agent (always)
- testing agent (if code changed)
- clang-tidy via Bash agent:
python tests/lint/clang_tidy.py(if C++ changed)
- Wait for all agents to complete
- Address any issues found
- Stage changes
- Generate commit message
- Commit and verify
Stage Changes
Related changes together:
git add path/to/file1.cpp path/to/file2.h
git diff --staged # Review
Cross-layer pattern (C++ + Python + Type stubs + Tests):
git add include/pypto/ir/expr.h python/bindings/ir_binding.cpp \
python/pypto/pypto_core/__init__.pyi tests/ut/ir/test_expr.py
Never stage: Build artifacts (build/, *.o), temp files, IDE configs
Commit Message Format
Structure: type(scope): description (≤72 chars)
Types: feat, fix, refactor, test, docs, style, chore, perf Scope: Module/component (ir, printer, builder) Description: Present tense, action verb, no period
Good examples:
feat(ir): Add unique identifier field to MemRef
fix(printer): Update printer to use yield_ instead of yield
refactor(builder): Simplify tensor construction logic
test(ir): Add edge case coverage for structural comparison
Bad examples (avoid):
❌ feat(ir): Added feature. # Past tense, has period
❌ Fix bug # Missing type prefix
❌ WIP # Not descriptive
Commit
# Short message
git commit -m "feat(ir): Add tensor rank validation"
# Detailed message (in editor)
git commit
In editor:
feat(ir): Add tensor rank validation
Validates tensor rank is positive before setting shape.
Raises ValueError for invalid ranks.
Updates tests with edge case coverage.
Co-Author Policy
❌ NEVER add AI assistants: No Claude, ChatGPT, Cursor AI, etc.
✅ Only credit human contributors: Co-authored-by: Name <email>
Why? AI tools are not collaborators. Commits reflect human authorship.
Post-Commit Verification
git show HEAD # View commit
git log -1 # Check message
git show HEAD --name-only # Verify files
Fix issues (only if not pushed):
git commit --amend -m "Corrected message" # Fix message
git add file && git commit --amend --no-edit # Add forgotten file
⚠️ Only amend unpushed commits!
Checklist
- Changed files analyzed (code vs docs/config only)
- Code review completed
- Tests passed (if code changed) or skipped (if docs/config only)
- Clang-tidy passed (if C++ changed) or skipped (if no C++)
- Only relevant files staged
- No build artifacts
- Message format:
type(scope): description(≤72 chars, present tense, no period) - No AI co-authors
Remember
A good commit is thoroughly reviewed, groups related changes, has clear "why" message, and attributes only human authors.