| name | fixing-diff-cover-failures |
| description | Use when make test-diff-coverage fails, CI reports diff coverage below 100%, or a GitHub PR is blocked on uncovered changed lines. |
Fixing diff-cover failures
Overview
GitHub merge requires 100% test coverage on the PR diff (branch vs base). Overall project coverage (e.g. 85%) is separate and does not unblock merge.
Loop until green
git fetch origin
make test-diff-coverage
- Read uncovered lines printed by
diff-cover (same signal as CI).
- Add or extend tests so every changed line executes in tests (branches, errors, edge cases).
- Re-run
make test (full suite), then make test-diff-coverage.
- Repeat until both pass.
Use python3 -m pytest --cov=pulp_tool --cov-report=term-missing only to debug specific files; make test-diff-coverage is the authoritative gate.
Red flags
- Declaring PR-ready after
make test alone (without diff-cover)
- Relying on "nearby" coverage from unrelated tests
- Stopping after one green run before a later fix broke coverage
- Using
# pragma: no cover to avoid writing tests (only if truly unreachable and justified)
Common rationalizations
| Excuse | Reality |
|---|
| "Overall coverage is high" | Merge gate is 100% on the diff only |
| "That line is hard to test" | Add a focused unit test or refactor for testability |
| "CI might pass anyway" | diff-cover locally matches CI; fix before push |
References
- Merge policy: CONTRIBUTING.md
- Full check script:
git fetch origin && ./scripts/check-all.sh
