Part 6 of 13 · 2 min read
What the example found
The three real branches' diff sizes, real pytest results, checker flags, and description checks.
How to read the branch-size table: each row is one of three real branches built with git and holding the same underlying bug fix, with different amounts added around it.
| Branch (toy repo) | Files | Changed lines | Where the lines are |
|---|---|---|---|
| Good: fix and tests | 2 | 22 | source 12, tests 10 |
| No test: same fix | 1 | 12 | source 12 |
| Bloated: fix, reformat, rename, print | 3 | 51 | source 13, router reformat 34, tests 4 |
What to notice: the bloated branch's actual source fix is 13 lines, almost identical to the other two branches' 12. So literally all of the extra size (38 of 51 lines) is unrelated reformatting, a rename, and a debug print. None of those were needed by the stated bug fix.
How to read the pytest-run table: each row is a real pytest invocation against one branch or one combination of code and tests.
| Real pytest run | Result |
|---|---|
| main, existing tests | 3 passed |
| main + good branch's tests | 2 failed, 3 passed |
| good, own tests | 5 passed |
| no-test + good branch's tests | 5 passed |
| bloated, own tests | 3 passed |
| bloated + good branch's tests | 1 error (import of the renamed pick) |
What to notice: the bloated branch's own test run shows "3 passed" with no apparent problem. But that is only because it rewrote the tests to match its own rename. Running the good branch's unmodified tests against the bloated code produces an import error. That exposes how the rename silently breaks anyone who imports the old name.
How to read the checker-flags table: each row is one of the three branches; columns are what the diff checker and the commit-message checker flagged.
| Checker | Good | No test | Bloated |
|---|---|---|---|
| Diff flags | none | no_tests | formatting_churn, debug_leftover, undocumented_api (false alarm) |
| Commit flags | none | not_imperative ("Fixed parsing") | subject_period, not_imperative |
What to notice: the checker flags the no-test branch for missing tests and the bloated branch for noise. But it never catches the one thing that actually matters most about the bloated branch, the breaking rename. That only showed up when the good branch's own tests were run against it.
How to read the description table: each row is a fictional pull-request description at increasing levels of completeness.
| Description (fictional) | Flags |
|---|---|
| "fix bug" | no issue link, no why, no testing, too short |
| One sentence with an issue link | no why, no testing, too short |
| Issue link, problem, change and test evidence | none |
What to notice: adding just the issue link to a one-sentence description still leaves three flags standing. A link alone doesn't answer why the change is needed or how it was verified. That is why the checker treats these as three separate, independent requirements rather than one combined "has enough information" check.
{{svg:path|A contribution path and a ladder of contribution types.}} {{svg:lab|Three pull requests built with real git and run with real pytest.}}