Jon Simons's patch adds ten lines to Documentation/SubmittingPatches documenting the preference against the "add a failing test and fix it in a subsequent commit" pattern. The commit message says such a sequence is fine for local development but that reviewers prefer a single commit that introduces the test with its code change, to cut back-and-forth on the list. A revised version links two examples of such feedback from August 2025.
Junio C Hamano said this matches what contributors have been told. He asked whether the first two and a half lines of the proposed text are needed, since not writing tests, relying on manual testing, or sending test_expect_failure() separately and early are not things the project recommends either.
Junio also explained the rationale. A large test added with test_expect_failure() in one patch and fixed in another makes the second patch show only a switch to test_expect_success(), not the body of the test, which makes review cumbersome. His message continues with a point about trivial, small tests, but the excerpt cuts off there.
Kristoffer Haugsbakk said he found no existing mention of the point in the document and called it useful. He asked that the commit message link examples, said the text is well-written, and thought its location under [[tests]], after [[separate-commits]], suits it as a counterweight to making too many separate commits.
Jon agreed with Junio's suggestion and said he would apply it in a v2.