frankenbit/release-toolkit

Watch

Fork 0

bug(prep): the [Unreleased] drop refusal prints a CRLF file's \r, so a terminal overwrites the line it names #1507

New issue

Closed

opened 2026-10-02 12:28:50 +02:00 by claude · 1 comment

claude commented

2026-10-02 12:28:50 +02:00

Member

rt prep refuses [Unreleased] text it would drop (prep.UnreleasedDropError, from #1486), naming each line. The refusal prints each UncarriedLine.Text verbatim. On a CRLF CHANGELOG.md, every record keeps its trailing \r, and that \r reaches the terminal.

Measured

This was a Go probe on main @ 2060ed1, 2026-10-02. It made the same calls rt prep makes: parser.UnreleasedContent(clBytes) (cmd/rt/prep.go:195), changelog.UnreleasedStartLine(clBytes) (:298), then prep.Compose. The CHANGELOG used CRLF line endings, with this [Unreleased]:

Prose.

### Notes

- n

Compose returns *prep.UnreleasedDropError. Detection is correct, with the same lines as on LF input. The refusal's second line is:

"  CHANGELOG.md:5: Prose.\r    <- before the first ### heading"

In a terminal, the \r returns the cursor to column 0. The reason after it is then drawn over the start of the line. Here the reason is the longer part, so CHANGELOG.md:5: Prose. is no longer visible.

Why it matters

The refusal exists to point the operator at the exact lines to fix (#1486). #1485 made this repo check out LF, but an adopter's CHANGELOG.md can still be CRLF. There, the refusal shows reasons without the lines they belong to.

Acceptance criteria

On a CRLF CHANGELOG, rt prep's refusal contains no raw \r, and it names the same file lines as on LF input. — DONE @ #1508 (de49186, reviewed as de49186). TestCompose_refusalOnACRLFChangelog reads the CHANGELOG with the two calls rt prep makes and asserts that the CRLF refusal is byte-equal to the LF one, naming lines 5, 7 and 9.
UncarriedLine.Text stays verbatim; only the rendering strips the \r. — DONE @ #1508 (de49186, reviewed as de49186). The new UncarriedLine.Render strips it at print time, and both rt prep and the fold-back refusal use it. The test checks that each CRLF Text keeps its \r.
LF output stays byte-identical. Its tail is already pinned in internal/prep/unreleased_drop_test.go. — DONE @ #1508 (de49186, reviewed as de49186). The new test pins the whole LF refusal and passes against 2060ed1's code. The literal tail pin is unchanged.
Tests are red on main and mutation-verified (AGENTS.md section 8, Axis B). — DONE @ #1508 (de49186, reviewed as de49186). The test is red against 2060ed1's code at the raw-\r and CRLF-vs-LF assertions. Mutants: 6/6 killed at 5682563. At de49186, the red-on-main check, M1, M5, M7 and an added-kind probe were re-run, and review 7169 reproduced them.

Notes

#1493 solved the same problem in changelog.FoldBackDropError.Describe, with an unexported unCR in internal/changelog/foldback.go. One exported rendering helper used by both refusals would keep them from drifting. #1493 did the same for the accepted-headings note and the remedy (changelog.AcceptedHeadingsNote, changelog.UncarriedRemedy).
Raised in the review of #1493 (review 7158 finding 2; re-review 7160).

`rt prep` refuses `[Unreleased]` text it would drop (`prep.UnreleasedDropError`, from #1486), naming each line. The refusal prints each `UncarriedLine.Text` verbatim. On a CRLF `CHANGELOG.md`, every record keeps its trailing `\r`, and that `\r` reaches the terminal. ## Measured This was a Go probe on `main` @ `2060ed1`, 2026-10-02. It made the same calls `rt prep` makes: `parser.UnreleasedContent(clBytes)` (`cmd/rt/prep.go:195`), `changelog.UnreleasedStartLine(clBytes)` (`:298`), then `prep.Compose`. The CHANGELOG used CRLF line endings, with this `[Unreleased]`: ``` Prose. ### Notes - n ``` `Compose` returns `*prep.UnreleasedDropError`. Detection is correct, with the same lines as on LF input. The refusal's second line is: ``` " CHANGELOG.md:5: Prose.\r <- before the first ### heading" ``` In a terminal, the `\r` returns the cursor to column 0. The reason after it is then drawn over the start of the line. Here the reason is the longer part, so `CHANGELOG.md:5: Prose.` is no longer visible. ## Why it matters The refusal exists to point the operator at the exact lines to fix (#1486). #1485 made this repo check out LF, but an adopter's `CHANGELOG.md` can still be CRLF. There, the refusal shows reasons without the lines they belong to. ## Acceptance criteria - [x] On a CRLF CHANGELOG, `rt prep`'s refusal contains no raw `\r`, and it names the same file lines as on LF input. — DONE @ #1508 (`de49186`, reviewed as `de49186`). `TestCompose_refusalOnACRLFChangelog` reads the CHANGELOG with the two calls `rt prep` makes and asserts that the CRLF refusal is byte-equal to the LF one, naming lines 5, 7 and 9. - [x] `UncarriedLine.Text` stays verbatim; only the rendering strips the `\r`. — DONE @ #1508 (`de49186`, reviewed as `de49186`). The new `UncarriedLine.Render` strips it at print time, and both `rt prep` and the fold-back refusal use it. The test checks that each CRLF `Text` keeps its `\r`. - [x] LF output stays byte-identical. Its tail is already pinned in `internal/prep/unreleased_drop_test.go`. — DONE @ #1508 (`de49186`, reviewed as `de49186`). The new test pins the whole LF refusal and passes against `2060ed1`'s code. The literal tail pin is unchanged. - [x] Tests are red on `main` and mutation-verified (AGENTS.md section 8, Axis B). — DONE @ #1508 (`de49186`, reviewed as `de49186`). The test is red against `2060ed1`'s code at the raw-`\r` and CRLF-vs-LF assertions. Mutants: 6/6 killed at `5682563`. At `de49186`, the red-on-main check, M1, M5, M7 and an added-kind probe were re-run, and review 7169 reproduced them. ## Notes - #1493 solved the same problem in `changelog.FoldBackDropError.Describe`, with an unexported `unCR` in `internal/changelog/foldback.go`. One exported rendering helper used by both refusals would keep them from drifting. #1493 did the same for the accepted-headings note and the remedy (`changelog.AcceptedHeadingsNote`, `changelog.UncarriedRemedy`). - Raised in the review of #1493 (review 7158 finding 2; re-review 7160).

claude added the

labels

2026-10-02 12:28:50 +02:00

claude referenced this issue

2026-10-02 12:29:16 +02:00

recover-pending-cut fold-back silently drops [Unreleased] text outside an accepted ### heading #1491

claude referenced this issue from a commit

2026-10-02 12:40:21 +02:00

fix(prep): print the [Unreleased] refusal's lines without a CRLF file's \r

claude referenced this issue

2026-10-02 12:43:40 +02:00

fix(prep): print the [Unreleased] refusal's lines without a CRLF file's \r (#1507) #1508

claude referenced this issue from a commit

2026-10-02 12:56:28 +02:00

test(prep): build the CRLF refusal golden's tail from the shared helpers

release-bot referenced this issue

2026-10-02 13:24:13 +02:00

chore(release): v0.64.0 #1494

claude commented

2026-10-02 13:27:34 +02:00

Author

Member

Landed on main as de4918684856ff3ef364de2381866e5e7576bd42 via #1508 (rebase merge, 2026-10-02 13:23:04).

Landing vs review. It was a fast-forward: 5682563 and de49186 landed with their reviewed SHAs. git range-diff 2060ed1..de49186 2060ed1..origin/main shows = for both, and the trees match.
Approval. alex, review 7171: APPROVED, official, bound to de49186, the merged head. It was submitted at 13:08:48, after the HOLD for the review-7168 nit was released (HOLD comment 113946 at 12:52:20, released by 113949 at 12:57:37), so it follows the AGENTS.md section 6.2 order.
Other reviews. Claude reviews 7168 (at 5682563) and 7169 (at de49186) were posted as COMMENT, because that account authored the PR. Both give the verdict APPROVE, and 7169 re-read the head after the HOLD was released.
Checks. PR CI on de49186: 30/30 success, read right before the merge. push:main on de49186, at the time of writing: 14 success, 1 failure, and 1 pending (release / fire-cut, waiting to run). The failure is go-ci / record reviewed vs landed commit (push). Its log (run 18842, job 2) reads comment page 2 repeated the previous non-empty page; endpoint did not advance; refusing to grade, which is the recorder defect in #1495, not this change. check-self-bootstrap (push) is green, as the PR's re-pin section predicted.
Rolling PR. The push:main prep regenerated #1494 at the landed head, and its body now carries this change's fragment line. So rt prep picked up the change from main with no re-pin.
Closing references. Before the merge, the PR title, body, both commit messages, both comments and all three review bodies were scanned with the AC-closure gate's regex: no closing keyword. The CI ac-closure-check context on de49186 passed.
Branch. claude/1507-prep-refusal-crlf was deleted after the range-diff check.

All four acceptance criteria are ticked above (DONE @ #1508). Closing by hand.

Landed on `main` as `de4918684856ff3ef364de2381866e5e7576bd42` via #1508 (rebase merge, 2026-10-02 13:23:04). - **Landing vs review.** It was a fast-forward: `5682563` and `de49186` landed with their reviewed SHAs. `git range-diff 2060ed1..de49186 2060ed1..origin/main` shows `=` for both, and the trees match. - **Approval.** alex, review 7171: APPROVED, official, bound to `de49186`, the merged head. It was submitted at 13:08:48, after the HOLD for the review-7168 nit was released (HOLD comment 113946 at 12:52:20, released by 113949 at 12:57:37), so it follows the AGENTS.md section 6.2 order. - **Other reviews.** Claude reviews 7168 (at `5682563`) and 7169 (at `de49186`) were posted as COMMENT, because that account authored the PR. Both give the verdict APPROVE, and 7169 re-read the head after the HOLD was released. - **Checks.** PR CI on `de49186`: 30/30 success, read right before the merge. push:main on `de49186`, at the time of writing: 14 success, 1 failure, and 1 pending (`release / fire-cut`, waiting to run). The failure is `go-ci / record reviewed vs landed commit (push)`. Its log (run 18842, job 2) reads `comment page 2 repeated the previous non-empty page; endpoint did not advance; refusing to grade`, which is the recorder defect in #1495, not this change. `check-self-bootstrap (push)` is green, as the PR's re-pin section predicted. - **Rolling PR.** The push:main prep regenerated #1494 at the landed head, and its body now carries this change's fragment line. So `rt prep` picked up the change from `main` with no re-pin. - **Closing references.** Before the merge, the PR title, body, both commit messages, both comments and all three review bodies were scanned with the AC-closure gate's regex: no closing keyword. The CI `ac-closure-check` context on `de49186` passed. - **Branch.** `claude/1507-prep-refusal-crlf` was deleted after the range-diff check. All four acceptance criteria are ticked above (DONE @ #1508). Closing by hand.

claude closed this issue

2026-10-02 13:27:34 +02:00