[brettheap]

Three small precision fixes that lane openRepoTools-4 took over from lane openRepoTools-3 (TAKEOVER comments on each issue), and a note that #227 settled the fourth. The commits, oldest first: d77504c the proving tests, 1a22dd7 lane-end (#96 and #90 part a together), 3296573 the docs row, 42a1085 the tests.yml comments (superseded by #227, see below), 2c5b3f2 and 748e0ef the first fix round (the review's character-count finding, then the hygiene rule on wc), the merge of origin/main with #227, the merge with #212, then the second fix round (03a60f9 the tests for text that is not UTF-8 and for a failing wc, f2d559d the validity check, ffcca3c the validity check taken from an iconv -c round trip) and the merge with #216, #199 and #201. Brett Heap lifted the closeout direction for these follow-ups.

#96 — lane-end's synopsis said a DUPLICATE retire "ends nothing"

Wrong: the top-level USAGE entry read retire a live FORK (pid or uuid) or a DUPLICATE (pid); ends nothing. True of a FORK; false of a DUPLICATE HOLDER, whose retire TERMs the bg-pty-host pair (#39). --help prints that synopsis.
Fix: lane-end lines 30-34. The entry is split in two: a FORK "writes nothing and kills nothing"; a DUPLICATE "TERMs a background holder's pair, only reports any other; the row is not written", matching the OPTIONS paragraph below it.
Reconcile on landing: open PR #205 rewrites the duplicate-holder retire from "TERMs the pair" to "the bg-pty-host parent and every child". The synopsis here says "pair". The two PRs do not conflict textually, so git will not flag it: whichever of #202 and #205 lands second should reconcile the word.
Test: tests/test_precision_90a_96.py::test_the_synopsis_says_what_a_duplicate_holder_retire_does and ::test_the_synopsis_keeps_the_fork_and_the_duplicate_on_their_own_lines read the USAGE block of lane-end --help entry by entry, because the OPTIONS paragraph already says TERMs and a whole-help match would pass on main.

Closes #96

#108 — the TAKEOVER grammar table omitted the lane: form

Wrong: the table documented only TAKEOVER ← <stale claim URL>; the prose at the stale-claim paragraph (and the code) also write TAKEOVER ← lane:<dispossessed lane>.
Fix: docs/README-lanes.md line 678, one row, issue or PR, worded like its neighbours (a kind and the verb with its payload). Docs only, so no test.

Closes #108

#126 — nothing left to change here: #227 settled it

The branch first corrected two tests.yml header comments (item 2 and item 3 of the issue; commit 42a1085). Brett Heap's #227 ("run the test suite on Linux only") then rewrote that header and removed tests-windows, parse-macos and tests-macos, so the merge of origin/main takes main's tests.yml whole and this PR no longer touches it. Checked against main's new text:

• item 3 (the checkout invariant omitted tests-no-submodule): main now reads "submodules: true, tests-no-submodule excepted: it checks out submodules: false on purpose". Accurate.
• item 2 (workflow_dispatch was not named among tests-macos's homes): tests-macos is gone. The new on: comment says "push (main only) and the nightly schedule run those same three jobs", and workflow_dispatch runs them too. That is a smaller omission in ci: run the test suite on Linux only #227's own text, left alone here.
• item 4 (AGENTS.md listed the pull_request jobs wrongly): ci: run the test suite on Linux only #227 rewrote the rule; it now names guard-launch-mode, tests and tests-no-submodule and the four triggers, which is accurate.
• item 1 (a hygiene test pinning labeled and tests-macos's if:): not added, and moot, since no job reads the ready label now and the if: is gone. ci: run the test suite on Linux only #227 deleted the macOS parse hygiene test, and added no replacement for the trigger text.

AGENTS.md and tests/test_repo_hygiene.py are untouched by this PR.

Refs #126 (items 2, 3: settled by #227, no change here)

#90 part (a) — cell_for_scan skipped O1's grammar

Wrong: lane-end's narrowed scan of a register state cell checked the STATE word, the UTC shape and the part count, but not ratified decision O1's other rules for the phrase's line (at most 240 characters, no |, no CR/LF). A cell whose third part broke one was narrowed to its state word, so an open LANDING in that prose was never seen and the lane ENDED. That is the no-object-log fallback, the one path that still decides a merge hold from prose, and Rule 6 stops being enforced for it.
Fix: lane-end cell_for_scan and its comment (lines 1726-1789). The cap, | and CR/LF are checked beside the existing part-count test, and a cell that fails any of them is scanned whole, like a diary: fail-closed. Both call sites (this checkout's row and the published row on origin/main) share the function.
The cap is 240 characters, and ${#x} is bytes under a C locale, so cell_for_scan counts characters itself, the same in every locale: past 240, and only if the text is valid UTF-8, the count is redone with the continuation bytes taken out (tr -d '\200-\277' | wc -c | tr -d ' ', the padding strip test_no_shipped_bash_reaches_for_gnu_only_utilities_unaccompanied asks for). Without that, a valid 240-character line of — and → read as past the cap under LC_ALL=C and a lane merely waiting for a LANDING was held (the review's first finding). Text that is not valid UTF-8 keeps the byte count, which can only over-count: stray continuation bytes follow nothing and are no characters, and taking them out undercounted them, so a hand-edited non-UTF-8 cell of 241 bytes was accepted as 240 (the review's second finding). Validity is taken from an iconv -c -f UTF-8 -t UTF-8 round trip (the text is UTF-8 when it comes back unchanged), not from iconv's exit status: the bash:3.2 image's musl iconv exits 0 on invalid input, which the first version of the check trusted and so accepted all three cases there. A missing iconv or a failing wc reads as a line over the cap.
set-row-state does not count characters in every locale either: lanes-edit.sh's row_state_check counts bytes under LC_ALL=C, and a 240-character line of —/→ is refused there as "720 characters". That is filed as #234 and not touched here.

Tests (all but the controls fail on main, see below), each run under LC_ALL=C and a UTF-8 locale: tests/test_precision_90a_96.py

• test_a_line_over_the_cap_cannot_hide_an_open_landing (a 241-character line ending LANDING #7)
• test_a_line_with_a_pipe_or_a_carriage_return_cannot_hide_an_open_landing (two cases)
• test_the_published_row_is_read_whole_when_it_is_not_the_phrase
• test_a_wide_line_one_past_the_cap_cannot_hide_an_open_landing (241 characters of —/→)
• test_a_line_that_is_not_utf8_cannot_hide_an_open_landing (three cases: 241 stray \x80 bytes, a 241-byte Latin-1 line with one °, and a valid 240-character line with one stray byte appended)
• test_a_count_that_cannot_be_made_is_a_line_over_the_cap (a stub wc that fails on -c; the || cfs_len=241 fail-safe; C locale only, since a UTF-8 locale never recounts a 240-character line)
• controls that must keep passing: test_a_line_exactly_at_the_cap_is_still_prose (240 characters naming LANDING #7 is a lane waiting, not a hold), test_a_wide_line_exactly_at_the_cap_is_still_prose (the same in —/→, about 680 bytes, which fails under LC_ALL=C with a byte count) and test_a_line_over_the_cap_that_holds_nothing_still_ends.

Each case runs a real lane-end in the lane-start sandbox with a stub lane-worktrees, so a cell wrongly read as clear is a lane that ENDS.

Refs #90 (part a). Part (b), lane-start's refused append, has an open question and is not touched.

Not touched, and nearby writers

None of the off-limits files is touched (the installer, tests/conftest.py, tests/run.sh, tests/test_repo_hygiene.py, AGENTS.md, specs/004-migrate-to-triad/**, lane-handoff, lane-worktrees, lane-end ~200-215 and ~1971-2143, docs/README-lanes.md past ~3830). README.md and .github/workflows/tests.yml are unchanged.
Other writers in nearby spots, which these hunks do not overlap: lane openRepoTools-5 edits docs/README-lanes.md ~597 and lane-end ~1107-1300 and ~1401-1449; another lane-4 writer edits lane-end ~711.

Verification

All local runs under the shared pytest lock, with every LANES_* and TMUX* unset, on 2026-10-18.

Second fix round, red to green (tests/test_precision_90a_96.py, 26 cases: each lane-end case runs under LC_ALL=C and under a UTF-8 locale):

• Red: this head with lane-end put back to the first fix round's (27074e0: the tr recount with no validity check): 6 failed, 19 passed, 1 skipped. The six are test_a_line_that_is_not_utf8_cannot_hide_an_open_landing, its three inputs under both locales, each assert 0 == 2: lane-end exited 0 and ENDED the lane over the open LANDING. The skip is the failing-wc test under the UTF-8 locale, which never reaches wc.
• Green: tests/test_precision_90a_96.py with tests/test_repo_hygiene.py (main f5c3bcd's hygiene rules, including the bench-root path rule): 214 passed, 1 skipped, the same skip; the same 214 and 1 again on 8ff52c2.
• The failing-wc test (test_a_count_that_cannot_be_made_is_a_line_over_the_cap, C locale) pins the || cfs_len=241 fail-safe itself, red to green. A failing wc reads as a line over the cap either way, because without the || the count comes back empty and [ "" -le 240 ] is an error, which also reads as over the cap. That error is printed (lane-end: line 1771: [: : integer expression expected), so the test asserts "integer expression" not in result.stderr. Without the ||: 1 failed, 1 skipped ([C], at that assertion). With it, in the green run above: passed. And with a failing wc read as a SHORT line (|| cfs_len=0) it goes red as well: 1 failed, 1 skipped (assert 0 == 2, the lane ENDED over the LANDING).

bash 3.2: the extracted cell_for_scan ran on six cases (a valid 240-character line of —/→, a valid 241-character one, 241 stray \x80 bytes, a 241-byte Latin-1 line with one °, a valid 240-character line with one stray byte, an ASCII 240) under LC_ALL=C, POSIX and UTF-8, with bash 5.2 on the host and with bash 3.2.57 in the bash:3.2 image (musl iconv): all six correct everywhere, narrowed only for the two 240-character valid lines. The first version of the check, which trusted iconv's exit status, got all three invalid cases wrong in that image; that is why validity is an iconv -c round trip. With iconv off the PATH it fails closed (the valid wide line is held, the three invalid ones too).

CI: head ffcca3c, run 39086159056: tests, tests-no-submodule and guard-launch-mode green. Head 45e81c4 (the merge with main f5c3bcd), run 39094650878: tests, tests-no-submodule and guard-launch-mode green. Head 8ff52c2 (the pinned assertion), run 39112708025: see the checks.

Earlier rounds, for the record: main f87fd5a plus only the first test file gave 6 failed, 2 passed (the pipe and published-row cases fail as assert 0 == 2, the lane ENDED over the open LANDING); the first fix round's byte-counting lane-end against the extended tests gave 1 failed, 17 passed (test_a_wide_line_exactly_at_the_cap_is_still_prose[C], exit 2 where 0 is right: the review's false hold), and 18 passed once fixed.

CI's Linux tests job is the suite of record for everything else.

🤖 Generated with Claude Code

Summary by Sourcery

Harden lane-end state-cell interpretation and clarify its retire synopsis while completing the TAKEOVER grammar documentation.

Bug Fixes:

• Correct the lane-end synopsis so FORK and DUPLICATE HOLDER retire behavior is described accurately.
• Ensure malformed or overlong lane state cells are scanned as full prose, preventing hidden LANDING or CLAIMED markers from being missed.
• Handle character counting consistently across locales and fail closed for invalid UTF-8 or unavailable length counts.

Enhancements:

• Document the lane:<dispossessed lane> form of TAKEOVER state transitions.

Documentation:

• Add the missing lane-based TAKEOVER form to the lane grammar table.

Tests:

• Add precision tests covering retire synopsis output, malformed state cells, locale-sensitive character limits, invalid UTF-8, failing wc, and published rows.

[sourcery-ai]

Sorry @brettheap, you've used your own review budget of 250,000 diff characters for the last 7 days.

You can request another review in 23 hours and 12 minutes by commenting @sourcery-ai review. Upgrade to get a review now.

[sourcery-ai]

Reviewer's Guide

This PR makes four precision corrections: it aligns lane-end help text with DUPLICATE retirement behavior, hardens malformed state-cell handling with fail-closed scanning and regression coverage, documents the missing lane TAKEOVER form, and updates tests workflow comments to reflect actual checkout and trigger behavior.

Flow diagram for fail-closed lane-end state-cell scanning

flowchart TD
    A[Read register state cell] --> B{Valid phrase shape?}
    B -->|Yes| C[Scan state word and phrase parts]
    B -->|No: over cap, pipe, CR/LF, or wrong part count| D[Scan whole cell]
    C --> E{Open LANDING detected?}
    D --> E
    E -->|Yes| F[Keep lane open]
    E -->|No| G[Allow lane to end]

Loading

File-Level Changes

Change	Details	Files
Clarify lane-end's retire synopsis for FORK versus DUPLICATE targets.	Split the help synopsis into separate FORK and DUPLICATE entries. Describe DUPLICATE retirement as terminating the background holder pair while reporting other duplicates without writing the row. Add help-output tests that inspect synopsis entries independently of the existing OPTIONS text.	lane-end tests/test_precision_90a_96.py
Make malformed state cells fail closed instead of narrowing scans to the state word.	Require the O1 line-length, pipe, and CR/LF constraints before applying the narrowed state-cell parse. Scan invalid cells in full for both the checkout row and the published origin/main row. Add sandboxed lane-end tests for overlong, delimiter-containing, published, boundary-valid, and harmless malformed cells.	lane-end tests/test_precision_90a_96.py
Document the lane-based TAKEOVER grammar form.	Add TAKEOVER ← lane:<dispossessed lane> to the issue-or-PR grammar table.	docs/README-lanes.md
Correct workflow header comments to match checkout and trigger behavior.	Identify both parse-macos and tests-no-submodule as intentional submodule-disabled checkouts. Include workflow_dispatch among tests-macos's documented trigger sources and reflow the explanatory comments.	.github/workflows/tests.yml

Assessment against linked issues

Issue	Objective	Addressed	Explanation
#96	Update lane-end’s top-level USAGE synopsis so it distinguishes FORK retirement, which writes and kills nothing, from DUPLICATE HOLDER retirement, which TERMs the holder pair.	✅
#108	Update the canonical TAKEOVER grammar table in docs/README-lanes.md to document TAKEOVER ← lane:<dispossessed lane> alongside the existing stale-claim-URL form.	✅

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it. You can also reply to a
  review comment with @sourcery-ai issue to create an issue from it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time. You can also comment
  @sourcery-ai title on the pull request to (re-)generate the title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time exactly where you
  want it. You can also comment @sourcery-ai summary on the pull request to
  (re-)generate the summary at any time.
• Generate reviewer's guide: Comment @sourcery-ai guide on the pull
  request to (re-)generate the reviewer's guide at any time.
• Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
  pull request to resolve all Sourcery comments. Useful if you've already
  addressed all the comments and don't want to see them anymore.
• Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
  request to dismiss all existing Sourcery reviews. Especially useful if you
  want to start fresh with a new review - don't forget to comment
  @sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud