[brettheap]

What and why

This is part 2 of #170, stacked on #174, so its base is feat/lane-worktrees-residue. #174 takes every item that can destroy or publish something, or that decides a gate. This PR takes the rest of #170: report honesty, portability, the suite wrapper, the daily start, test honesty and nits. With both landed, nothing in #170 is left open, so this PR says Closes #170.

Items 3, 9, 10 and 11 were not in #170's PR 2 list. They are taken here so that "Closes" is true. (d) and B8 were begun by lane openRepoTools-3's writer on a local branch that was never pushed. They were finished and proved here.

Items, commits and tests

"Red" means the case failed with the base's (499d2ab, #174's head) lane-worktrees, lanes-edit.sh, lane-start, AGENTS.md and tests/run.sh, run with this branch's tests, in proof run 37507138793, job red.

item	where	commit	test	red at base
(d) a FOREIGN clone's date used %cs (git 2.21+)	lane-worktrees report	1d3b321	test_lane_worktrees_report.py::test_a_foreign_clones_date_needs_no_git_2_21	yes
B8 the report read .lane-state only beside checkouts the shape walk sees; an empty section said "none" beside unread records	report	0d623aa	test_an_empty_section_says_what_could_not_be_read_never_none	yes
C2 the nested-lane case used a shape the old walk handled	report test	0d623aa	test_the_report_finds_every_kind_of_leftover_and_changes_nothing (now a checkout in a plain directory inside a repository)	yes
G3 a FAILED register git grep read as "no events"	report _closed_lanes	6a32842	test_a_register_grep_that_failed_is_a_finding_never_no_events	yes
G10, G11, G12 strict UTF-8 on the workspace .gitignore and both --post writes	report	6a32842	test_bytes_that_are_not_utf8_never_abort_the_report	yes
E10 a bare os.listdir of the estate root	report evidence scan	6a32842	test_an_estate_root_that_cannot_be_listed_is_a_finding	yes
item 9 only ignored DIRECTORIES were evidence	report evidence scan	6a32842	test_an_ignored_evidence_file_is_found	yes
item 10 the .gitignore repair command's path was unquoted	report hygiene row	6a32842	test_the_gitignore_repair_command_survives_a_space_in_the_path	yes
E4 rmtree's retry chmod-ed through a symlink	sweep	bffab3e	test_a_failed_removal_never_chmods_what_a_symlink_points_at	yes
E5 one gh pr list --limit 1000 missed older branches' PRs	sweep prs	bffab3e	test_a_branchs_pull_request_is_found_however_many_the_repository_has	yes
E6 the NOT WRITTEN fallback line was unquoted	sweep	bffab3e	test_the_hand_written_register_line_is_quoted_for_the_shell	yes
E7 note_line did not pass LANES_SESSION	sweep	bffab3e	test_the_register_line_is_written_as_this_session	yes
C5 no case had a process scan that cannot be made	sweep (seam LANE_WORKTREES_SEAM_NO_PROCSCAN)	bffab3e	test_a_process_scan_that_cannot_be_made_keeps_every_tree	yes
C1 "LANDED beats OPEN" never asserted the remote branch's fate	test	bffab3e	test_merged_is_the_register_or_gh_never_ancestry	no, by design (see below)
C4 rescue-before-removal proved only negatively	test	bffab3e	test_the_rescue_is_on_origin_at_the_moment_of_removal	no, by design
C6 pid 999999 assumed dead	tests	bffab3e	test_killed_suite_sandboxes_whose_owner_is_gone_are_removed, test_a_tmp_dir_with_no_mark_of_a_suite_is_listed_until_it_is_old	no, by design
E1 a non-ASCII staged .pyc passed (C-quoted path)	lanes-edit.sh bytecode_in_pathspec	8e5a3c7	test_pathspec_check.py::test_a_staged_bytecode_file_whose_name_is_not_ascii_is_refused	yes
E2 add '<path>' parsed in the caller's locale	lanes-edit.sh	8e5a3c7	test_the_add_lines_are_read_in_the_c_locale	yes
G4 the offered .gitignore command's path was unquoted	lanes-edit.sh	8e5a3c7	test_the_offered_gitignore_command_survives_a_space_in_the_path	yes
item 3 stop_group reaped a TERM-deaf leader before the KILL	tests/run.sh	ea26b52	test_run_wrapper.py::test_a_leader_that_ignores_term_is_killed_at_the_deadline	yes
G8 a caller's --basetemp won over the wrapper's	tests/run.sh	ea26b52	test_a_callers_basetemp_is_refused[form0,form1]	yes (both)
G14 a terminal run's leftover child outlived the run root and lock	tests/run.sh	ea26b52	test_what_a_suite_on_the_terminal_leaves_running_is_stopped	yes
item 11 the daily report's directory inherited the umask	lane-start	4f6d122	test_the_daily_report_is_private_and_survives_its_panes_hangup	yes
E8 the report died with its pane's hangup	lane-start	4f6d122	the same case	yes
C3 "no report ran" proved by a one-second sleep	report tests	4f6d122	test_a_stamp_from_today_runs_no_second_report_and_yesterdays_is_replaced, test_a_failing_or_switched_off_report_never_fails_a_start, test_a_dry_run_start_runs_no_report	no, by design
E9 the by-hand suite line left caches	AGENTS.md, manual	0292005	test_repo_hygiene.py::test_the_suite_wrapper_takes_one_lock_and_names_it_where_agents_read_it	yes
G15 the manual missed the report's per-checkout parents	docs/README-lanes.md	0292005	(manual)	n/a

Passing at the base by design. C1, C3, C4 and C6 are test-honesty items. Each makes a case assert what it claimed to prove, and the base's code already behaves correctly there. So 7 of the 29 selected cases pass at the base, and that is the point: they now pin that behaviour positively instead of vacuously.

• C1 now asserts the remote branch is kept.
• C3 reads the start's own synchronous word instead of sleeping.
• C4 uses a git that checks origin at the moment of worktree remove.
• C6 uses a pid it started, reaped and proved gone.

Copilot round 1 on this PR

There were three threads, all real defects in this PR's own code, and each is fixed with a case. "Red" means the case failed with lane-worktrees and lane-start at 0292005 (the head before the round) and this branch's tests, in proof run 37513097772. At e66195c in that run, the lane_worktrees, run_wrapper, pathspec_check, repo_hygiene and openrepotools_command cases pass (376). The one failure is the macOS-job hygiene test, which reads the trimmed proof workflow itself.

thread	what it found	commit	test	red at 0292005
r4199103207 Report.markdown	any unread thing, an ignored-file listing included, turned an empty orphan or aging section's "none" into "unknown"	5b1fa79	test_lane_worktrees_report.py::test_a_failure_that_is_no_record_leaves_an_empty_section_none	yes
r4199103056 lane_daily_report	a chmod 700 that failed still started the report	68be6c8	test_a_report_directory_that_cannot_be_made_private_runs_no_report	yes
r4199103154 Sweep._note	the NOT WRITTEN line to paste dropped LANES_SESSION	e66195c	test_lane_worktrees.py::test_the_hand_written_register_line_is_quoted_for_the_shell (now wants the session word)	yes

Copilot round 2 on this PR

Review 5433148700 on e66195c opened no inline threads. Its two "previously missed" findings are real defects, and both are fixed. "Red" means the case failed with lane-worktrees and tests/run.sh at e66195c, in proof run 37514530247. At 7ac5cbe in that run the same selection passes (378); the one failure is the macOS-job hygiene test, which reads the trimmed proof workflow.

finding	commit	test	red at e66195c
lane-worktrees:5113: the report's untracked listing was taken whatever its exit	19e53be	test_lane_worktrees_report.py::test_an_untracked_listing_that_failed_is_a_finding_never_evidence	yes
tests/run.sh:175: the G14 stray scan counted its own subshell, ps and awk, so every terminal run waited out the ten-second deadline	7ac5cbe	test_run_wrapper.py::test_a_terminal_run_that_leaves_nothing_waits_for_nothing	yes (20.7 s: the scan waited out its deadline twice)

Design notes

• G14. (The scan runs in the wrapper's own shell and passes over its own children since 7ac5cbe.) A non-interactive bash never hands a job the terminal. Measured: set -m with fg left the suite stopped on SIGTTIN. So the wrapper reads its process group's members before the suite starts. Afterwards, whatever is in the group that was not there before, and is not a sibling the caller started, is TERMed and then KILLed after ten seconds, before the cleanup.
• E8. The report runs under nohup, and under setsid where there is one. Linux has it; macOS does not, and there nohup alone keeps the hangup away.
• E9. AGENTS.md stays at its 316-line cap: the one line changes in place, and a hygiene assertion holds it there.

Verification

tests/run.sh on the workstation was starved by other projects' suites, so CI is the record. Proofs ran on temporary proof/openRepoTools-1/* branches with a trimmed tests.yml, each deleted once read:

• run 37507138793, on this head:
  • head (-k 'lane_worktrees or run_wrapper or pathspec_check or repo_hygiene'): 291 passed and 1 failed. The failure is the macOS-job hygiene test, which reads the trimmed proof workflow itself.
  • red: 22 failed and 7 passed (the by-design cases above).
  • parse-macos: bash 3.2 parses every shipped bash file.
  • helpers: the shell suite under plain python3 -m pytest -k lane_helpers, as CI runs it. 22 passed.
  • helpers-base-runsh: the BASE tree's shell suite through tests/run.sh. 4439 ok and 1 FAIL, "the row's stamp states clause (c)'s two facts for a person, in its TAIL". That failure is the base's own, not this PR's. Under run.sh's long run root, the lane row's stamp is cut before its dir …; window … tail. The pre-rebase proof saw the same failure on this branch through run.sh. It is not touched here; it is reported to the lane.
• An earlier run, 37501364279, on this branch's pre-rebase head over 685cbe3, gave the same head and red results.

This PR's own CI is the landing record. I have not applied ready.

Closes #170
Refs #174 #176 #162

Lane: openRepoTools-1

🤖 Generated with Claude Code

Summary by Sourcery

Harden lane reporting, sweeping, daily startup, and test execution for honest failure handling, portability, privacy, and reliable cleanup.

Bug Fixes:

• Make lane-worktrees reporting accurately surface unread state, register, filesystem, and evidence data without treating missing information as clean results.
• Improve sweep safety by preserving trees when liveness is unknown, protecting symlink targets, finding pull requests across all branch histories, and emitting complete shell-safe register commands.
• Make daily reports private, retryable when permissions cannot be secured, and resilient to terminal hangups.
• Harden the test wrapper against TERM-resistant suites, caller-supplied temporary directories, and child processes left behind by terminal runs.
• Fix bytecode path checks across locales and non-ASCII filenames, and quote offered .gitignore repair commands.

Enhancements:

• Expand report coverage to discover lane state and evidence throughout nested estate layouts while remaining non-mutating and portable across older Git versions.
• Replace timing-based report tests and unreliable fake dead PIDs with positive behavioral assertions.

Documentation:

• Update contributor and lane documentation for clean test execution, session-aware sweep records, comprehensive report discovery, report privacy, and process cleanup behavior.

Tests:

• Add regression coverage for report honesty, portability, filesystem and encoding failures, sweep ordering and safety, shell path handling, daily report lifecycle, and test-wrapper cleanup.

[brettheap]

@codex review

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[Copilot]

Copilot review overview

🔵 Needs a closer look

Terminal cleanup always detects its own scan processes, and failed untracked-file scans can still produce incomplete reports.

Review effort: Balanced
Findings: None

Resolved since last review (3)

• Track record scan failures separately from other unreadable errors
• Preserve LANES_SESSION in recovery fallback command
• Abort report when directory permission tightening fails

Previously missed (2)

In code that hasn't changed since last review

Handle failures when reading untracked files

lane-worktrees:5113

The untracked listing is still treated as complete without checking its status. If this git ls-files fails, its empty or partial stdout silently removes untracked evidence from the report, whereas the ignored listing immediately above correctly records the failed read. Record this failure too and consume stdout only on success.

Exclude scanner processes to avoid guaranteed ten-second delays

tests/​run.sh:175

strays sees the ps process (and the command-substitution shell) that it starts in the same terminal process group. Those PIDs were not in TTY_BEFORE, so even with no leftover suite process every call stays nonempty, executes all 50 sleeps, and adds a guaranteed ten-second delay to each terminal run. Capture the scan first and exclude the scanner and direct helper children before applying the baseline filter.

[brettheap]

Copilot round 2 (review 5433148700 on e66195c). This round opened no inline threads. Its two "previously missed" findings are both real defects, both fixed in 7ac5cbe's push, and each has a case:

• lane-worktrees:5113, the untracked listing taken whatever its exit. Fixed in 19e53be. A failed listing is now a row of "Records the report could not read", and its output is never offered as evidence. Case: test_an_untracked_listing_that_failed_is_a_finding_never_evidence.
• tests/run.sh:175, the stray scan seeing its own subshell, ps and awk. Fixed in 7ac5cbe. The scan now runs in the wrapper's own shell, writing to a file under the run root, and passes over the wrapper's own children (bash 3.2 has no $BASHPID). Case: test_a_terminal_run_that_leaves_nothing_waits_for_nothing (under 8 s).

Proof run 37514530247 (temporary branch, deleted once read):

• Head: 378 passed. The one failure is the macOS-job hygiene test, which reads the trimmed proof workflow.
• With lane-worktrees and tests/run.sh at e66195c, both cases fail. The terminal run took 20.7 s, because the scan waited out its deadline twice.

This was the second and last Copilot round for this PR. Lane openRepoTools-1.

@codex review

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[sonarqubecloud]

Quality Gate passed

Issues
30 New issues
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