Files
stack/agents/darkwing/work/queue-162/scratch-activate-ed5b41d4-notes.md
jason.woltje 158205c5cc lead: commit darkwing's untracked evidence (348 files: queue-137 queue-141 queue-150 queue-162 queue-165 queue-166 queue-167 queue-169 queue-179 queue-180 queue-181 queue-182 queue-185 queue-186 queue-187 queue-188 queue-190 queue-93)
Unclaimed files left in the canonical checkout. Inventory: canonical-cleanup-inventory.md (commit-as-evidence groups). detect-secrets scan clean on 2026-11-02: hits are sha256 pins, git object ids, test sentinels already in tracked tests and unit-file lines.
2026-11-02 15:05:14 -06:00

64 lines
3.3 KiB
Markdown

# Window B P3 rerun on the landed tool (sage-1070, mos-0867 R2)
Scratch only. Run by darkwing 2026-10-19T22:14:50Z to 22:14:59Z in
`~/darkwing-scratch/r1070`. The 4b16e848 run's files are unchanged beside
these.
- Tool: a `git clone --shared` of the canonical checkout at 4b738e2da
(the landed row 166), `refactor` in the clone at d195d171c.
`scripts/stack-release.sh` blob ed5b41d4677ee2bd85860e120c48387d440f9446.
bus-service.sh, webui-service.sh and stack-copy.sh blobs are the same
as in the 4b16e848 run.
- Copies: 5fa9cc8f2 and ca15d73e9, made fresh with `stack-copy.sh copy`.
Outside node_modules they match the r1004 copies byte for byte (`diff -rq`).
- Business file: `scratch-mkbiz.mjs` on the committed example (blob
3a81ca0faea6) with `{"usage-source":["transcript"],"channel":["cli","webui"]}`.
It matches scratch-business-B.json except the nosecret directory, so
its sha256 differs.
- Harness: scratch-B-harness with paths moved to r1070 and three changes
to run.sh. The backup dir comes from the data root plus the log's id
(addendum 1's fix). Step 2a runs `sha256sum -c ../SHA256SUMS` from the
backup's bus/ (addendum 2's fix). Step 0 records the unit reads the
row 166 checks see. teardown.sh is new.
Every file's sha256 is in SHA256SUMS-ed5b41d4.
## Result
All steps exit 0: plan, activate --pm --migrate (3h to 3i, backup
checked OK), D3 counts 0/0/0, rollback --restore --plan, rollback
--restore (back to 3h, pointer on 5fa9cc8f2). Teardown left no process
and closed port 47391.
Against the 4b16e848 run, after normalizing paths, times, ids and pids
(scratch-activate-ed5b41d4-vs-4b16e848.diff), the tool's output is the
same line for line except the `drift:` count in both status parts (39
and 47 then, 100 and 108 now), because `refactor` in the clone moved
from 6e7c23d7d to d195d171c. The other differences are the new parts
0 and 2a, a header line, and the D3 stamp now read from the right path.
## Row 166's W4 checks
`--plan` does not run them. `do_activate` and `do_rollback` exit after
`check_target; plan` (stack-release.sh:717, :727). `precheck`
(`drop_in_override`, then `read_units` with `unquoted`) runs at :718
and :728, and `status` calls `read_units`. So no `--plan` output
in this run or any other shows them.
| Check | Run here | Branch taken |
|---|---|---|
| drop-in scan, steps 2 and 6 | yes | pass: one unit path, no `.service.d` dirs, `DropInPaths=` empty |
| unreadable unit path or drop-in dir/file refuses | no | needs a mode-000 dir or file |
| drop-in with ExecStart or WorkingDirectory refuses | no | needs a `.conf` |
| no quote or backslash in bus/Console ExecStart argv | yes | pass |
| no quote or backslash in bus Environment= | yes | pass |
| a quoted or backslashed value refuses | no | the stand-in prints Environment= unquoted, unlike real systemctl, so it can't show this faithfully |
| gate copy-check names the target (item 3) | yes | pass on ca15d73e9, then 5fa9cc8f2 |
| bus stops during the held read (item 2) | no | needs a stop timed inside the read |
This run doesn't cover the refusal branches. The tests in
stack-release.test.mjs at 4b738e2da do: T4 (an override `.conf`), T28
(unreadable or relative unit paths and drop-ins), T29 (a bus that stops
during the held read), T30 (copy-check on another copy) and T31 (quoted
or escaped Environment= and ExecStart=).