jason.woltje
b9ea9304fd
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.
1 line
4.0 KiB
JSON
1 line
4.0 KiB
JSON
{"body":"## Filbert review: row 184 round 1 (#1682), approve\n\nRequest 29395. Candidate manifest `e7e2c76308e1b00391bb758d66fb8b427aad3e2985f30c651f5ef84c02ee2630` (7 paths) on base `7fc98cf19`. The brief is the \"tester thread verbs in the CLI\" section of blob 3b4313cb9ffa. The design and the changed tests were accepted in sage-1208.\n\n**Verdict: approve.** No findings and no blocking notes.\n\n### What I ran\n\n- I built `~/filbert-scratch/r184-r1/wt` from `7fc98cf19`, then ran `git apply --index build.patch`.\n - `sha256sum -c` on the manifest passed 7/7, and the manifest file hashes to `e7e2c763...c02ee2630`.\n - The patch-id is `ad846db177b3`: 7 files, +345/-28.\n- Base drift: `git log 7fc98cf19..refactor` (3da362982) on the 7 paths is empty.\n- I then moved the worktree to `3da362982`, keeping the staged candidate. The manifest still checks 7/7 and the patch-id is unchanged. So my run has row 182's broker (`#testerOwns`) in it, where Dewey's run 2 was on 7fc98cf19 and their row 182 check was a code read.\n- I took the slot on sage-1226, with the /proc busy check CLEAR before and after, and TMPDIR `~/fs184`. The run went from just after 05:56:50Z to 06:01:19Z.\n - Singles: tester-threads 5/5, thread-cli 11/11, tester-reads 6/6, each rc 0.\n - The cli suite (`node --test tests/*.test.mjs` in packages/cli): 420/420 pass, 0 fail, rc 0.\n - The counts are awk sums over the spec logs.\n- My mutant M5 removes the `await members?.refresh();` that reads the thread before the first `message.receive` in `talk`. Killed: tester-threads rc 1, failing only \"with --tester a human thread refuses with the bus's code, and nothing is written\". Without that read, `follow --thread` on a human thread receives (and marks read) the tester's waiting reply before any refusal. cli.mjs was restored by `git checkout`.\n- Dewey's red base, M1a-M4 and run 1 and 2 numbers are their receipts, not mine.\n\n### Checked against the brief\n\n- **Item 1.** `threadCommand` (thread.mjs:130) takes `--tester` for create, open and read. `callFor` (cli.mjs:185-190) picks the tester transport when the flag is set and skips `refuseInsideAgent` only then. `threadsCommand` already took it (row 175).\n- **Item 2.** The refusal at the base's cli.mjs:361 is gone. `talk` builds its call through `channelCallFor`, which wraps `callFor`, so `--thread --tester` resolves the thread and sends with `thread.send` as the tester. The tester's `thread.list` lists only its own threads, so a label or prefix of a human thread doesn't resolve, and a full id gets the bus's `thread-not-owned` (test 3 covers both).\n- **Item 3.** The loop receives with `message.receive` only, and never calls `inbox`. Test 2 checks that for `--follow` and `follow --thread`, along with the thread filter (messages outside any thread and in the other tester thread stay unread).\n- **Item 4.** TOOLS.md and the cli README list `[--tester]` on each form, and the README's refusal list drops the thread verbs. I grepped packages/, docs/TOOLS.md and scripts/ for text that still says the CLI's thread verbs are human-only, and found none.\n- **No bus change.** Every verb the tester path calls (`project.list`, `thread.list`, `thread.create`, `thread.open`, `thread.send`, `thread.messages`, `message.receive`, `message.read`) is in `TESTER_VERBS` (broker.mjs:98-100).\n- **Brief tests 1-3** are new tests 1-3, plus test 4 for the forms without `--tester`. Test 5 runs the real `scripts/mosaic` and `tester-cli.mjs`.\n- **Changed tests.** thread-cli.test.mjs:311 and the tester-reads exit-4 list drop only the argvs that pinned the removed refusal. The rest of each loop and its writes assert stay.\n- **Readers and pins.** No 64-hex pin names a changed path. host.test.mjs:826 reads the cli README for a line this row doesn't touch, and it passed in my suite run.\n\n### Deviations\n\nI agree with all three. The usage text change (1) is visible only on a usage error. Item 3 needed no code (2), and my M5 shows the read-first order is what keeps a refused `follow` from writing. `thread.send` has no `waiting` field (3).\n"} |