5.2 KiB
Row253 re-integration onto 385a7194c
Authority: sage-2659. Branch gander/row253-385a preserves all previous seat branches. Source tip before this packet: 141b79c293. Ten commits replayed from 0fa7bf421..8de88458d onto 385a7194c.
The first design commit conflicted only on BUILD-LOG.md. Kept all upstream bytes and appended the row253 phase; later appends replayed. BUILD-LOG and DEFERRED keep 385a7194c as exact byte prefixes. BUILD-LOG stays outside the manifest and lands by append after code. README merged automatically; its four row253 replacement pairs are exactly the same as before. The row278 thread-claims documentation stays present. No runtime/test conflict or repair.
range-diff.txt identifies two context-only changes (commits1 and6), both BUILD-LOG context. All other pairs equal. patch-ids.json records all ten old/new pairs. The full patch-ids for commits1 and6 change because their BUILD-LOG context changed. Excluding BUILD-LOG, all ten patch-ids match. No runtime/test change.
The new seven-path manifest digest is 72a70e99ba0416ceb1f63ed737132ebbeb212332d1441c362b3dfa6c8b00209c. Only DEFERRED and README differ from8de88458d. The two host scripts, two test files and shared sudo fixture are byte-identical to Duckworth's approved candidate. manifest-drift.diff is the required 0fa7bf421..385a7194c diff restricted to the seven landing paths; upstream changed only DEFERRED(+79) and README(+9). drift.json lists every upstream changed path separately.
Test reads and pins, source-only inspection
The focused case1/case2 tests read the host scripts and temporary fixture files. case2 also reads case2-precheck.sh and the two service templates; those three assets did not change in 0fa7bf421..385a7194c. Both tests write their own config.mjs and webui cli.mjs stand-ins into scratch copies. They do not import the real Console cli/serve/public modules, and row278's real Console changes require no focused fixture change.
The earlier dependency claim is widened here: both tests import cli/tests/helpers.mjs, which imports Store/Broker. Store reads schema.sql and the migration chain and checks schema bytes at module load, even when a particular test only calls tmp(). Store, broker, schema.sql and 3j-3k.sql changed upstream. drift.json lists the conservative literal-import closure and its nine moved modules (including business access/credentials/role/vocabulary and library mapping). The closure includes generated-source literals and is not an execution trace. case2-precheck.sh also dynamically imports the actual cli config and business index; business imports changed and the whole focused file must cover them. cli config itself did not change.
case2 imports seat/src/sandbox.mjs, as Duckworth0116 N9 identified. It did not change since0fa7, and its sandbox test still may skip on sandboxProblem(). Report that exact skip in the gate; do not assume zero. The actual sandbox binder reads the repository file inventory, so changed dependencies are covered by the whole-file run if available. No host experiment was run here.
Whole CLI covers changed format/grants/instances/lifecycle/tester/stack-release tests and real cli modules. Whole WebUI covers row278 and its existing shell fixture repair on this base. stack-release.test.mjs T46 at916 reads actual schema.sql and classifies every table; Store's schema hash and migrations are another live consumer. Whole-shell calls can load queue/business/bus consumers too. No row253 test has a literal 3k/3l version pin to update by assumption. The relevant whole-CLI schema fixture strings in stack-release-fixture.mjs model old target copies intentionally; they are not pins on the current bus schema. No assertion is changed to anticipate264a.
264a's 3k-to-3l schema change has not landed on this base. Before CLEAR, compare its store/schema/migration and any test or asset it moves, and integrate on a new branch if Sage names a new base. A seven-manifest-path-only check is insufficient because the focused import closure reads schema at load time. No result carries from the historical row253 gates after the C4 product repair.
Gate preparation, no runs
SHELL-LIMITS.json uses the same successful whole-file receipt set as row278, plus every row278 run3 whole-shell result from385a7194c. Each limit is ceil(3longest)+120 seconds. Foundation's newer39.901s receipt replaces39.708726s; its limit remains240. Other maxima remain the row196 receipts. All ten scripts have receipts;900 is the no-record fallback only. Each current script hash is pinned. No scripts/test-.sh changed.
window.py now uses these whole-shell wrappers: timeout -k10 LIMIT bash scripts/test-X.sh, with the outer harness wait30s later. driver.py marks exits124/137 as timeout and never reports them as kills. These are harness-only changes; no product/test/fixture changed. No command was executed by either harness. All40 proposed string targets still occur exactly once, checked as bytes in uniqueness.json without running syntax or tests. Full C4 timing and kills remain unverified.
The next action is Duckworth's integration read, followed by Sage's CLEAR after281 and264a. No Node, test, syntax command, browser, probe, queue write, helper push or live action ran during this re-integration. Publishing the new branch needs a gap because the repo helper runs Node.