Luci · Find the runtime regression suite + run it to see 5 f
Luci
Find the runtime regression suite + run it to see 5 failures.
Luci
rtk filtering names. Bypass to get failure names.
Luci
6 failures, 2 clusters: snapshot-reconcile semantics (MC-4656 caused) + workbench JS asset tests (maybe pre-existing). Check if cluster B pre-exists on parent commit.
Luci
Commit msg claims "5 pre-existing unrelated" — suspect, since 66d8e08 touched workbench JS/html/ticket.html. Verify against parent in temp worktree.
Luci
All 6 fail on parent too → MC-4656 introduced **zero** new failures; they're pre-existing. But 3 are squarely in MC-4656's domain (needs_input reconcile semantics). Read the tests to decide correct fix.
Luci
Tests define the correct intent: snapshot marks stale but must NOT flip to needs_input (stay in_progress). That's MC-4656's actual goal. Read app.py snapshot code + what 66d8e08 changed in tests.
Luci
MC-4656 didn't touch the 6 failing tests. The 3 needs_input tests are the canonical assertions of MC-4656's goal (snapshot must not flip in_progress→needs_input) and they're RED — fix belongs in this ticket. Read snapshot code + the needs_input test + investigate the 3 workbench tests.
Luci
Found the source flip: app.py:1071-1073 — `if stale: reconciled_status = "needs_input"`. That's the false-needs_input origin. Tests want stale in_progress to keep status (reconciled None) + set failure_reason, and let Control Room requeue. Read the other needs_input test + control_room recovery + the 3 asset tests.
Luci
`test_no_false_positive...` is a luci_operator test (different module). Get clean tracebacks for the 3 non-JS logic tests.
Luci
3 distinct false-needs_input sources, all squarely MC-4656 scope:
1. app.py snapshot flips stale in_progress→needs_input (2 tests)
2. luci_operator `_has_unanswered_worker_question` fires with no QUESTION: marker (1 test)
But changing the snapshot may strand MC-4656's recovery (it targets needs_input rows). Read the recovery code added by MC-4656.