You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 1f73d95
Browse filesBrowse the repository at this point in the historyBrowse files
fix(t3x): apply the sync test-timeout to every gate, not just fork CI (#32)
Follow-up to a393ff0, from an independent review of it. Three real defects,
two of them regressions that commit introduced.
1. The timeout override never reached the pipeline that needs it most.
a393ff0 fixed only t3x-ci.yml. The daily sync runs its own verify via
scripts/t3x/sync-upstream.sh (`$RUN $script`, plain `vp run test`), and the
resolver prompt tells the agent to run plain `vp run test` too. So the next
upstream delta would rebase cleanly, fail verify on the exact test that was
just diagnosed, and escalate as `verify-failed` — reopening issue #29's
class of failure through a different door. Threaded through all three.
2. `--testTimeout=60000` silently HALVED apps/server's budget.
apps/server/vite.config.ts pins testTimeout: 120_000, with a comment saying
its sqlite+git+orchestration tests "can exceed the default budget on loaded
CI hosts". The flag applies to every package, so 60s cut the heaviest suite
in half — the same failure class the override exists to prevent. Now
120000, the highest value any package configures, so it lowers nothing:
web 15s -> 120s, root 60s -> 120s, server 120s -> 120s. The old comment
claiming it "only stops apps/web's override" was simply false.
3. Both SEAMS.md churn commands always returned 0.
Git's approxidate silently ignores a relative suffix on an absolute date, so
`--since="<iso> -60 days"` means `--since="<iso>"` — an empty window. The
previous wrong-but-working command was replaced with two non-working ones,
and since the doc says "if the ledger and this document disagree, the ledger
is right", a literal regeneration would have zeroed every churn and risk
figure. Now uses `--since="@<epoch>"` arithmetic, with the trap documented
and a self-check value (ChatView should return ~59, not 0).
Also corrects SEAMS.md against the tree a393ff0 actually created: merge-base
b64ae88 (not 887dd6e), totals +1468/-112, claude.md row +78/-31. That
commit edited a seam file without updating its own row, so a new rule says to
update the row and totals in the same commit — the ledger measures the tree a
commit creates, not the one it started from.
Finally, records in the runbook that an upstream sync is the sanctioned
exception to AGENTS.md's "do not run repo-wide checks" — a rebase can break any
package, so the full suite is the point.
Every file changed here is fork-owned, so the ledger totals above stay accurate.
Co-authored-by: Claude Opus 5 (1M context) <[email protected]>
|`apps/mobile/…/T3ComposerEditorModule.kt`| +1 | 1 |**1**| Event-name list entry |
70
-
|`docs/providers/claude.md`| +76/-31 |~0 |**~0**| Fixes the broken multi-account recipe. Near-frozen upstream (last touched 2026-04-29) — **the one row worth upstreaming**, which would remove it |
75
+
|`docs/providers/claude.md`| +78/-31 |~0 |**~0**| Fixes the broken multi-account recipe. Near-frozen upstream (last touched 2026-04-29) — **the one row worth upstreaming**, which would remove it |
0 commit comments