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 ff354ee
Browse filesBrowse the repository at this point in the historyBrowse files
Fix Android showcase capture and rebuild v2 queued rows
The Android showcase job died two seconds after the environment server
started, printing one line: "cannot rollback - no transaction is
active". That error was the masking, not the failure. The seed opens
state.sqlite while the server it just started is still writing, so
BEGIN IMMEDIATE loses the write-lock race and returns "database is
locked"; the catch block's unguarded ROLLBACK then threw over it. A
local probe against a fresh server reproduces both strings — 1 failure
in 40 attempts on an idle machine, and none in 120 once the connection
carries a busy timeout. Give the seed a 30s timeout, guard the
rollback, and print stacks from the harness: it only fails in CI, where
the line that threw is the whole diagnosis.
Queued tasks rendered in Thread List v2 with the v1 row design — a bold
`text-lg` title, a zinc "Pending" pill, and a chevron — so the one row
you cannot act on shouted louder than the live work above it. Rebuild
them in the v2 card idiom (project line, title, `branch · machine`)
under a "Pending" divider that mirrors "Settled", with "Queued" in the
status slot. The label stays uncolored: v2 reserves hue for act-now,
in-motion, and broken, and a queued task asks nothing of the user.
The showcase fixture had no settled threads, so the list screenshotted
as one undifferentiated block. Seed three (one per project, 5h/28h/2d)
with an explicit settled override, and drop the oldest unlabeled active
thread so the settled tail starts above the fold on a phone.
Co-Authored-By: Claude <[email protected]>
0 commit comments