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 460a65b
Browse filesBrowse the repository at this point in the historyBrowse files
perf(mobile): thread list resumes from the saved cache, and large gaps reload over gzip HTTP
On a cold start every client downloaded the full thread list even when its
saved copy was a few events behind. The shell cache now records the server
snapshot its rows descend from (`origin`: server version, database
migration, omissions, and when the snapshot was taken), and a client whose
cache matches the current subscription and is under 12 hours old resumes
from the saved cursor instead. Replay never renews `snapshotAt`, so drift
from skipped older-settled cuts and rebuilt projections stays bounded.
When the gap is over the server's replay cap, the server used to answer the
resume with a full snapshot over the socket, which iOS does not compress.
A client that sends `httpSnapshotFallback` with `afterSequence` now gets a
single `snapshot-required` item instead; it drops the origin, loads the
gzip HTTP snapshot, and subscribes again from its cursor. Old servers ignore
the flag and old clients never send it.
Items buffered from a superseded session are dropped rather than applied to
the next session's rows, and a resume that catches up bumps `generation`
so thread families and search stop waiting for a snapshot.
Thanks to @Bil0000 for #30732 and @emmems for #30317, which found the
cold-start cost and the uncompressed socket fallback and shaped this fix.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
Co-Authored-By: Claude Opus 5.5 <[email protected]>
0 commit comments