mirror of
https://github.com/zeronsh/comet.git
synced 2026-11-16 18:57:19 +00:00
* Remote chats: show the sync, not an outage, while a room catches up
Opening a chat hosted on another device over a slow edge showed a blank
transcript, then "Reconnecting — messages will send once the connection
recovers." with nothing sent, and every revisit repeated it. Each rejoin
of a chat first opened cold also rewound its cursor and downloaded its
rows again.
A remote chat that has not caught up now shows a loader in its
transcript and "Syncing…" in its composer. It reads as stalled only once
its room dial has stayed down past the engine's own connect and hello
budget, and the composer mentions sending only when there is something
to send. The engine certifies the cursor on its first verified write, so
the legacy repair runs once per snapshot. Side chats get the same, and a
fork opens its tab at once with its progress, error and Retry.
* Remote chats: before/after captures of the sync states
`remote-sync-fixture` attaches the production shell to a real engine, as
at launch, and plays the edge and the host over it: a chat opened cold,
revisited while its room rejoins, an outage with and without a draft, a
side chat opened cold, and a fork on a slow host. The same fixture ran
on main and on this branch; the captures are in
docs/screenshots/remote-chat-sync.
* Regression doc: say how the mutation runs were grouped
* Chat sync: pin the cursor an unverified row opens with
Once a write lands, the legacy cursor repair no longer runs for the
doc, so a stored cursor above the rows it holds would never be fetched
again. ChatPersistence::new opens a row whose cursor is not verified at
0, and no test covered it. With the zeroing removed, the new test fails
(left: 500, right: 0), and every other test still passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Regression doc: the zeroing the engine fix relies on, and where a rejoin still rewinds
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Remote chats: a sync that never delivers always stalls; calmer offline and fork-failure states
- The stall clock no longer restarts when the engine moves a room back to
waiting for a sync slot (contention handoff), so a room that never joins
reads "Reconnecting…" after 35 s instead of "Syncing…" forever, and a
pending send can still surface "Not delivered". Opening the chat still
restarts the hold and the stall; only delivery ends the clock.
- Transcript loader: hard offline shows the amber dot the composer and
sidebar use, not a spinner.
- Pending fork failure: a plain empty state ("Couldn't fork on {host}" with
the engine's reason as detail) and a ghost Retry, instead of the raw error
in danger red.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
* Regression doc: the fork-failed capture predates the restyle
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Wing <44917547+wingleeio@users.noreply.github.com>
Co-authored-by: zeronsh[bot] <346184242+zeronsh[bot]@users.noreply.github.com>
|
History
|
|
|---|---|---|
| .. | ||
| after-cold-open.png | ||
| after-fork-failed.png | ||
| after-fork-waiting.png | ||
| after-outage-draft.png | ||
| after-outage.png | ||
| after-revisit.png | ||
| after-side-chat.png | ||
| before-cold-open.png | ||
| before-fork-failed.png | ||
| before-fork-waiting.png | ||
| before-outage-draft.png | ||
| before-outage.png | ||
| before-revisit.png | ||
| before-side-chat.png | ||