livqeno_chat 0.2.0

Livqeno Chat for Flutter — real-time messaging with durable history, presence, typing indicators, reactions, read receipts and threads. Connect, send, and listen without writing WebSocket code.

0.2.0

• Renamed from raven_chat to swalvo_chat (Raven/Livqeno is now Swalvo). Change the dependency to swalvo_chat and the import to package:swalvo_chat/swalvo_chat.dart. Every public Raven* type is now Swalvo* (RavenChat → SwalvoChat, …); the old names remain as deprecated typedefs until 1.0, so existing code compiles with deprecation warnings only. Wire contracts are unchanged: RAVEN_* error code strings, data-channel labels and protocol ids, and E2EE key-derivation labels stay as they were, so this release interoperates with every server and every client built on raven_chat.

Phase 3: the Flutter SDK catches up with @swalvo/chat, feature for feature. Everything is additive; an app written against 0.1 compiles and behaves the same.

• Catch-up recovery, which was missing. 0.1 reconnected and re-joined but never replayed anything: a message sent while the socket was down was simply never seen until the next history load. The client now keeps a resume point per room and, after every reconnect, replays missed messages from history (oldest first, de-duplicated), then the edits, deletes, reactions, pins and moderation events around them from the server's event log. Live frames that arrive meanwhile are held so they cannot overtake the replay. recovered reports the outcome.
• Event ordering. Conversation events carry an eventSeq; they are applied in order, duplicates are dropped, and a hole that does not close by itself within gapFillDelay (300 ms) is backfilled. When the log has expired (EVENTS_EXPIRED), the newest history page is re-read and published on resyncs.
• Outbox. send() queues every message until the server confirms it. A send the connection drops fails with queued: true and goes out again on reconnect with the same clientMessageId, so it is stored once. A send the server refuses is failed until outbox.retry() or outbox.cancel(). Implement RavenOutboxStorage to keep the queue across restarts.
• Inbox. queryMyConversations() and a live chat.inbox with unread counts and mentions, counted once across the page/socket overlap by seq.
• Moderation. block / unblock / listBlocks, flagMessage, and ban / unban / mute / unmute for moderators. New error codes muted, banned, blocked, slowMode (with retryAfterSeconds), messageBlocked (with reason), eventsExpired. A ban or a revoked membership arrives on removedFromRoom, and the room is no longer re-joined.
• Push. registerDevice / unregisterDevice / listDevices, setNotificationLevel, muteConversation.
• Messages. mentions on send, pin / unpin / listPins and the pins stream, search, paginated threadPage, linkPreview, replyCount, lastReplyAt and seq on RavenMessage, and messageUpdateEvents carrying the update's reason.
• Large rooms. presenceCounts and a paginated listPresence. Audience tokens (isAudienceViewer) join their room without naming it and refuse member-only calls locally.
• Attachments, which were missing. createUploadTicket, uploadAttachment (ticket, PUT straight to storage, confirm) and getAttachmentDownloadUrl.
• Fixed: HTTP errors were mapped on the canonical RAVEN_* code the API sends over REST, so every REST failure surfaced as unknown. The chat code (legacyCode) is preferred now. A bodiless 4xx/5xx is an error rather than an empty success.
• Fixed: a read frame's messageId was dropped, leaving RavenReadState.lastReadMessageId null for every live read receipt.

0.1.1

Two fixes from an external integration, both in behaviour the documentation promised and the code did not deliver.

• Fixed: presence was lost on every reconnect. Presence is socket state, and the new socket a reconnect opens knows nothing of what was set on the old one — so a client that survived a blip went silently invisible to everybody else for the rest of the session, while its own connectionState still read connected. setPresence() now remembers what it was asked for and re-sends it after each reconnect, ahead of the room rejoins so the server never processes a join for somebody it believes is absent. A client that never called setPresence() still announces nothing.

• Fixed: startTyping()'s documentation said it was safe to call on every keystroke. The server disagreed — typing frames are limited to 20 per 10 seconds, which a fast typist passes inside a sentence, and the 21st came back as an error banner mid-message. The repeats are now dropped client-side rather than sent and refused: at most one frame every two seconds per room, well inside the limit and well short of the indicator's own TTL. stopTyping() clears the window, so stopping and starting again still signals immediately.

• Documented: connect() joins a conversation; it does not announce the user. Two connected clients each see only themselves until they call setPresence(). That is deliberate and shared with every other Livqeno Chat SDK — a client that reads without appearing is an ordinary thing to build — but it was nowhere in the API documentation.

0.1.0

Initial release (Livqeno Phase 13).

• RavenChat — connect, join rooms, send and receive, with bounded exponential-backoff reconnection.
• Message history with cursor pagination, editing, soft deletion, threads.
• Presence, typing indicators, reactions, read receipts.
• Idempotent sends: a retry after a reconnect returns the original message instead of posting a duplicate.
• Typed errors sharing the web SDK's codes.