…d PK.

Still need to think it through. Might need to find a better way than iterating all updates in entire DB periodically.

…se', callback).

Could not take latest as it generates double types. Found out we needed version 9.3.1 exactly.

* Renamed updTable to updatesTable
* Renamed id / rowId to parentId

* To lookup whether a certain document is open and find it

* updates
* awareness
* sync
* websocket

…ngs change)

Move all yjs related modules to an yjs directory

* updatesTable is not meaningful outside of client
* server need to know the parent table to verify access
* parent table + yDoc prop combined forms the same uniqueness as updatesTable.

Referring to dexie-cloud-common a better way for pnpm: with workspace:^

…n compressing docs.

* Update receivedUntil after getting server changes over syncWithServer() but not when getting chaotic server changes on a particular document over websocket.
* Before syncWithServer, compute an Y.js state vector from the server changes that has arrived after receivedFrom so that the server can avoid sending the same changes back.
* Update YSyncState after sync with the correct unsentFrom and receivedUntil values.

useYProvider() back to useDocument()
because the hook not only maintains a provider but also manages the lifetime of the document, keeping it loaded as long as the component is mounted.

* Opening a document on the server, providing a state vector of updates that we've got from the server over websocket but not through regular sync.
* Handling doc-open re-requests in case WebSocket goes down and up again while the document is open in an editor.
* Don't send the initial awareness state - wait for editor to manage awarenesses - except when the component is unmounted - then we remove our "client" from awareness state and also send a "doc-close" message

…ecome standard.

Store serverRevision on sync state for Y-tables in order to include it in the DocOpen request so that the client can get additional updates since last sync over the websocket immediately when the document is opened.

* After login, or when user is accepted to a new realm, all Y docs in the user's realms are downloaded in full over a binary stream from new endpoint in dexie-cloud-server with the ability to continue where it left off in case network would break in the middle.
* This is nescessary because clients only subscribe to updates on their current realms. When new access is given to user, it cannot just get the latest updates from last revision but need to fully download all the docs on the new realm.
* With non-YJS changes we are currently baking in the same flow in the ordinary sync, but for YJS we can safely do the downloading outside the normal sync API as the changes are omnidempotent and the CRDT types does not depend on an exact revision being in place.

* Downloading of y documents from dexie-cloud didn't work properly
* Didn't create dexie-cloud-syncer on updates table until after a long time
* Garbage collecting yDocs didn't work properly when dexie-cloud addon had been syncing.

…essages to server

…-hooks

Setting the three of them on the same alpha version

+ Clone state of provided Y.Doc into updates table
+ strip away physical Y.Doc property before storing to DB
+ throw when trying to replace an Y.Doc property on an existing object.
+ runtime type check of yDoc props.
+ new unit tests of the above

* Run it 10 seconds after Dexie instance created
* Then run it every 5 minutes.
* And don't update the timestamp lastCompressed unless anything actually could be compressed. Because otherwise we might starve the WAL log by writing every 5 minutes despite nothing is happening from the user's perspective.

…s integration isn't used.

…cs immediately

…he same client, and destroy doc if it is open.

…rying it out first like this.

…hat are never synced to server)