I've started with example code for Quill editor that works fine, but if I replace
const type = ydoc.getText('quill') with something like const mymap = ydoc.getMap("event"); const mytext = new Y.Text(); mymap.set("description", mytext); const type = mymap.get('description');, Quill won't work in collaborative mode. I see that messages are being exchanged through websocket, but editor value isn't updated in my case.Both versions of the code return YText as
type, in original/demo version it is stored in the root of YDoc, while my modified version stores the YText inside a YMap. I see that YDoc object is being updated via websocket on clients by console logging on update action, but I can't figure out why the editor won't update the content. Any ideas?I'm trying to create a bit complex structure of the YDoc and have multiple editors/properties handled via single YDoc, so I'd like to avoid having all those editors inside YDoc root, I'd rather like them to be a part of my document structure. Is this doable in any way?
@thebata_twitter Yes, this is possible. A client could, for example, create a new document as a Y.Map property and initialize it with some content. Another client could iterate through the map, find the document, and then start working on that document as well. This can be useful to create some form of directory structure.
When you execute the above code for every client, each client will overwrite the description map and write a new Y.Text object. So they all work on different objects. Instead, only one client should create the document. Other clients can read it. Overwrite it only if this is your intended behavior. Also note that you can't bind data to a deleted Y.Text type.
In your case, it probably makes more sense to create another top-level type `"event-description".
ydoc.on('synced', () => ..) to make sure that you received initial content.
Hey @dmonad, thanks for the help! I'd still like to use the mentioned approach with a bit complex structure, so the solution you provided sounds good. I've tried to implement it as per your suggestion ( ydoc.on('synced', () => {console.log('synced')}) ) , but 'synced' event seems not to be fired at any moment.
Could you please tell me when 'synced' event should be fired and by which entity? I've searched through the source code, but couldn't figure that out...
I'm using WebsocketsProvider if it matters, tried with local one as well as with the public one from the quill demo.
I'm working on a MVP now and single ydoc per object on my side at the moment makes more sense. I get the idea about multiple providers and top-level props, will try that approach also at some point.
Still getting to know this awesome library that you've made, it will take some time and experiments to figure out all the possibilities :)
Hi guys and @dmonad, I have quick question.
Let's say we have prosemirror document: <p>Hello World<p>
in yjs, paragraph represented as YXmlElement the text 'Hello World' as YXmlText, right?
Now, there are 2 users that change the document concurrently. One splits the paragraph at the end of the word 'Hello'. Another one appends !!! to the end of the word World.
End result becomes: <p>Hello<p><p> World!!!<p>
as Yjs doc: 2 YXmlElements and 2 YXmlTexts, right?
My question is that, does yjs split the YXmlText into two? I seems it does not, then how in the world yjs is able to insert the !!! at the end of newly created second YXmlText? It somehow works perfectly, but I'm unable to understand how the position where !!! is inserted is tracked when the YXmlText was created newly?
<p>Hello World<p>, user A appends !!! and user B wraps the paragraph in <blockquote />, the end result should be <blockquote><p>Hello World!!!<p></blockquote>. However yjs loses appended !!! because wrapping in blockquote replaces the node making information about the insertion position for !!! gets marked as deleted.
Right, that's one of the downsides of the structured representation. From a certain perspective it may make sense. For example, userA transforms a paragraph to a Quote, adds comment markers and the author of the quote. And userB transforms the sentence to indirect speech. The expected result should be just either of the representations (the blockquote in Yjs), but with the possibility to see the differences (which you can do with snapshots).
It is basically impossible to retain the intention of the user if there are concurrent actions. Yjs could be better at merging concurrent changes. In practice though, it doesn't really make a difference. In any case, concurrent changes that resulted from editing while offline should be reviewed using diffs & snapshots. Concurrent changes from live editing are resolved immediately. The user doesn't really care how these conflicts are resolved.
live editing are resolved immediately
That's right, prosemirror author martin had also mentioned that in his blogpost.
I see that flat representation seems to be more beneficial to get most out of yjs. If paragraph had a boolean attribute that telling this paragraph should be wrapped in blockquote then there is no need for nesting. Same for list items, those can be solved with a simple flag mostly.
For tables, I have no robust solution in mind yet. Eventually, it would be really great to have a robust document structure representation encompassing most of these use cases like Quill's but a bit extended for tables, list items, blockquotes etc...
I see that flat representation seems to be more beneficial to get most out of yjs. If paragraph had a boolean attribute that telling this paragraph should be wrapped in blockquote then there is no need for nesting. Same for list items, those can be solved with a simple flag mostly.
I guess you could represent the prosemirror binding differently. If xml.nodeName would be an attribute, you could simply change xml.nodeName = 'blockquote'. But there are other examples when this might not be what you are looking for. The only true solution is to add move operations on ranges of types/text. If you assign a comment to a paragraph, and another users moves the same paragraph to a different location, you would expect that the comment will be retained.
@dmonad quick question, whats the right way to check when a YDoc has loaded if there is nothing to sync against? It looks like .observe and provider.on('sync') only fire when there's some data already in the YDoc.
Context: when a user loads in, I want X to happen if there's nothing in the YDoc, Y to happen if there is something in the YDoc, and I can't consistently get an event to fire when the YDoc has loaded fully.
Some additional context for why I'm asking, every so often my yMaps return as empty and I am unable to figure out why or how to restore the data. I keep garbage collection on for my yDocs so I am wondering if maybe that could be the problem, but it would help to look inside levelDB as well
It is unlikely that y-leveldb misbehaves. You would probably notice that you can't apply further updates (because of update dependency) when y-leveldb doesn't store a document update.
I suggest that you try to reproduce the issue. You could implement a simple sync approach like this to reproduce the behavior. Then post it to the issues and I'll debug it.