Question about patching more than two models

[cristiano-belloni]

In the docs, you have this example of synchronization between users:

const blob = user1.toBinary();
const user2 = Model.fromBinary(blob).

user1.api.str(['text']).ins(5, ' Alice');
user2.api.str(['text']).ins(5, ' Charlie');

const patch1 = user1.api.flush().toBinary();
const patch2 = user2.api.flush().toBinary();

// Users can exchange patches and apply them
user1.applyPatch(Patch.fromBinary(patch2));
user2.applyPatch(Patch.fromBinary(patch1));

What if there existed another user that forked the first model immediately and wanted to synchronize with the two users at a later time? Something like:

const blob = user1.toBinary();
const user2 = Model.fromBinary(blob)
const user3 = Model.fromBinary(blob)

user1.api.str(['text']).ins(5, ' Alice');
user2.api.str(['text']).ins(5, ' Charlie');

const patch1 = user1.api.flush().toBinary();
const patch2 = user2.api.flush().toBinary();

// Users can exchange patches and apply them
user1.applyPatch(Patch.fromBinary(patch2));
user2.applyPatch(Patch.fromBinary(patch1));

const patch3 = user1.api.flush().toBinary();
user3.applyPatch(Patch.fromBinary(patch3))

Would this work? Because if it doesn't, all the users have to keep all the patches at all the various stages (which is a problem in case the users are separated programs) and if it does, I can't understand how the patches keep the whole non-incremental set of changes efficiently. I'm used to yjs, where users exchange vector clocks first to know where they are at, then generate ad-hoc patches based on the received vector clock. How does it work in json-joy's CRDT?