[messaging] Matrix.org: Decentralized group chat

Trevor Perrin trevp at trevp.net
Mon Mar 9 18:57:39 PDT 2015


On Sat, Mar 7, 2015 at 4:13 PM, Ximin Luo <infinity0 at pwned.gg> wrote:
>
> For a private dynamic group chat, the merging requirements are not exactly the same as in a collaborative environment like Google Wave:
>
> - for a basic application, one does not *need* to merge arbitrary state. The only thing you need to merge is an unordered set (the membership set).


Yeah, I agree that merging in group messaging is simpler than the
"merging of edited documents" in revision-control and
collaborative-editing [1].

It's true that group members might simultaneously send messages or
change group "fields" (add/remove user, change group name).  There
might also be connectivity problems / delays between group members.
You could view these as edits to the group state in different
branches, which have to be "merged" by recipients into a consistent
view of (message history, membership, group name).  But this is
simpler than merging source code:

 (a) Text messages from users don't have to be edited together, they
just have to be displayed.  You could display the causal order in some
way, or display messages in arrival order.

 (b) For any group fields that can only be changed by a single admin,
merging isn't necessary, since causal order will indicate the latest
value.

 (c) If group membership can be added but not deleted, then merging is
easy (union of group members).

There are cases where these aren't sufficient, e.g. if group
membership could be added+deleted by multiple parties.  Ximin touched
on that, and it's worth discussing more.  But even that has fairly
specific features (e.g. unordered set), so makes me skeptical of the
need for general-purpose merge algorithms.

Trevor

[1] http://en.wikipedia.org/wiki/Operational_transformation


More information about the Messaging mailing list