What steps will reproduce the problem?
[Theoretical, never done in practice]
1. Create a wave with several blips, add participant X.
2. User X logs in to two clients, then goes offline (e.g. disconnect network).
3. X opens the wave on both clients, reads some blips in one client, reads
other blips in the other client.
4. X goes back online on both clients.
What is the expected output? What do you see instead?
Expected output (ideally) is that the reading actions merge concurrently. e.g.
any blip read on either client becomes read on both clients. Each client
should display this resolution live, without either client needing to re-open
or refresh. Additionally, re-open or refresh should be a no-op after this
resolution: all read-state should be the same.
Instead, theoretically, both clients will have no change on reconnection, but
when re-opening that wave, the reading actions from only one of the two clients
will be preserved.
The bug is simple: the embedding of WaveletReadState objects in the m/read
documetn does not resolve multiple <wavelet> elements with the same id.
A fix for this must preserve the cleanup-on-write property. i.e., active
resolution of the XML, by merging the data in the XML as soon as duplicates are
observed, is not acceptable. This makes a fix not completely trivial.
Since the data types embedded inside each per-wavelet read-state object element
are monotonic and superimposable, one solution is for WaveletBasedSupplement to
collect all WaveletReadState objects that exist for the same id, and wrap them
in a composite WaveletReadState object, that distributes all queries over each
component read state object. On any write subsequent action, for any wavelet
with multiple <wavelet> elements, a new <wavelet> object can be added that
contains the sum information of all the others, and then all the original
<wavelet> elements can be deleted.
Doing this completely correctly, with respect to composition, intermediate
transitive states, left-to-right broadcast of document events, and a sensible
resolution of multiple WaveletReadState objects, may or may not be difficult.
Original issue reported on code.google.com by david.he...@gmail.com on 2 Dec 2010 at 5:12
Original issue reported on code.google.com by
david.he...@gmail.comon 2 Dec 2010 at 5:12