Re: RFC: AbiTalk communication protocol suggestion

From: Tomas Frydrych <tomasfrydrych_at_yahoo.co.uk>
Date: Sat May 21 2005 - 19:00:27 CEST

Hi Martin,

> There is one big issue with the token-based blocking of local
> changes to the document. We need intercept calls inside
> PD_Document and check whether it's OK to proceed. I guess this is
> not too much code, but it more intrusive than the other changes
> needed to implement this.
>

I have done some more work on the protocol to minimise the effects of
locking, but AFAICS it cannot be completely avoided. This is because in
order for the conversation partners to be able to maintain local
mappings for coordinance translation they have to establish an initial
synchronic state, and due to network lag I think this cannot be achieved
without a temporary lock out (there is some more on this in the latest
version of the proposal, which now lives at http://www.abisource.com/~tf/).

I have had a look at the inkboard website; from what I found in the
documents, there are four different protocols discussed in an outline.
They are all fine in principle, but none of them has any error
resiliency: they are based on broadcasting one-way messages, and there
is no mechanism ensuring the integrity of the documents (on reflection,
this might be less of a problem for an svg tablet than our document
stream). I think error resiliency is essential to this working well; we
should not assume that all parties will receive all messages or that
none of them will get corrupted in transit -- we need to reject messages
that are based on document that is out of sync.

I need to reflect on the conflict resolution, which with the reduced
locking needs to be rethought, and the inkboard work is very helpful in
this regard.

The inkboard project is also very tightly integrated with Jaber, and
while we should not commit ourselves exclusively to a particular
transmission medium, Jaber lends itself well to real-time collaboration,
so there is lot to be learned there.

What is emerging from my thinking about all of this, and from the
various comments made, is that there are quite a few things that would
need to be implemented in AbiWord before collaboration becomes viable; I
have added section to the doc that starts listing these; these are I
think issues that we might want to consider for 2.6

Tomas
Received on Sat May 21 19:00:36 2005

This archive was generated by hypermail 2.1.8 : Sat May 21 2005 - 19:00:38 CEST