On Jun 07, 2007, at 12:17, Marcos Caceres wrote:
2) event-source is indeed one of the latest choices (see
availability in Opera: http://www.subbu.org/weblogs/main/2006/09/
server_side_dom_1.html). There are other server push techniques,
the oldest and most tested of them all is described on
pushlet.com. Yet this approach drains mobile's battery and has
many other problems, among them - timeouts by proxies and
firewalls, connection restore (spotty wireless coverage), etc.
All really good points. These is the kind of implementation
feedback we are trying to gather so please keep it coming.
Another option for server updates is XMPP. It is much more general
purpose, powerful, interesting, and better defined (well, the latter
really isn't hard) than SSE/event-source. It was demonstrated as an
approach at XTech in a way that was truly convincing. Amongst others,
one interesting aspect is that you can also do client to client
eventing. It could also encapsulate REX messages (or SSE messages,
whenever those become specified in a less half-baked manner).
The spec would prescribe that widget ID (as defined in config.xml)
be used in a URL to point to a widget, like: widget://localhost/
widget-id and that the widget engine would perform a "show"
operation for this widget. Nothing earth shattering, yet will
remove the most painful area with mobile widgets. I would even
venture to say - that without this capability the mainstream
adoption of mobile widgets is impossible.
I find this idea of the widget:\\ scheme very interesting. However,
we need to make sure we have explored what we currently specified
and make sure that the problem cannot be solved without specifying
something new.
Much, much agreed. Adding a new scheme should be avoided unless
really, really necessary.
This about it - the spec defines UNIQUE IDENTIFIER for a widget
(id element) and yet it is not a URL! Tell that to Tim Berners
Lee ...
I'm pretty sure he monitors the WAF work (he made a comment about
one of our specs a few weeks ago). If Tim had an issue he would
have raised it.
I wouldn't make that kind of assumption. Just because Tim had a
comment to make on something doesn't meant that he follows everything
and reads every email.
We've looked at various options internally (at Joost) and we've
reached agreement that using a URI (preferably http:// based) is the
best option, and a simple one too. It's also common and classic.
--
Robin Berjon - http://berjon.com/
------------------------------------------------------------------------
Pr. Farnsworth: Those delightful birds, with their chirp chirp chirp
and their tweet tweet splat.
-- Futurama
