Re: Integrating with the event loop

On Wed, 22 Sep 2010 17:47:18 +0200, Robin Berjon <robin@berjon.com> wrote:
> this is, again, a message from the DAP WG.
>
> We've been looking at tightening up loose ends in our more mature  
> specifications so as to bring them to last call hopefully soon. One  
> thing that's been dangling on my plate for a while is integration with  
> the event loop (task queues and all that).
>
> On the one hand I understand how the model is a useful abstraction, and  
> I see a number of places where it may clarify things. On the other hand,  
> I note that some specifications that don't use it — Geolocation comes to  
> mind — don't appear to have specific interoperability issues (in this  
> respect at least).
>
> What I can't seem to figure out is what is the criterion one uses to  
> decide whether describing one's process in event loop terms is  
> useful/necessary, and when it's optional. My first reaction was  
> asynchronicity, but that doesn't seem to be sufficient.
>
> Any guidance on this topic will be welcome,

I think the Geolocation API ought to have integrated with it as well. It  
probably works because it is fairly predictable how it ought to integrate,  
but that does not make it well defined. For instance  
http://www.w3.org/TR/geolocation-API/#clear-watch makes it seem as if  
currently queued tasks still ought to be processed if they are in the  
queue, but I somewhat doubt that is the intent or matches what is  
implemented.


-- 
Anne van Kesteren
http://annevankesteren.nl/

Received on Wednesday, 22 September 2010 19:41:07 UTC