Re: [Ietf-caldav] Last Call comment on Etag requirements in draft-dusseault-caldav-12

On Jun 20, 2006, at 9:56 AM, Julian Reschke wrote:

> Lisa Dusseault schrieb:
>> Xythos WFC and Chandler (the Zanshin library that does WebDAV in  
>> python) behave this way and make the assumption I describe.  How  
>> else would you expect a caching or synching client to behave after  
>> doing a PUT, when the implementors of those clients were pretty  
>> sure that WebDAV servers stored the content without mucking with it?
>
> Chandler is not released and obviously operating based on what the  
> CalDAV spec currently says.
>
> Regarding the Xythos client: I just did some tests, and as far as I  
> can tell the behavior is the same independantly of whether the  
> server returns an ETag in PUT: the client always assumes that  
> content was not rewritten, and in the absence of an ETag uses the  
> Last-Modified date to check. So it seems that it doesn't handle  
> content-rewriting servers at all, right? (one needs to manually  
> purge the cache to get the actual content).

My assertion was that if a strong ETag is returned, Xythos WFC  
assumes that what it PUT was what the server stored, and it seems you  
agree.  You found that if a Last-Modified is returned instead, WFC  
makes the same assumption -- naturally, they're very similar.

You're probably quite right about the general case, that existing  
WebDAV clients don't handle content-rewriting servers at all.  What's  
the best thing a content-rewriting server can do in this situation?   
I would hope that if a client receives neither an ETag nor a Last- 
Modified in a PUT response, then the next time it synchronizes and  
sees an ETag that it's never seen before, the client downloads the  
resource.  This allows the content to eventually get synchronized  
although perhaps not as fast as would be ideal.

But CalDAV clients will have to handle content-rewriting servers at  
least handling events (calendar component resources), because during  
protocol development we heard from a couple server developers that  
they'd need to add custom iCalendar properties to an event as soon as  
it was stored, thus rewriting the content.

lisa

Received on Tuesday, 20 June 2006 23:56:00 UTC