On 2013-02-19 12:06, Dave Cridland wrote:
On 19 Feb 2013 08:47, "Ashley Ward" <ashley.w...@surevine.com
<mailto:ashley.w...@surevine.com>> wrote:
>
>
>
> On 18/02/2013 18:19, "Ralph Meijer" <ral...@ik.nu
<mailto:ral...@ik.nu>> wrote:
>
> >On 2013-02-18 17:34, Kevin Smith wrote:
> >>
> >> Thoughts?
> >
> >Maybe this option is not very well suited for the case where there are
> >potentially multiple publishers, as one of them going offline would
> >result in the retraction of items published by others, too.
> >
> >I agree this should be described better.
>
> Maybe it would make more sense if purge_offline applied to the node owner
> rather than the publisher?
>
Except that you can have multiple node owners, too.
So seeing as I can think of use cases where multiple publishers and
partial purges happen, I think we should be sending retractions for
partial purges and purge notifications for full purges - both subject to
the notify_retract option, of course.
Agreed.
I'd like to reiterate that this particular feature is probably useful
only for some very specific use cases. The changelog that mentions the
feature being added mentions 'extended presence'. This more or less
implies PEP nodes with a single publisher/owner and a tight availability
presence coupling.
Having "partial purges" might be quite an expensive operation with many
persisted items and I don't really feel like addressing that at this
point (e.g. with new protocol).
Even though I know that at least ejabberd's pubsub support has this
feature, I am not aware of any applications using it.
We can either clarify the intended use while mentioning some potential
issues mentioned in this thread, or deprecate the feature altogether.
--
ralphm
