(A discussion group for Salmon is on my to-do list -- will bump priority
up.)

In the case of PubSubHubbub, if you assume fidelity of GUIDs (atom:id) then
updates become nearly trivial -- you key off of the GUID and replace the old
version with the new version (or append a revision, or whatever strikes your
fancy).  So yes, an update with an empty content, title, and summary would
be almost like a delete -- but that would also be an invalid atom:entry :(.
 I believe issues like this are what drove the deleted-entry mechanism.

--
John Panzer / Google
[email protected] / abstractioneer.org <http://www.abstractioneer.org/> /
@jpanzer



On Sat, Oct 24, 2009 at 7:59 AM, igrigorik <[email protected]> wrote:

>
> +1 for a Google group for Salmon.
>
> How about handling updates? At least intuitively, a delete is a
> special case of update -- what if someone updates their comment, how
> do we propagate that?
>
> As far as existing examples go, FriendFeed is one of the few
> publishers that I'm aware of that actually distinguishes between new
> posts and updates:
> http://friendfeed.com/api/documentation#realtime
>
> Perhaps we could make the delete request as an update with empty body?
>
> ig
>
> On Oct 23, 6:20 pm, John Panzer <[email protected]> wrote:
> > My proposal would be:
> > 1. Support the existing Atom Tombstones spec extension (
> http://ietfreport.isoc.org/all-ids/draft-snell-atompub-tombstones-06.txt)
> as
> > the 'official' way to indicate that a source has deleted something, if it
> > cares to tell downstream subscribers;
> >
> > 2. Add language that explains how the hub deals with the deleted-entry
> > brethren to real entries -- essentially, I think this should say that
> > they're treated just like atom:entry for purposes of deltas, atom:source,
> > etc. etc. so they get passed through.
> >
> > I think that's it.
> >
> > --
> > John Panzer / Google
> > [email protected] / abstractioneer.org <http://www.abstractioneer.org/>
> /
> > @jpanzer
> >
> > On Fri, Oct 23, 2009 at 2:59 PM, Jeff Lindsay <[email protected]>
> wrote:
> > > It's not. I'm wondering what kind of behavior they expect from this
> kind of
> > > feature in the protocol. If it's just a way to notify subscribers a
> feed
> > > item was deleted and they deal with that however they like, that seems
> like
> > > it would be much easier to get into the spec. Are they any AtomPub-like
> > > semantics for this sort of thing?
> >
> > > My experience with this project so far is that if you want to get
> something
> > > like this in the spec and there is general consensus it's a good idea,
> you
> > > can write up a proposal of what can go in the spec and then Brett
> and/or
> > > Brad will work on integrating it.
> >
> > > -jeff
> >
> > > On Fri, Oct 23, 2009 at 2:52 PM, Jeff Eddings <[email protected]
> >wrote:
> >
> > >> In talking with Flickr today, deleting items from a feed seems to be a
> > >> must-have for them for implementing PuSH.  It sounds like it isn't in
> the
> > >> spec (yet).  Being more of a interested observer rather than a
> dedicated
> > >> contributor, how do we get such a feature into the spec?
> > >> Thanks,
> >
> > >> Jeff
> >
> > >> On Fri, Oct 23, 2009 at 12:40 PM, John Panzer <[email protected]>
> wrote:
> >
> > >>> Yes, that makes sense -- most of these issues are generic.
> > >>> One other one which I just remembered:  How to deal with deletion of
> > >>> entries.  Specifically, I want to be able to push a deletion
> upstream, and
> > >>> then broadcast it out, the same way that a new or changed comment is
> pushed.
> > >>>  In practice this probably means adopting one of the tombstone specs,
> e.g.,
> > >>>
> http://ietfreport.isoc.org/all-ids/draft-snell-atompub-tombstones-06.txt.
> > >>>  This in turn would require that the hub understand some semantics
> about the
> > >>> deleted-entry element (a child of feed), as opposed to the entry
> element,
> > >>> which is all the current spec talks about.  Thoughts?
> >
> > >>> (This is actually fairly important for spam control in the case of
> Salmon
> > >>> -- if a comment is published and pushed via PubSubHubbub, but
> subsequently
> > >>> discovered to be spam, we need a claw-back mechanism.  It doesn't
> have to be
> > >>> honored by downstream subscribers of course, but they should have the
> > >>> opportunity to be informed.  It's not always possible to filter spam
> before
> > >>> publishing, especially when you're trying to be real time.)
> >
> > >>> --
> > >>> John Panzer / Google
> > >>> [email protected] / abstractioneer.org <
> http://www.abstractioneer.org/>/ @jpanzer
> >
> > >>> On Fri, Oct 23, 2009 at 12:18 PM, Pádraic Brady <
> [email protected]
> > >>> > wrote:
> >
> > >>>> I started watching the Salmon Protocol (you guys need a mailing
> list!) a
> > >>>> while back when I noticed a few posts about it on Twitter.
> >
> > >>>> In relation to the issues you raised, subscriptions can be given a
> > >>>> limited time to live, but the issue is in who defines the TTL - some
> Hubs
> > >>>> may assume unlimited TTLs for all subscriptions so it can become
> important
> > >>>> to make sure Subscribers are passing a realistic lease_seconds
> value. I'm
> > >>>> not sure that Hubs can manage this any other way, other than to
> periodically
> > >>>> purge inactive feeds through some garbage collection tasks - such a
> task is
> > >>>> undefined in the specification but maybe automatic re-subscriptions
> will
> > >>>> ease in the behaviour most suitable for comment feeds. It still
> needs
> > >>>> Subscriber support for the most part in being intelligent about
> their lease
> > >>>> period.
> >
> > >>>> It's really a very messy area. Can Hubs filter subscriptions?
> Eliminate
> > >>>> subscriptions to feeds without updates within X months, etc? As
> someone on
> > >>>> another topic noted, Subscribers do need to pay attention to their
> > >>>> subscriptions and manage them. I think this is a good practice so
> Hubs are
> > >>>> not unduly burdened with tracking subscriptions to dead or inactive
> feeds,
> > >>>> or feeds with a limited useful lifetime (in terms of activity not
> content).
> >
> > >>>> On Hubs changing the source feed - most probably won't. It costs
> little
> > >>>> to maintain an entry as-is for distribution. Some Hubs may basically
> > >>>> condense the feed changes to a narrow spectrum of elements -
> removing the
> > >>>> rest. It's not really specified so it's up to Publishers to select
> > >>>> appropriate Hubs. You also can't rely on Subscribers tracking the
> different
> > >>>> behaviour of individual Hubs. The PHP Hub I'm writing at the moment
> won't
> > >>>> perform any filtering since doing so potentially robs feeds of
> information
> > >>>> that looks unnecessary in one use case, but becomes essential in
> another
> > >>>> (e.g. Salmon).
> >
> > >>>> Paddy
> >
> > >>>> Pádraic Brady
> >
> > >>>>http://blog.astrumfutura.com
> > >>>>http://www.survivethedeepend.com
> > >>>> OpenID Europe Foundation Irish Representative<
> http://www.openideurope.eu/>
> >
> > >>>> ------------------------------
> > >>>> *From:* John Panzer <[email protected]>
> > >>>> *To:* [email protected]
> > >>>> *Sent:* Fri, October 23, 2009 7:30:34 PM
> > >>>> *Subject:* [pubsubhubbub] Salmon Protocol
> >
> > >>>> The Salmon Protocol (http://salmon-protocol.org) leverages
> PubSubHubbub
> > >>>> and web hooks to build a real time, decentralized commenting and
> annotation
> > >>>> system.  The basic idea is that commentary swims upstream to the
> thing being
> > >>>> commented on, which can then redistribute comments back out via
> PubSubHubbub
> > >>>> to interested subscribers.
> > >>>> There's a demo available athttp://
> salmon-playground.appspot.com/roswhichposts comments back to a Blogger
> blog (by proxying to an existing API,
> > >>>> just for demo purposes).
> >
> > >>>> Aside from evangelizing Salmon, I'm also interested in getting
> feedback
> > >>>> on the use of PubSubHubbub.  One specific question:  At the moment,
> Salmon
> > >>>> specifies that subscribers should follow the rel="comment" link for
> each
> > >>>> entry in an Atom feed to find the comment feed to subscribe to.
>  This could
> > >>>> lead to a lot of subscriptions that have initial activity but then
> die out;
> > >>>> not sure if this is a problem for PubSubHubbub or not.  It might be
> better
> > >>>> to define a feed for, and there fore a way to subscribe to, "all
> comments
> > >>>> for items in this feed" as a whole -- but there's no standard way to
> do this
> > >>>> at the moment.
> >
> > >>>> Also, Salmon definitely needs some extensions to get passed through
> > >>>> un-modified (like crosspost:source, thr:in-reply-to, etc.) so the
> discussion
> > >>>> about what hubs may/may not change is very relevant.
> >
> > >>>> Comments, critiques, feedback of all kinds welcomed.
> >
> > >>>> --
> > >>>> John Panzer / Google
> > >>>> [email protected] / abstractioneer.org<
> http://www.abstractioneer.org/>/ @jpanzer
> >
> > > --
> > > Jeff Lindsay
> > >http://webhooks.org-- Make the web more programmable
> > >http://shdh.org-- A party for hackers and thinkers
> > >http://tigdb.com-- Discover indie games
> > >http://progrium.com-- More interesting things
>

Reply via email to