A. Pagaltzis wrote:
By and large, I agree with him that it would have been beneficial to specify Atom just a little more closely at a model level. But I also agree with you that aspiring to much higher rigor beyond the Infoset level is unnecessary. My retrospective opinion is that the correct approach would have been to specify that an Atom Feed Document consists of a series of completely independent Atom Entries, each of which can be interpreted in isolation of any others as well as of the Atom Feed Document that they are found in (modulo Person Construct inheritance). This would explicitly allow consumers to rely on this atomicity (pun intended), thus preventing any extensions from crossing these boundaries. This goes beyond the mere interchange of messages to a definition of a standardised model, though as you can see, it’s a very loose one. I contend it is also the model used implicitly by any useful aggregator which tracks feed content over time.
I think this group would not have been able to agree on what "a little more" was, the current situation makes sense to me. IMO, the Infoset is no basis for a model; it's there for spec writers and marshalling off the wire.
cheers Bill