#1 : when reading from disk(and assuming DCP knows if it is reading from disk or memory), i would guess that we need special case it, and call it as Document Update(even when it is just a metadata update).
#2 : As of now - all meta.* and system-xattrs and user-axattrs. (but read onto #3)
#3 : good question. Essentially this is a corollary to #2. If we can differentiate between the 3(meta, user-xattr and system-xattr), it will be good. But a P0, would be just differentiate between meta(both meta.* and all types of xattrs) and body updates. (Tagging Matt Ingenthron for any thoughts on this)
Priya Rajagopal / Adam Fraser : can you add some more details w.r.t SG here.
The background to this MB is essentially that SG updates the xattrs(system) and this causes mutations on the DCP. For Eventing, this is a problem, since for non-idempotent operations in Eventing will lead to confusion as multiple updates are seen by Eventing. Eventing being a handle to consume the events, this requirement becomes critical.
Also, I believe that this request can be satisfied only for new events, and not for historic events.