RE: [xri] Backtracking

["Drummond Reed" <drummond.reed@cordance.net>]

I talked to John not long ago, and I believe we have closure on this issue
he raised. Mea culpa, as it appears text I added to section 11.5 of ED06
(the Recursion and Backtracking subsection of Redirect and Ref Processing)
is the source of the problem. It says:

*************
.	When a resolver first encounters a Redirect or a Ref, this is called
the First Recursion. As specified in sections 11.2 and 11.3, the resolver is
obligated to resolve the highest priority Redirect or Ref to see if it can
satisfy the resolution query. If EITHER resolution of the Redirect or Ref
fails OR if resolution is successful but the resolver is ultimately unable
to continue or complete the orginal resolution query, the resolver MUST try
the next highest priority Redirect or Ref. The resolver MUST continue until
either it is successful or all Redirects or Refs have failed.
*************

That text makes it sound like backtracking can back "all the way up" from
the final SEP selection phase back into the authority resolution phase.

That is NOT CORRECT. Thankfully the flowcharts in ED06 appear to be correct
about this, but the above text definitely needs to be revised to the
following clear:

* Authority resolution and final SEP selection are two separate phases of
resolution. (The fact that the authority resolution phase uses a "dedicated"
form of SEP selection during it can be confusing, but the dedicated SEP
selection that takes place during authority resolution is still 100% part of
the authority resolution phase).

* Backtracking during Ref processing is limited to THE PHASE IN WHICH THE
REF IS ENCOUNTERED. In other words, if you follow a Ref during authority
resolution, and you are not able to complete authority resolution (you can't
find an authority resolution SEP), you backtrack as needed to complete
authority resolution. ONCE YOU HAVE COMPLETED authority resolution
(including any XRD-level Ref in the final XRD), ONLY THEN do you move into
the final SEP selection phase. Once in that phase, you again follow Refs as
needed and backtrack as needed to find the final SEP, but you NEVER
BACKTRACK BACK UP into the authority resolution phase.

Once again, many minds make for a good spec. I appreciate John pressing on
clarification on this issue even though some of us weren't clear what the
problem was. And I apologize that I was the source of the conflicting text.

This returns us to ZERO OPEN ISSUES (and let's hope we stay there). The job
now is to get all this -- and Gabe's RelaxNG schemas -- rolled into ED07 by
Monday night. If you encounter ANYTHING ELSE that you think needs revising
in the spec, please email the list ASAP. I'll try and check it a few times
over the weekend, but definitely Monday morning.

=Drummond (going offline now to a high school football team reunion)