W3C home > Mailing lists > Public > ietf-dav-versioning@w3.org > July to September 2001

Clarification of DAV:update-merge-set when rerunning MERGE

From: Roy Seto <Roy.Seto@oracle.com>
Date: Fri, 14 Sep 2001 15:25:34 -0700
Message-ID: <3BA283DE.13FDA2AD@oracle.com>
To: ietf-dav-versioning@w3.org
I'm looking at the behavior of the MERGE method when it is
rerun. I'd like clarification of the DAV:update-merge-set
postcondition in Section 11.2 for this scenario. 

Specifically, if a merge target was already checked out
before the second or later MERGE, and its DAV:merge-set or
DAV:auto-merge-set already includes the merge source, does
the merge target appear in a DAV:response XML element in the
response body? 

I feel this is ambiguous in draft-18. The
DAV:update-merge-set postcondition does apply to this merge
target, because it was already checked out before the MERGE,
and its DAV:checked-out version is not a descendant of the
merge source (assuming a client has not moved the merge
source from DAV:merge-set to DAV:predecessor-set between the
MERGE requests). However, the state of the merge target does
not actually change in the second MERGE request, since
adding a source to a DAV:merge-set which already contains it
doesn't change the DAV:merge-set property.

I prefer that the DAV:update-merge-set postcondition require
the response body to include the merge target in this case.
I think that makes it easier to implement clients that cache
minimal state and support users who wish to restart the
overall merge process where they left off.  Here is a use
case:

1. A user's client issues a MERGE request, including
   the DAV:merge-set and DAV:auto-merge-set 
   properties in the DAV:prop element of the request 
   body. The client remembers the resources listed
   in the response body which have nonempty 
   DAV:merge-set or DAV:auto-merge-set properties.

2. The user iterates through the conflict targets 
   identified in step 1, performing a logical merge of 
   their content and dead properties and moving merge 
   sources from DAV:merge-set or DAV:auto-merge-set to 
   DAV:predecessor-set as he or she modifies each 
   checked-out target resource.

3. Before the user is done with step 2, the client 
   crashes and its state is lost, or the user wishes
   to finish step 2 from a different client instance
   in a different physical location. At this point,
   the new client instance does not know the set of 
   targets which still have nonempty DAV:merge-set
   or DAV:auto-merge-set properties. If the client
   can rerun MERGE with the same request-URI and
   source, and be guaranteed to be given the set of 
   targets with nonempty DAV:merge-set or 
   DAV:auto-merge-set properties, this would be an 
   attractive way to let the user continue where
   he or she left off.

Alternative options I considered for supporting restart:

 - The client could do a PROPFIND Depth:infinity
   and look for resources which have nonempty 
   DAV:merge-set or DAV:auto-merge-set properties,
   but it seems difficult to make this efficient.
   It also requires the client to detect the restart
   case and implement it with a different code path.

 - The client could do a DASL query to optimize
   the PROPFIND above if DASL were standardized.

--

I have a similar question about the behavior of the
DAV:merge-preview REPORT after the corresponding MERGE has
already been done. If a merge target requires a merge, and
its DAV:merge-set or DAV:auto-merge-set already includes the
merge source, does the response body include it in a
DAV:conflict-preview element? (For reasons similar to the
DAV:update-merge-set case, I prefer that it must.)


Thanks,
Roy
Received on Friday, 14 September 2001 18:22:00 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 8 January 2008 13:57:42 GMT