Re: Revision names

Chris Kaler (ckaler@Exchange.Microsoft.com)
Tue, 12 Oct 1999 10:31:07 -0700 

Message-ID: <FD7A762E588AD211A7BC00805FFEA54B041DD982@HYDRANT>
From: "Chris Kaler (Exchange)" <ckaler@Exchange.Microsoft.com>
To: "'jamsden@us.ibm.com'" <jamsden@us.ibm.com>, ietf-dav-versioning@w3.org
Date: Tue, 12 Oct 1999 10:31:07 -0700
Subject: RE: Revision names


I think these are somewhat separate discussions:
1) Do we separate label namespace from id namespace
2) Do we allow labels and/or ids in headers
3) Can labels and/or ids be combined to construct URLs

My opinion on these...

1) I would like these to be separate namespaces.  My
   argument has been that the server needs to be able
   to determine if X is an 'id' or a 'label'.  We need
   a way to differentiate them -- context is fine if
   it works.

2) Headers are a good thing.  I should be able to alter
   the behavior of GET by including a header that indicates
   which version I want.  This has a lot of advantages in
   that I can separate URL from version (revision).  Also,
   I can have separate headers, or special syntax in the
   header to address #1.

3) As stated below, I'm against requiring this, unless
   the server has said that they support property collections
   in which case the server has signed up to some additional
   requirements.

Just my two cents... :-)

-----Original Message-----
From: jamsden@us.ibm.com [mailto:jamsden@us.ibm.com]
Sent: Tuesday, October 12, 1999 10:20 AM
To: ietf-dav-versioning@w3.org
Subject: RE: Revision names




If the revision name (id or label) is provided in a separate header, then
would
we still need to construct these URLs? I think we should avoid this if we
can.





"Chris Kaler (Exchange)" <ckaler@Exchange.Microsoft.com> on 10/12/99
01:00:35 PM

To:   Jim Amsden/Raleigh/IBM@IBMUS, ietf-dav-versioning@w3.org
cc:

Subject:  RE: Revision names



I am strongly opposed to the specification describing what a
revision id must look like.  I know Jim only said that they
should be valid path segments, but, in retrospect, I think
that is too severe.  If my sever wants to use a URI to refer
to a revision, then I should be allowed to.  However, this
may not be a "valid" URL segment.

As well, the specification shouldn't require the ability for
a client to be able to "glue" these with other information and
expect to have a URL that works. Now if a server supports
property resource collections, then they better conform to this.
But a server that doesn't, shouldn't have to.

This approach will limit servers from providing ids that are
optimized to their store's.

Chris

-----Original Message-----
From: jamsden@us.ibm.com [mailto:jamsden@us.ibm.com]
Sent: Tuesday, October 12, 1999 7:53 AM
To: ietf-dav-versioning@w3.org
Subject: Re: Revision names




We can think of the server as another collaborator in distributed authoring
systems, one that provides a set of services. In particular, there can be
little
distinction between a revision name (something that distinguishes one
revision
from another in this context) specified by some other client and one
specified
by the server. In both cases there is the possibility for collisions, and in
both cases, there is the desire to use the revision name to select a
revision.
As Tim points out below, there is also a desire for the syntax of label
names to
be consistent with URLs, and to be able to marshall revision names in
request
and response entity bodies (in XML) as well as in headers. The only
difference I
can see is that the server's revision name, the revision id, can't be moved
or
reused - its an immutable or fixed label that ensures revisions can always
be
distinguished. Any attempt to move or reuse the label id results in an
error.
This makes potential client/server collisions even safer than client/client
collisions as there is no possibility of some other client getting an
unexpected
revision because some other client or the server moved the label id.

The only reason to separate id and label name spaces seemed to be to avoid
client/server label name collisions. But it is clear that client/client name
collisions are much more likely to happen, and have greater consequences (as
measured by the principle of least astonishment). I continue to find it hard
to
justify the complexity separate label namespaces adds to the protocol vs.
the
problems it solves. Does anyone else see other issues that separate
namespaces
would avoid?





Tim_Ellison@oti.com (Tim Ellison OTT) on 10/12/99 10:26:17 AM

To:   ietf-dav-versioning@w3.org (ietf-dav-versioning)
cc:

Subject:  Revision names




As mentioned by both Jeff (by phone) and Geoff (in an earlier positing),
revision id's must be legal URI path segments if we envisage the ability to
refer to a revision by a URL (i.e. DAV:history's revisions collection "/"
DAV:revision-id).

Maybe we will also want to refer to a particular labelled resource by a URL
in a similar fashion.

If we choose to differentiate labels and revision id's by extra syntax
surrounding the value this would lead to bizzare looking URLs.


Having listened to the discussions, I think that the argument for avoiding
collisions between labels & revision ids has been largely debunked; and the
protocol would undoubtably be simpler if there was not requirement to
separate namespaces.  However, labels and revision ids have different
characteristics from the client's perspective and it would be immensely
reassuring to know which you are dealing with at all times.  I just don't
see yet how this would fit into the protocol.

Tim