RE: Internal and External Members

I agree with the comments that the current external member mechanism is
broken.  I also agree with Judith's summary and her conclusion that some
sort of "internal name resolves to external member" naming mechanism is the
basis for the proper solution.

At the same time, however, I feel that one can build a very useful DAV
server without support for external members. Additionally, based on mail on
the list, I believe consensus exists that one can also build a very useful
DAV server without support for ordered collections. This indicates to me
that these features are optional value-adds and thus can be safely placed in
their own specification. 

Additionally by putting these features in a separate specification we can
move on with the Distributed Authoring last call and thus also begin the
work of creating the Versioning, ACL, and schema specifications, all of
which are completely dependent upon the Distributed Authoring specification.
In fact the reason why the versioning features were separated out is that we
found that any change to the distributed authoring features necessitated a
change to the versioning system. Rather than play a hopeless game of "patch
the problem" we decided to get the distributed authoring features
standardized and then use that foundation to build versioning, ACLs, etc. I
believe the same logic applies to both external members and ordered
collections.

I realize that there is some hesitation in having a separate spec for
ordered collections and external members but I would like to assure everyone
that a separate spec in no way deprecates these features.

Just as the Distributed Authoring protocol specification is built on top of
HTTP/1.1, and the versioning functionality of WebDAV is built on top of the
Distributed Authoring protocol specification, so features like ordered
collections and external members can be built on top of DAV.

As such I would like to propose the creation of a new specification to
define both external members and ordered collection functionality. I offer
myself as a co-author if it is felt that my services are needed. I am
interested in co-authoring because I know that products such as Microsoft
Site Server will need both ordered collection and external member
functionality. I also invite Jim Davis, Ellis Cohen, and Judith Slein to be
co-authors as they have repeatedly demonstrated a solid grasp of the issues
and a strong motivation to see them resolved.

If this proposal is accepted then ADDREF, DELREF, and the externalmembers
property would be removed from the Distributed Authoring specification. No
further changes would be needed as external members will have internal names
and thus appear completely normal to Distributed Authoring compliant
clients.

				Yaron


> -----Original Message-----
> From:	Judith Slein [SMTP:slein@wrc.xerox.com]
> Sent:	Tuesday, January 27, 1998 10:21 AM
> To:	ejw@ics.uci.edu; w3c-dist-auth@w3.org
> Subject:	Internal and External Members
> 
> I've been spending some time thinking about how to implement external
> members of a collection so that expected behaviors just naturally happen.
> 
> For me, the point of having external members is to allow multiple
> collections to share the same resource.  It's to save each collection from
> having to keep a physical copy of the same resource, and to insure that
> any
> change in the resource will automatically be reflected in any collection
> that has it as a member.
> 
> ---------------
> 
> We're looking for an implementation of external members and internal
> members that yield similar semantics for both.  Either they should both be
> implemented as properties on the collection, or they should both be
> implemented as URIs relative to the collection (as Dylan Barrell suggested
> a few days ago).  Let's look at some scenarios for each implementation.
> 
> PROPFIND
> PROPPATCH
> COPY
> LOCK
> 
> --------------
> 
> 1. Treat both internal and external members as properties of the
> collection.
> 
> A Depth=0 COPY will copy a collection and the properties that list its
> internal and external children.  The intuitive meaning of Depth=0 is that
> only the collection, none of its children, should be copied.  So the spec
> would have to make some special, explicit requirements about what happens
> when you COPY Depth=0 a collection: the internalmembers and
> externalmembers
> properties must be empty on the new collection.
> 
> A Depth=infinity COPY copies both the internalmembers and the
> externalmembers property for the collection and for each internal member
> that is a collection.  It copies the resources referenced by the
> internalmembers property, but not those referenced by the externalmembers
> property.
> 
> A Depth=0 LOCK should mean that you cannot add / remove members from the
> collection.  Since a LOCK on an object also locks its properties, this
> should have the expected effect.
> 
> A Depth=infinity LOCK locks the collection resource itself and all
> resources referenced by the internalmembers property, recursively.
> 
> PROPFIND should list all members of a collection, both internal and
> external, with the requested properties for each member.  PROPFIND would
> examine both the internalmembers and the externalmembers properties, and
> for each URI on either list, it would retrieve the requested properties.
> 
> PROPPATCH.  Presumably once you find the URI for either an internal or an
> external member, you can do a PROPPATCH on that URI to set a property on
> the member.  There is no way to set a property of a member-in-a-collection
> (a property that applies to the resource only as a member of that
> collection, e.g., who added it to the collection and when).
> 
> ------------------
> 
> 2. Treat both internal and external members as URIs relative to the
> collection.  That is, when an external member is added to a collection, a
> new resource is created relative to the collection.  (Its URI could be one
> requested by the client or one determined by the server.)  This resource
> has a membertype property = external and a location property = the URI
> where its content is located, and no content.  (Or, as Dylan suggested,
> the
> URI that points to the content location could itself be the content of the
> member resource rather than a property on it.)  Note that this approach in
> effect results in all members, even external members, having names (URIs)
> in the collection, a point that has been contentious in the past.  It does
> allow external members to carry a set of properties separate from the
> properties on the resource they point to, a capability Mary Hofstede
> argued
> for recently.
> 
> A Depth=0 COPY will copy a collection itself, but none of its members,
> either internal or external.  This is the behavior that we expect for
> Depth=0.
> 
> A Depth=infinity COPY will copy all the resources that are members of the
> collection, whether internal or external, recursively.  For the external
> members, this does not involve following the URI to the content location.
> It just means copying the resource that is the member of this collection.
> If an external member points to a collection, nothing special happens.
> It's still only the resource in this collection that gets copied.  (What
> is
> the resourcetype of an external member that points to a collection?)
> 
> A Depth=0 LOCK should mean that you cannot add / remove members from the
> collection.  However this is expected to be implemented today for internal
> members, the same approach would be taken for external members.
> 
> A Depth=infinity LOCK locks the collection itself and all of its member
> resources, recursively.  For external members, it locks only the URI
> relative to the collection (so that none of its properties or content can
> be changed), but does not lock the resource pointed to by its location
> property (or by its content on Dylan's implementation).
> 
> PROPFIND should list all members of a collection, both internal and
> external, with the requested properties for each member.  For external
> members, do we display only properties on the URI relative to the
> collection, or do we follow the URI in the location property and display
> its properties?  I think the latter.  For a Depth>0 PROPFIND where an
> external member points to a collection, we do expect to see the members of
> that collection.  (What is the value of resourcetype for an external
> member
> that points to a collection?)
> 
> PROPPATCH.  Presumably once you find the URI for either an internal or an
> external member, you can do a PROPPATCH on that URI to set a property on
> the member.  Now external members do have a URI within the collection as
> well as a URI for the content location, so clients can set properties on
> either URI.  The external member's URI within the collection can carry
> such
> properties as who added it to the collection and when, while the URI it
> points to can carry such properties as the resource's author, title,
> content size, etc.
> 
> ------------
> 
> I am inclined to think that we get desired behavior more naturally from
> implementing each member, whether internal or external, as a resource in
> the collection.  This works well for COPY, LOCK, and supports a separate
> set of properties on an external-member-in-a-collection from the
> properties
> that reside on the resource the external member points to.  PROPFIND needs
> more thought, since it's no longer clear whether the client wants to see
> the properties on the external-member-in-a-collection of the properties on
> the resource the external member points to.
> 
> Note that DMA has chosen to implement both types of membership as
> properties on a collection.  For those who are interested, here are some
> excerpts from the DMA 1.0 specification.
> 
> --------------
> 
> DMA distinguishes between Direct and Referential containment:
> 
> "Direct containment. This models a 1:N, or one-to-many, relationship. A
> containing object may contain multiple objects, but an object is directly
> contained within at most one containing object. A containing object that
> directly contains another object is the parent of that object. An object
> that is directly contained within a parent object is a child of that
> parent. Multiple directly contained objects are children. Direct
> containment defines a strict hierarchy of objects with no cycles. Document
> spaces implementing direct containment must prevent cycles. For example,
> moving a parent into one of its children must be disallowed. DMA does not
> limit the depth of the direct containment hierarchy, but a document space
> may choose to impose a limit." 
> 
> "Referential containment. This models an N:M, or many-to-many,
> relationship. Objects that are referentially contained within a containing
> object are referred to as containees. A containing object that
> referentially contains another object is referred to as a container of
> that
> object. A container may contain multiple containees. A containee may be
> contained within multiple containers. Cycles in referential containment
> relationships are not disallowed by DMA, because they could be expensive
> to
> prevent. A document space may choose to disallow them, however, in which
> case it is obligated to enforce this. DMA defines operations that modify
> containment hierarchies in such a fashion that cyclical referential
> containment can be supported."
> 
> The point of having referential containment is to let multiple containers
> share some of the same content.  This understanding of referential
> containment also allows the same object to belong more than once to a
> single container, and allows the same object to belong to the same
> container once directly and one or more times by reference.
> 
> DMA also has the object infrastructure to allow properties to attach
> either
> to the member itself or to the member/container relationship.
> 
> DMA implements both direct and referential containment using properties on
> the container, one enumerating the container's childrem (direct members)
> and the other enumerating the container's containees (members by
> reference).
> 
> --Judy
> 
> 
> 
> Name:		Judith A. Slein
> E-Mail:		slein@wrc.xerox.com
> Phone:  	(716) 422-5169
> Fax:		(716) 422-2938
> 
> Xerox Corporation
> Mail Stop 105-50C
> 800 Phillips Road
> Webster, NY 14580

Received on Tuesday, 27 January 1998 18:41:46 UTC