FW: RE: Baby steps

Caught by the spam filter -- the email address
BCragun.ORM2-1.OREM2@gw.novell.com has been added to the accept2 list for
the list, so future messages from this address shouldn't be bounced.

- Jim

-----Original Message-----
From: Bruce Cragun [mailto:BCragun.ORM2-1.OREM2@gw.novell.com]
Sent: Thursday, August 06, 1998 7:21 AM
To: ejw@ics.uci.edu; rubys@us.ibm.com; w3c-dist-auth@w3.org
Subject: [Spam?] RE: Baby steps


This explanation of VPortals helped me a lot in understanding their purpose.
But it returns me to a question I asked in Redmond: What is the method for
getting an earlier (i.e. not the tip) version of a document?  And does
WebDAV truly support not only getting a non-tip version but also the ability
to return that version, with changes, to the same location?  Stated another
way, can I edit a non-tip version and return it to the collection as that
same version, or does WebDAV force a new version for any change?

This issue is critical to us at Novell, so I just want to make sure nobody
believes we have placed such limitations into this spec.

>>> Jim Whitehead <ejw@ics.uci.edu> 08/05 1:56 PM >>>

> Suppose I am in my favorite Integrated Development Environment
> (IDE) and want to checkout the latest version of x.c.  Nothing
> fancy, no branches.
>
> Everything I have read so far seems to start at "once you have found the
> Vportal, here are the set of operations you can perform"
>
> ...
>
> I believe I can follow the chain from portal to vgraph to cache, but
> can't quite make the leap from the conceptual resource to the portal.

The simple answer is you can discover the Vportals by performing a PROPFIND
on its containing collection to retrieve the members of the collection.  The
PROPFIND will list the Vportals, along with all other resources (e.g.,
collections, regular resources, etc.) that are contained in the collection.
This is the typical WebDAV mechanism for retrieving the members of a
collection.

Examples are always helpful to fully understand versioning, so here is an
example which will hopefully shed some light on this issue.  The example
will start off assuming that no versioning is taking place (i.e., show how
things would be organized and accomplished using existing DAV and HTTP
features), then add versioning into the picture.

So, let's imagine that your "x.c" is part of project "my-utility", and
belongs to an HTTP URL hierarchy that looks like this:

              my-utility
                   |
  -----------------+-------------------
  |         |           |             |
x.c      makefile   my-util.h        y.c

If this URL hierarchy were located on the server "devel.swcorp.com", then
there would be the following URLs:

http://devel.swcorp.com/my-utility/           The collection holding the
resources
http://devel.swcorp.com/my-utility/x.c        The C source resource "x.c"
http://devel.swcorp.com/my-utility/makefile   The makefile for the project
http://devel.swcorp.com/my-utility/my-util.h  The include resource for the
project
http://devel.swcorp.com/my-utility/y.c        The C source resource "y.c"

Using DAV PROPFIND with Depth "infinity" on
http://devel.swcorp.com/my-utility/, a complete list of the members of the
collection would be returned, along with the properties requested in the
PROPFIND.

OK, now versioning is going to be added into the picture.  Assume that
"x.c", "makefile", "my-util.h", and "y.c" are versioned (ignoring
configuration management issues for the moment, and hence leaving the
my-utility collection unversioned), and has a history of several previous
revisions.

>From previous discussions on this list, and working group meeting
discussions, there is broad agreement that each revision of a resource is
itself a separate resource, with its own URL.  This approach has many nice
qualities -- links can be constructed to previous revisions of a resource,
and they integrate well with the existing cache infrastructure.

But, this creates a naming problem.  Ideally, when working on a project, you
want to have a structural view of it -- in this example, the project has
been hierarchically structured by placing the program files (x.c, y.c,
makefile, my-util.h) into a collection (my-utility).  With the addition of
versioning, each source code resource in the project now has multiple
resources associated with it -- each conceptual source code resource is now
really a set of resources.  For example, "x.c" is now really a conceptual
handle for the resources "x.c version 1", "x.c version 2", "x.c version 3",
"x.c version 4", etc.  When working on the program, the programmer thinks of
all these resources simply as "x.c", but for the Web server, there needs to
be a separate URL for each of these resources.

Since the HTTP URL space only has a single form of organization,
hierarchical, all of the versions of a resource, each with its own URL, need
to be placed somewhere in an HTTP URL hierarchy.  There are several choices
here.

One choice is to "URL munge", essentially hiding all but the tip (or default
published) version of a resource.  In this scheme, you could add a
",v{version id}" to the end of a URL to retrieve a specific version.
Performing a retrieval without adding ",v{version id}" would return the tip
or default member.  This approach is bad because it makes URLs non-opaque,
and raises the risk of namespace collisions with other mappings of ",v" in a
URL.  As a result, this approach has time and again been rejected by this
working group.

Another choice is to place all of the versions of a resource into the same
collection as the original resource.  In this example, a PROPFIND of
"my-utility" would list "x.c version 1", "x.c version 2", "x.c version 3",
"y.c version 1", "y.c version 2", etc.  This approach is bad because it
mixes together the structure of a project and the versions of project
members into the same collection.  It would also cause the collection
listing to become quite large over time.

It seemed to me, looking at this problem, that what was needed was an
abstraction which would allow a programmer to work on their projects as if
they were unversioned, i.e., if the programmer worked on an unversioned
source code resource called "x.c", then they should still be able to work on
"x.c" even when it was versioned.  To avoid the URL munge approach, and to
avoid mixing together the structure view with the versions of resources, it
would be necessary to stash the version resources in a location in the URL
hierarchy which was different from the "structure" location. The Vportal is
that abstraction.

A Vportal provides the bridge between the structural organization of a
project, and the location in the URL namespace where every individual
version of a resource is located (i.e. the version space, making the Vportal
a link between the structure space and the version space).

So, how would the my-utility project look if it were versioned using
Vportals?  The structural view would still be the same -- a PROPFIND on the
collection http://devel.swcorp.com/my-utility/ would still return the
following resources:


http://devel.swcorp.com/my-utility/x.c        Vportal for the C source
resource "x.c"
http://devel.swcorp.com/my-utility/makefile   Vportal for the makefile for
the project
http://devel.swcorp.com/my-utility/my-util.h  Vportal for the include
resource for the project
http://devel.swcorp.com/my-utility/y.c        Vportal for the C source
resource "y.c"


So, when versioning a project, each resource is replaced with a Vportal,
which is a pointer into the version graph for that resource. Returning to
the original question, "how would you discover a Vportal", the answer is
"use PROPFIND".

The Vportals also imply that there is a stash of versioned resources
someplace in the URL hierarchy.  For this example, these versions are
located in collections under "/version-cache/", as follows.

The revisions of "x.c" are located in the collection
"/version-cache/my-utility/x.c/".  So, version 1 would have the full URL:

http://devel.swcorp.com/version-cache/my-utility/x.c/x.c-vers1

There is nothing magic about this naming convention -- I do not expect it to
be standardized, and I do expect it to vary from server to server.  This is
OK, and does not lead to loss of interoperability since the version graph
contains a complete list of the members of the graph, their is-derived-from
relationships, and their complete URLs.

When editing a versioned resource, the Vportal resource receives CHECKOUT
and CHECKIN commands.  So, to begin editing "x.c", a CHECKOUT would be sent
to "http://devel.swcorp.com/my-utility/x.c".  When editing was finished, a
CHECKIN would be sent to the same URL.  From the perspective of the person
editing versioned "x.c", the Vportal "x.c" acts just like unversioned "x.c".

Well, this probably answered your question in more detail than you were
expecting :-)

- Jim

Received on Thursday, 6 August 1998 12:38:50 UTC