Afternoon (1:35 PM)

AOLpress and AOLserver - Long

Long started by informing everyone that GNNpress and GNNserver are now known as AOLpress and AOLserver.

In AOLserver, the current version of resource is stored in the file system, while the meta-data, indices, etc. are stored in a database.

Long noted that people think of their web sites in terms of a hierarchy, and hence it is disorienting for the server to place a resource at a location other than the one specified by the user.

With regard to server-side includes and other dynamic content, AOL has no strong solution for getting to the actual source. Their current approach disallows editing a page with dynamic content. However, AOLpress users seem to be doing it anyway, although it is not clear what mechanism they use. His recommendation for how to get a document's source before server side includes processing is to use content negotiation when requesting the source, and asking for MIME type text/x-html-ssi. Unfortunately, this approach does not handle the case where the source is not HTML.

A discussion then ensued about pretty printing (canonicalizing) the HTML written by an authoring tool. There were some questions about how much canonicalizing an authoring tool should attempt. Long mentioned that AOL used to get grief from customers because their tool automatically cleans up its HTML output, but they no longer do, since the tool now additionally pretty-prints the HTML source. It appears that users perceive enough extra value in the pretty-printing of the generated HTML that they are willing to accept the cleanup.

Masinter: There are people who are claiming to do HTML authoring who want their mail client and mail servers to send around marked-up mail. They are using a kind of HTML that they make human readable through operations such as centering HTML source text between <CENTER> tags.

Whitehead: Is there an interoperability concern with regard to pretty printing?

Long: Probably not. I would be happy if we just made it so that everybody's client could do a PUT to everybody's server. That would be the most basic form of interoperability -- everything else is icing on the cake.

Nielsen: The W3C could put up a playground where everyone could test out the PUT interoperability of their authoring tools, so long as we don't get into the test suite business.

Long mentioned that the focus of his talk is not interoperability. He followed Dan Connolly's suggestion to go through your system and find out where you feel fuzzy and the least likely to be doing something in the same way as everyone else.

Continuing his presentation, Long described how moving pages brings up the issue of how to move relative links. Moving image files is easy, because the IMG tag indicates the image is considered part of the document, and should be moved. However, moving audio files is not as easy because they are included via a A HREF tag, which could refer to a close or distant resource. The best solution to this problem is to employ collections and move mini-webs. Unfortunately, many beginning users do not use the AOLpress mini-web functionality. Another solutions is to have a usage practice where relative URLs are employed for references internal to a collection and external URLs are used for references outside of it.

This led into a discussion on the semantics of

"/x/y" style relative URLs

I.e., they're not "../foo" or "http://.."

Current versioning implementation slide:

Long reported on a prototype effort to add versioning capability to the AOLserver. In this prototype, reads employ a "Content-Version" header to retrieve a stated version of a resource. Writes provide a "Derived-From" header so the server knows how to check-in the version of the resource. The server prevents dirty writes. This implementation solves the "CREATE" problem (i.e., it does not need a separate CREATE method for the initial creation of a resource, because specifying a null Derived-From header directs the server to perform a resource creation). Since this prototype was completed, the header syntax has changed for HTTP 1.1. An incomplete report on this is available at URL:

http://staff.navisoft.com/users/d/dave/w3cwd/wd-version-960128.htm

As an aside, Long mentioned that users really wanted to know if they were going to overwrite a document. This was accommodated by providing a dialog box with "OK" and "Cancel" options whenever a document was about to be overwritten. This required performing a HEAD operation on the resource before a PUT could take place.

At this point there was a discussion of entity tags (Etags, an opaque string which identifies a snapshot), and their potential use in versioning. Long believes entity tags are sufficient to avoid lost updates. Masinter recommended as an action item the review of entity tags to see if they are adequate for versioning. Nielsen stated that use of entity tags may have some interaction with caching that would be undesirable.

Long continued his presentation by describing locking within AOLserver and AOLpress. The AOLserver implements LOCK and UNLOCK methods. If a resource is currently locked, a "Locked-By" header is returned stating who currently has the lock. During an unlock, if you have write permission on a resource, you can unlock it. If you do not have write permissions, a message is displayed stating that the resource is currently locked, and by who. An OPTIONS method is used to detect server support for locking.

Access Control slide:

It is possible to (allow, deny) a (user, group, netmask) for a particular URL.

There is a forms based user interface for access control functionality. There was a discussion of the value of this approach. The main benefit is that clients do not need to understand the content of the forms, they just merely need to be able to put them up.

Versions slide:

Resource revisions are time stamped and saved. Revisions are accessed through a prefixed URL, with a time stamp in it. Relative links and images are resolved as of that revision's time stamp.

PUT vs. POST slide:

Long expressed his position that PUT is a much better method for writing content than POST. PUT simplifies access control. PUT simplifies infrastructure such as caching proxies, gateways, etc., since it is ambiguous whether an arbitrary POST is transmitting an order for a pizza or writing a resource. There are also general-use servers which are starting to support the PUT method.

Namespace methods slide:

An example of the result of a BROWSE method is the return of an application/x-navidir MIME type.

application/x-navidir

{MIME type} {name}

...

Dave then displayed a slide on the issue of whether we should standardize Server API's. As rationale for why this might be desirable, Long mentioned that if, as an author, you need to do something to the server for your pages, it would be nice to use an existing API rather than talking to the site administrator for every change. ISAPI was mentioned as a possibility.

Variants slide:

PICS support slide:

Should raters provide well-known URLs to forms-driven interfaces for generators that produce rating-specific encodings?

Nielsen mentioned that the W3C has a tool where you can plug-in a PICS rating system, and it pops up the correct buttons, sliders, etc., and allows you to set the rating for a page.

Miscellany slide:

A question was raised about redirecting URLs, and how to tell if a resource has been permanently moved or only temporarily moved.

Brown: It would be useful to specify what is the guaranteed URL.

Masinter: You can register your URLs with the OCLC Persistent URL Service at URL:

http://www.purl.org/

Nielsen: You can abstract the Web space from the file space by having the server remap the resources that it owns.

FrontPage - Schulert

Release 1.1 was shipped this Spring, and release 2.0 will be shipped sometime this Fall. Vermeer was bought in January by Microsoft.

FrontPage was a client-server web authoring tool from the beginning. Ideally, FrontPage wants to be server independent -- they do not want to be in the server business, and would prefer to use HTTP as the base upon which they build FrontPage.

When they first started development of FrontPage, using the POST method was their only choice, since it was widely available on existing servers. They implemented an RPC-like mechanism on top of POST. There are somewhere between 20 and 30 entry points into FrontPage using this RPC-like mechanism. FrontPage has three server extensions executables, one for each of the three levels of access control available.

During design, FrontPage was aimed more towards the mainstream user, rather than the power user. The design was approached as a standard client-server development, where functionality was placed on whichever side of the wire it most made sense.

The FrontPage server extensions include much functionality, such as the ability to return a link map for the site, enhanced semantics for some operations, and the ability to set access control from the client. It does not include all features of SSA client control. There are capabilities for link repair, such as being able to inform FrontPage that a given URL has changed, and having FrontPage fix it in all of the pages on the server. There is a bulk upload (multi-resource at a time) capability.

FrontPage features webbots, which are objects that can be dropped into HTML and which are active when a page is uploaded to the server. An example webbot is the "day last changed" bot, which automatically inserts the date the web page was last changed into the source HTML for the page. Other webbots include a "substitution" bot, which can, for example, replace a company name in all web pages on a server, and a "table of contents" bot.

Schulert stated that the FrontPage group would like to get out of the business of maintaining the CGI scripts and server extensions.

Whitehead: How tightly are the webbots tied to using POST?

Schulert: We just need a way of invoking the webbot behavior.

Masinter: If you want to get out of maintaining those three CGIs, could you make FrontPage work with PUT?

Schulert: We can see a path out of it, using server-side Java or Visual Basic.

Schulert stated that you can post a web up to a server using FTP. You can post just changes that way, too. This solves the problem of directly editing the server when you don't want to make direct changes and a versioning system is not available. Hence we could use PUT in a very basic way, like using FTP, to have an interoperable write capability with other servers.

FrontPage currently has a bundled search engine, but Schulert does not view FrontPage as being in that business either -- the search engine is just enough to provide basic functionality. Ideally, Schulert would like version control, and search engines to be pluggable on the server.

In FrontPage 1.1, there is conflict detection and collision prevention, but no resolution help.

There was some discussion surrounding the problems of what you author is not what you are getting -- there can be things that the author knows that the server doesn't get. For example, if I delete a slide from a presentation, or a message in a discussion, the adjacent next and previous links are munged, and the server may know nothing about remedying that. Schulert sees there being an HTTP-level problem, in the long term, in this change of abstraction and also having conflicting server-side extensions. How do people's extensions cooperate with each other without knowing about each other?

Schulert stated that there is a core set of functionalities beyond PUT.

Suggestions:

Goal of Working Group - Whitehead

Whitehead next led a discussion about the goals and membership of the working group.

Whitehead displayed a slide which stated that the goal of the working group should be to make distributed authoring as pervasive as browsing is today.

Brown: I don't know about pervasive.

Long: I keep getting E-mail from customers who are using Netscape and notice a typo or spelling mistake and want to fix it. That's the world we want to address.

Masinter: We could be modest and not change the world, and simply have interoperability among our tools.

Fein: But we want to have interoperability for things we would like to do, not just what we are trying to do.

Masinter: Yes, we should look about two years ahead.

Nielsen: We should have something at level 0 very quickly -- I can't see farther than six months out.

Whitehead then recommended that the group adopt as its goal/objective/aspiration: to ensure that distributed web content authoring tools are broadly interoperable.

Masinter: We should change ensure to enable, so it would read: enable distributed web content authoring tools to be broadly interoperable .

Seiwald: We ought to keep that goal in mind, as a mission statement.

There was a suggestion to add a statement about standardizing features that exist today.

Nielsen asked that interoperability be limited to the HTTP framework.

There was a discussion about what was meant by interoperability, and what level of interoperability should be strived for by the working group. From this discussion, it was clear that there are different kinds and scopes of interoperability, and that working towards broad interoperability was a reasonable statement of the group's goal.

The working group adopted as its goal:

Enable distributed web content authoring tools to be broadly interoperable.

Sponsorship of Working Group by World Wide Web Consortium

Whitehead next led a discussion about whether this working group should seek sponsorship by the W3C. Nielsen mentioned that the W3C has a web page on the process used to create a W3C working group, at URL:

http://www.w3.org/pub/WWW/Consortium/Process/WorkingGroup.html

Nielsen mentioned that being a W3C working group implies that you have to honor the W3C agreements, including the meeting procedures, and it has to work on a focused output.

Seiwald: What is the difference between being 14 random people or being affiliated with the Internet Engineering Task Force (IETF)?

Masinter: If you are a working group, there are legal safeguards. For the IETF you have to have a chair, an editor, and a proposal and a draft. They don't require an initial meeting, and you are subject to the approval of the IETF Steering Committee for being appropriate to the IETF.

Nielsen: Becoming a W3C working group makes sense because it is within the domain of the area of interest of W3C.

Seiwald: Does the result go into the IETF?

Masinter: You can, if you need to make an Internet standard. There must be an open public review at the IETF, so that is another gate. This effort shouldn't be uncoordinated from the HTTP Working Group, however, it is not just HTTP that is of concern here.

Whitehead favors W3C sponsorship since it is the natural focus on Web-related issues. Hopes to not have quite the formality of IETF.

Fein: What does this obligate us to on behalf of our employers? What other conditions are there?

Masinter: If you work for a company that is a member of W3C, then your company has already aligned with the W3C agreements.

Hamilton: I just want to know what the ground rules are, rather than finding out retroactively.

Whitehead stated that he will investigate the intellectual property rights issues of non-W3C members participating in a W3C working group.

Meeting Goals, Criteria for Completion - Whitehead

Whitehead next put up a slide listing desirable meeting goals, and started a discussion on what would constitute criteria for completion of the working group's activity.

Masinter mentioned that he likes the idea of a having a demonstration of interoperability. For example, multiple authoring tools working against multiple servers and one document being edited by multiple authoring tools. It would be desirable that there be round-trip preservation of content features. For this demonstration of interoperability, the authoring tools should of the same level of quality that people already have.

Whitehead and Burns agreed that this is certainly ambitious, and it doesn't require versioning, or other advanced capabilities.

Burns: Having this demo actually run seems like a proper exit condition

Masinter: The IETF condition for advancing to a draft standard is that there be multiple interoperable implementations

Whitehead mentioned that a second demo is that N authoring tools do simultaneous editing of the same resource on one server, which would test lost update capabilities, collision handling, etc.

Brown: It appears there are several cases:

  1. A single client used against multiple servers.
  2. Multiple clients used against a single server.
  3. Multiple clients used against multiple servers.

Whitehead: We will need to refine these cases to make them more concrete.

Long: The deliverables are specifications by which each of these demonstrations can claim conformance.

Whitehead: Let's look at key interoperability issues before we specify deliverables.

Key Interoperability Issues (Nielsen's List)

Nielsen wrote a list of what he considered to be the key interoperability issues on the whiteboard in the conference room. He openly admitted that there was significant bias in his choice of issues.

The HTML item was added during group discussion. These interoperability issues then became the focus of discussion.

Schulert proposed that the GET-PUT working bullet is the following example: A user is browsing at a web page and sees a misspelling and wants, with simple steps, to edit the page and put the corrected page back out on the web.

Nielsen offered a scenario where there are a set of interrelated resources, and putting them back to a server atomically is desired.

Masinter: There are tasks below those - let someone author a new page, let someone do site maintenance.

Masinter then recommended that we go around the room and see what work people are willing to do as part of the working group.

Whitehead stated that he will write meeting notes for this meeting, but is not taking ownership of individual technical issues due to the number of issues, the amount of work required to coordinate the group, and his desire to concentrate on versioning issues.

Masinter is willing to help coordinate the activity of the working group, and contribute as an active member of the mailing list. He does not want to edit anything new.

Seiwald: I have more interest in the HTTP part than the HTML stuff and would be interested in writing up proposed HTTP changes. Willing to author proposals as needed.

Fein: There is more interest in HTML issues than HTTP issues in my case. I'm not as interested in protocols or back ends as I am in document content itself. I could take the Word requirements for HTML and write them down.

Masinter requested that Ron sort them out as content versus protocol etc. A good distinction between what is expected behavior for understanding, for not understanding, and for sort-of understanding a tag would be helpful.

Nielsen: That goes for all existing distributed authoring tools.

Dawson and Nielsen: Fein, Schulert, and Long all have lists of what they want to do in authoring tools - Word, FrontPage, and AOL. They could take on a task to sort out requirements.

Hamilton: I propose that Fein, Schulert, and Long (Word, FrontPage, AOL) use Fein's model (employed on his slides) for listing functional requirements and features and possible solutions.

Long and Burns are going to work on assembling something that is wordsmithed on functional requirements and scenarios. The tasks / scenarios that provide the cover for the features list.

Nielsen mentioned that he has three scenarios at three different levels in his slides.

Dawson volunteered to edit the scenarios document.

Seiwald invited everyone to submit two scenarios, not including the three we have.

Masinter: The task scenario document could be long. There is nothing wrong with that. We should also invite people to contribute. There may be substantial contributions from others who are not here.

Nielsen: I think coming up with a requirements document is nice. However, there are some things that need to be fixed and we could also start working now on the technical issues.

Masinter: We don't have to wait for the scenario document to be complete to start work on the technical issues. However, we will need the scenarios document to tell other people why we are doing this and how we will know when we are done.

Looked at Get for Edit versus Get for Browse. It's important to lay out what the alternatives are so you can say you discarded them.

Nielsen gave his suggestions for scenarios which should belong in the scenarios document:

  1. One person changing a misspelling in a document they found while browsing.
  2. Checking-in a group of resources which are related.
  3. Deleting an object from a web.
  4. Two people editing changes to the same resource.

Specified Deliverables

The group came to agreement that the following set of activities and deliverables should be produced by the working group:

  1. Task-oriented list of scenarios which interoperable distributed authoring tools will be able to perform.
  2. Collate lists of "key functionality" among AOLpress/AOLserver, FrontPage, Word, as well as other distributed authoring tools, such as Netscape

Looking at early September for next meeting, for at least the groups working on the deliverables above.

*** Meeting Adjourned ***


University of California, Irvine
Jim Whitehead <ejw@ics.uci.edu>
Department of Information and Computer Science
247 ICS2 #3425
Irvine, CA 92697-3425

Last modified: 23 Jul 1996