W3C | TAG | Previous: 3 May teleconf | Next: 25 May

Minutes of 12-14 May 2004 TAG FTF meeting in Boston

Nearby: Teleconference details · issues list (handling new issueswww-tag archive

TAG depicted in front of Stata
Center in Cambridge at 14 May 2004 ftf meeting. (full-size version)

Roll call from left: Paul Cotton, Norm Walsh, Roy Fielding, Mario Jeckle, Chris Lilley, Tim Berners-Lee, Ian Jacobs

Paul Cotton joined the meeting 13,14 May. Also periodically attending: Stuart Williams (by phone) and Dan Connolly (by phone, IRC, and iChat).

1. Administrative

  1. Resolved to accept minutes of the 3 May teleconf
  2. Resolved to accept draft monthly summary.
  3. Accepted this agenda
  4. Next meeting: 25 May 2004.
  5. Resolved: TAG plans to meet 5-7 Oct at a European location.

Minutes:

Completed/moot actions from previous meetings:

  1. Action PC, CL 2004/05/03: Review presentation materials when DC makes them available.
  2. Action CL, PC 2004/05/03: Read TBL's TLD essay on .mobi; send comments to www-tag.
  3. Action TBL 2004/05/03: Send pointer to .mobi essay to www-tag.

2. 12 May Minutes

2.1 Last Call Issues

Resources:

  1. Last call issues list
  2. Last Call Draft with inserted issues
  3. 10 May 2004 Editors' Draft
  4. Last Call Draft with inserted proposals
[IanJ]
TBL: We need to devote time to httpRange-14
[DanC_]
Tim, I agree that comments from PFPS and Hayes show dissatisfaction with our non-treatment of httpRange-14, but they don't really show anything new that would say why spending time on it today would be productive.
[tim_TAG]
The comments show that at least one more person seems to need a separation of the concepts of InformationObject and Thing.

Issues discussed here:

See also continuation of discussion of Last Call issues.

Issue msm2

Issue msm2

[IanJ]
TBL: I like the drop shadow
Action CL: Send IJ and SVG version of the first illustration in the document (Done)
Resolved: For issue msm2, we wish to keep the drop shadow.
[IanJ]
RF: Delete "This categorization is derived from Roy Fielding's work on "Representational State Transfer" [REST]."

Issue diwg2

Issue diwg2

[IanJ]
IJ: I did not incorporate the proposal in the 10 May draft.
RF: "The Spanish version" is a different resource.
IJ: I thought both are useful: conneg and URI for the Spanish resource.
CL: Yes, in general, show both sides of story.
[Roy]
A "tiny URI" example would be easier to convey as a reasonable use of multiple URI for one resource.
[Chris]
yes it would
tinyurl is the actual name ;-)
http://tinyurl.com/
[IanJ]
[General theme: when someone has said "why not this other way", ack the other way.]
Action IJ: Draft some text that talks about conneg as well as individual URIs.
TBL: Footnote perhaps.

Other changes the TAG accepted:

  1. s/resource owner/URI owner/
  2. s/URI ambiguity/URI overloading/

Issue stickler2

Issue stickler2

At their2 March teleconf, the TAG discussed changing the term "namespace document" to "namespace representation". This change was incorporated into the 10 May draft. The TAG discussed this change.

[IanJ]
Bug in 4.5.4

Namespace representation: If a namespace declaration binds a prefix to a URI, and that URI can be dereferenced to get a representation, then that is a namespace representation

IJ: We don't say "namespace document" in the 10 May draft. Does the web arch require the concept of "namespace document"
TBL: People use "namespace" to mean set of terms; "namespace document" to talk about those terms.

Issue stickler4

Issue stickler4

[IanJ]
NW: I think people should feel guilty if they don't provide representations!
Resolved: Rejected stickler4
Rationale: We meant this: people should - the community is poorer where representations are not available.
Action IJ - respond to reviewer.

Issue hawke1

Issue hawke1
TBL: I think that SH is asking for transparency. E.g., perhaps your agent should tell you when you are going to do a GET v. a POST. (e.g., change of cursor).
RF: In my dissertation - principle of visibility.
TBL: View source is good for other reasons - get people hacking and extending.
IJ: Also accessibility - emergency exit.
TBL: Sandro's comment is also about responsibility of agent to represent the user.
Action RF: Provide text for hawke1. There is general support for a visibility principle.

Issue dhm3

Issue dhm3
Proposed:
Resolved: Agreed. Editor will incorporate.

Issue dhm4

Issue dhm4

"The Web follows Internet tradition in that its important interfaces are defined in terms of protocols, by specifying the syntax, semantics, and sequence of the messages interchanged. "

[IanJ]
Reviewer asks "protocols as opposed to what"
[Chris]
its protocols rather than apis
[IanJ]
TBL: There's an important point here, but not protocol v. API.
[Chris]
but message passing rather than apis is another distinction with similar properties
[IanJ]
TBL: When we define protocols we are used to designing more resilient interfaces than when we design software APIs. (No reason we couldn't do the same for software.). Because protocols operate across boundaries, require scalability, etc. we are more used to designing flexible interfaces.
CL: Related point - in xforms they try to avoid the dom; they use message passing. Same rationale, even if there's no network .
TBL: They are defining error handling processes (doing their own operating system...) Put in a second paragraph in 1.2.4 - we do not use convention distributed protocol approaches when designing software APIs.
CL: Point to the client side example since distributed is not a requirement.
[Chris]
no, its not error handling at all!
[IanJ]
Action IJ: Add a second paragraph to 1.2.4 explaining what TBL said about resilience as typical design goal in protocols.
[Roy]
http://www.ics.uci.edu/~fielding/pubs/dissertation/evaluation.htm#sec_6_5_1

Issue rodriquez1

Issue rodriguez1

[IanJ]
Proposed: We don't think we need to address this comment in the Arch Doc.
[Chris]
XDI: Weaving the "Dataweb"
[IanJ]
[Stuart calls in from Crete]
Resolved: Reject rodriguez1
[Chris]
language for rejection (rationale)?

2.2 Findings Update

List of findings

The TAG did not do an update on:

2.2.1 How should the problem of identifying ID semantics in XML languages be addressed in the absence of a DTD?

How should the problem of identifying ID semantics in XML languages be addressed in the absence of a DTD?

[IanJ]
CL: Proposal - approve this and indicate that someone else is chartered to work on this. I've updated it as the XML Core WG has asked. They have a first draft of a document.
NW: Core WG will come back to TAG for review when work done.
Resolved: Accept finding.
Action CL: Add reference to the XML Core's working draft and publish approved finding. Announce finding (www-tag, public-tag-announce) (Done)
IJ to CL: Check out status section of this one: http://www.w3.org/2001/tag/doc/mime-respect.html
[Roy]
I'd prefer a shorter title: ID semantics in XML

2.2.2 The use of Metadata in URIs

The use of Metadata in URIs

[IanJ]
SKW: See my update, with summary of comments. I'll try to have another draft by mid-June

2.2.3 Versioning XML Languages

Versioning XML Languages

[IanJ]
NW: DO and I were working on this. The document needs an editing pass to incorporate comments from a DO/NW discussion a few months ago.
[Some discussion of editing]
Resolved: Ask Tim Bray and David Orchard if they would be willing to act in an editor role on the TAG, specifically on extant findings.
Action TBL: Talk to TB and DO about editor role.

2.2.4 Separation of semantic and presentational markup, to the extent possible, is architecturally sound

Separation of semantic and presentational markup, to the extent possible, is architecturally sound

CL: I have been working on this based on comments on arch doc section 4. I think arch doc oversimplifies; may need to say more about pros and cons. I expect a new draft soon.

2.2.5 Abstract Component References

Abstract Component References

The TAG plans to wait to hear from editor David Orchard.

2.3 Review of TAG's general issues list

Resources:

Issues discussed here:

See also continuation of discussion of TAG's general issues list.

The TAG indicated a desire to close these issues this year: 8, 14, 33, 34, 8, 15

2.3.1 HTTPSubstrate-16

Issue HTTPSubstrate-16

[IanJ]
Action RF: "Write a response to IESG asking whether the Web services example in the SOAP 1.2 primer is intended to be excluded from RFC 3205"
RF: An area director said "Yes, we intend 3205 to cover SOAP." Not sure where to take the action item beyond this.
TBL summarizes:
  1. People are tunneling through port 80
  2. There are benefits to one-port-per-protocol, but each time you create a new protocol you have to ask for a new port.
  3. The recommendation was thus not to use port 80 for other applications. Furthermore, there's a hint of "don't use http for things is was not intended for", but I disagree with that - promote reuse.
TBL: Abusive tunneling is another question. There are lots of times when you want HTTP functionality and you should darn well use it.
RF: I agree with TBL. But what do we do in practice? Write the IETF and say they are confused? (they might ignore). We could also submit a replacement draft, or our own description. It's difficult to find something specific now for something to respond to. I wish there were a suitable replacement strategy, but there isn't. What about if we write a few paragraphs expressing where we felt unease. I think Mark wants us to write up a general critique of the RFC and present the editors with comments from the TAG. I didn't see any way to reach consensus on that, but I could draft some comments.
IJ: How involved would this be?
RF: We could put this issue in limbo until someone tries to enforce 3205.
Resolved: Close Roy's action item on this issue
Resolved: Move HTTPSubstrate-16 to deferred state.
Action RF: Write descriptive paragraph explaining this issue's state.

2.3.2 fragmentInXML-28

Issue fragmentInXML-28
(NW: This touches on httpRange-14)
CL summarizes the issue.
Resolved:
  1. To the general question of "Do fragment identifiers refer to a syntactic element (at least for XML content), or can they refer to abstractions?" - see the section on fragment identifier semantics (i.e., the media type identifies a specification, which tells you their semantics).
  2. For the specific question in SVG, DC's observation is correct since it's licensed by the SVG spec.
Action CL: Write up a summary of the resolution.

2.3.3 binaryXML-30

Issue binaryXML-30

[IanJ]
Proposed: Close and state that the W3C has chartered a WG to investigate this.
CL's summary has been given to them as input.
[Roy]
The potato has been passed.
[IanJ]
Resolved: Move this issue to deferred status with a note that the TAG anticipates considering the Binary Char. WG's deliverables.

2.3.4 DerivedResources-43

Issue DerivedResources-43

[IanJ]
NW: Note that Xinclude no longer allows fragids.
[TAG tries to understand the issue]
Resolved: The TAG no longer understands the issue.
Action NW: Write to DO saying that XInclude no longer uses frag ids and the TAG is unable to construct from its meeting record what the issue was. We will discuss this further if we get help, but otherwise expect to close without action.

2.3.5 xmlChunk-44

Issue xmlChunk-44
NW: Core WG and TAG discussed this in Cannes. Not sure we made progress. I believe that there are wildly divergent ideas about what chunk equality means.
TBL: But we were in the same place re: URI comparison and the TAG did reach consensus.
NW: I hear TBL saying that he'd like a document summarizing different comparison approaches. There is another issue regarding comparing canonical forms.
TBL: Issue 44 is whether a chunk of XML has meaning out of context. Chunks only have meaning when XML is wrapped around them and there's a media type at the top. E.g., does xml:base propagate? If a chunk had a base in one document and it's moved, does the base move with it?
NW: Do we want to take a stab at it?
TBL: Note that there are two canonicalizations already. You need another one, e.g., including:
  1. must have namespaces (open issue about "unused" ns)
  2. xml:lang
  3. xml:base
TBL: If we were to propose a canonicalization, those are the sorts of issues we'd have to address.
CL: Note that XML doesn't have inherited attributes; they don't appear in the infoset. You could do the following (1) propagate attributes (2) canonicalize.
[TBL asks whether xml:lang and xml:base are to be inherited at the infoset level]
RF: "XML Fragment Interchange" is a CR...
NW: And it's been there for a long time. It's a related problem.
MJ: I can have a private equivalence function.
TBL: We are looking at a minimal equivalence function for all XML applications.
[Chris]
ok so this would define and name *an* equivalence function
[IanJ]
NW: If I were asked to define an equivalence function I would: (a) define at infoset level (b) not use the canonicalization specs and (c) include the possibility of error - you have to perform certain operations when you extract the chunk.
TBL: Note also, e.g., in a particular application additional information might need to be added to the chunk (e.g., CSS style for an SVG chunk).
[Chris]
hmm... 'in scope' xl pis for stylesheets ..... hmmm
[IanJ]
NW: Another point - infoset includes prefixes and we don't need the same prefixes. Let the minutes show that the TAG understands that Paul Cotton has concerns about this issue and is not currently present.
Action NW: Write up a named equivalence function based on today's discussion (e.g., based on infoset, augmented with xml:lang/xml:base, not requiring prefixes, etc.).: Write up a named equivalence function based on today's discussion (e.g., based on infoset, augmented with xml:lang/xml:base, not requiring prefixes, etc.).
NW: I don't expect to have text before mid-July.

2.4 Review of 10 May Editor's Draft

The TAG walked through changes in the 10 May Editor's Draft. See also continuation of discussion of 10 May draft.

  1. Summary of changes
  2. Diffs
[IanJ]
RF: Need to update RFC2396 reference
[DanC_]
(I skimmed one of the recent versions, only enough to realize I'd more time to read it carefully. I saw a _lot_ of changes.)
[IanJ]
DC: I read 10 may draft enough to know that there are a lot of changes and that I need to review more carefully.
RF: I read the first half.
[Zakim]
DanC_, you wanted to ask if anybody has read the 10May draft
[IanJ]
RF: I don't like the change from "orthogonal" to "independent". They don't mean the same thing. HTTP is dependent on the URI protocol, but they are orthogonal protocols. "loosely coupled" not same as orthogonal either. Orthogonality generally refers to things traveling on different paths.
TBL: The idea is that you change one spec without changing the other. That's a form of independence.
Resolved: Change "independent" back to "orthogonal". Say "may evolve independently" instead of "independent". Drop "loosely coupled."
[DanC_]
+1 "resource is a periodically updated report..."
transition sentence sounds editorial.
the "designers" bit seems editorial.
[IanJ]
RF: In 1.1.1, bullet 1, delete "i.e., " to end.
[DanC_]
s/document involve human activity/document that involve human activity/
[IanJ]
On "Many of the examples...."
IJ: This is from voice browser input on the doc.
CL: In voice browsing context, you only have input device (phone), client runs remotely.
[Chris]
client reporting errors is therefore much harder to do
[IanJ]
+1 to citing the work.
[Zakim]
DanC_, you wanted to suggest citing their work

Section 1.1.3

[DanC_]
Roy, how about citing [REST] from the acks section, ala "This document borrows several terms and concepts from [REST]."
ok, never mind.

Section 1.2.3 Error Handling

[Roy]
editorial - s/indicate through/indicate such consent through/
[IanJ]
TBL: I'm not happy with: "Agent recovery from error without user consent is harmful." E.g., TCP recovery from error....
[Norm]
Editorially, s/specifications/specification designers/, strikes me as not ideal
[IanJ]
DC: The user does give consent by using that protocol. Same as HTTP 404; that's part of the protocol.
TBL: I think that what we are saying here is "If an error occurs and some behavior happens which will mess up the user"; that's a bad thing. Introducing bad data into the system is a bad idea.
NW: We can distinguish "error correction" - error occurs and you can fix it although the error never occurred, v. "error recovery" where you can't correct the error...
TBL: How do you define the set of corrections that are "bad"? [See also issue schema1]
RF: I think if we distinguish correction v. recovery the problems go away.
TBL: User agents that correct errors without the consent of the user are doing the user a favor.
[DanC_]
"User agents that mask errors in a way that is not licensed by protocol documentation are not acting on behalf of the user."
(trying some scribbling... hmm...)
[IanJ]
TBL: User agents that mask an error from which they cannot recover, and that do so without the consent of the user, are not acting on the user's behalf.
Action IJ: Fix the text to make error correction/error recovery distinction.
[DanC_]
(hmm... is the connection between specs and consent written down? hmm.)
[Roy]
404 (not found) instead of "404 (resource not found)"

Section 2. Identification

[IanJ]
"The identification mechanism for the Web is the URI."
What about OWL/RDF methods of identification? {Comments from Sandro Hawke}
DC: OWL identification mechanisms don't give the same network effect benefits.
"Resources exist before URIs; a resource may be identified by zero URIs."
[DanC_]
"References from one part of the web to another should use URIs."
[IanJ]
TBL: OWL uses "indirect identification" as well as through URIs.
DC: OWL has other mechanisms, but they don't contribute to the network effect.
[DanC_]
(what did timbl ask?)
[Norm]
Timbl asked "can we use 'name' around here"
[IanJ]
TBL: What's a "part of the Web"?
[DanC_]
section 2.7.2. Assertion that Two URIs Identify the Same Resource discusses OWL, btw.
[IanJ]
TBL: How about "We refer from an information resource to anything"
[DanC_]
"In order to participate in the network effects of The Web, use URIs for naming."
[Zakim]
DanC_, you wanted to note "A URI must be assigned to a resource in order for agents to be able to refer to the resource." is wrong
[IanJ]
IJ: I am proposing to write some text that starts off slightly more generally with the desire to identify resources. URIs are the most important mechanism for identifying resources; we are seeing the growth of others. Here are the properties and benefits of the various identification mechanisms.
[Norm]
1: The primary ident. mech. is the URI
2. The ident. mech. that offers the benefit of the network effect is the URI
[Chris]
+1 to 'primary'
[marioj]
what about 'preferred' or 'recommended'?
[IanJ]
TBL: Value of giving something a URI is that you can refer to it from many formats.
[DanC_]
I much prefer to say, in the principle, what's better about URIs, rather than just say "primary" or "recommended"
[IanJ]
RF: The value of the network increases exp. with the value of nodes in the network.
[DanC_]
the point about formats was made in some drafts; it seems to be gone now. I'd like to see it come back.
[IanJ]
RF: This increases the value of the Web to society and also the value of specific resources by virtue of them being referred to directly by other resources.
TBL: People who use local ids only in their design, other people can't connect to them.
[DanC_]
yes, the point is not one string versus some mechanism using 3 strings; it's global vs. local naming.
[IanJ]
IJ: I am hearing the question as a larger one such as "why use URIs instead of queries"?
DC: One issue is local naming v. global naming. When you do global naming, you get interesting naming effects.
[DanC_]
"In order to participate in the network effects of The Web, use URIs for naming."
[IanJ]
DC: Another issue is which global naming system. For historical reasons, people are using URIs.
TBL: When you use global identifiers instead of local identifiers, you increase the potential for sharing.
[DanC_]
perhaps rather than "for historical reasons": "While the choice of identifier syntax is somewhat arbitrary, there is a lot of value in the existing network of URI-named resources"
[IanJ]
RF: You could rephrase as a definition: "The Web is defined as the set of resources identified by a URI"
TBL: "The World Wide Web is a global system and uses a set of globally-scoped names called URIs"
[Norm]
What's wrong with "The World Wide Web is a global system and uses a set of globally-scoped identifiers called URIs"
[DanC_]
principle: global naming leads to network effects. constraint: use URI syntax to participate in the existing network of URI-named resources
[IanJ]
The World Wide Web uses globally scoped names for resources; these are called URIs.
TBL: d/"for resources"/
TBL: I agree but you have to then explain in what way "the network effect"
TBL: Syntax and semantics
[DanC_]
principle: global naming leads to network effects. constraint: use URIs to participate in the existing network of URI-named resources
[timbl]
s/syntax//
[IanJ]
+1
[DanC_]
principle: global naming leads to network effects. constraint: use URIs to participate in the existing network, i.e. the Web
[IanJ]
Action IJ: Incorporate DC's suggested principle and constraint instead of constraint at beginning of section 2.
New "Principle: URI assignment"
[Norm]
"A URI should be assigned to anything to which agents may wish to refer"
[IanJ]
TBL: "It is beneficial to assign a URI because then others can then refer to it by URI."
[DanC_]
+1 beneficial, or something like that
[IanJ]
RF: (principle) "Shared resources improve the network effect." (good practice) It is beneficial to assign URIs to identify resources one would be interested in.
[DanC_]
I think the good practice is about assignment, not use, no?
[IanJ]
[IJ to work on this]

Section 2.1 URI Comparisons

[DanC_]
hmm... the Constraint at the top of 2.1 is a non-constraint.
[IanJ]
s/people/assignment of ...

[Doesn't matter who does it]

[DanC_]
+1 get rid of people/resources ala ndw's suggestion
people/agents
[Norm]
"Web architecture allows more than one URI to be assigned to a resource"
[IanJ]
Resolved: Delete the box (i.e., no longer a Constraint); leave the sentence.
CL: What about "Do not make the assumption that a resource is only identified by a single resource." Or "agents that assume..."
[Norm]
Constraint: URI Multiplicity; Web architecture allows a resource to be identified by multiple URIs"
[IanJ]
[DC excused]
[Norm]
s/comparison//
More explicitly, s/false negative comparison/false negative/
[IanJ]
NW: I don't think it's a good idea to draw attention to camel case of domain name.
IJ: What about all-caps instead of camel case?
NW: I'm uncomfortable with discussion of case in domain part.
RF: I suggest removing the example Example.Com and move reference to section 6 of [URI] in its place.
CL: Note that depending on what the agent is doing (as its job) the agent is likely to use (or not use) the extended comparison algorithms.

Section 2.1.1 URI aliases

[IanJ]
RF: Move rationale of 2.1.1 up front - talk about aliases and bifurcation.
[RF to suggest text]
But: s/unofficial/official

Section 2.2 URI Overloading

[IanJ]
RF: I don't know any way to avoid URI overloading.
TBL: That's a serious error.
RF: Whether or not it's an error, how do you avoid it?
TBL: Ownership system.
RF: There's no question that the authority should not use the same URI for multiple uses. This para is about user, not minter. People use URIs without finding out the authoritative meaning of the resource. You can't use a URI to refer to anything other than what it identifies. You can say that an authority should not go around changing the meaning of its resources.
IJ: Could be rephrased as "Before you start using a URI, find out what it identifies."
Action IJ: Rewrite "Avoid URI overloading" in terms of good practice for agents - avoid URI overloading by find out what a URI identifies before using it.
IJ: I deleted moby dick

Section 2.2.2 URIs in other roles

[IanJ]
RF: TBL used the phrase "indirect reference", which I prefer. TBL said "that's ok, but that's not what we're talking about here." URIs identify resources. In addition, URIs can be used to indirectly identify things other than the resource.
Action IJ: In 2.2.1, don't talk about Web arch, talk about indirect identification and that it's ok as long as the context is clear when it is happening. And that when we say "identifier" we mean "direct" henceforth. (URI may be one component in a phrase that is a series of references that strung together identify something)
NW: And that's one way to lead to overloading.
IJ: I will remove distinction of web protocols world / non-web protocols world.

Section 2.3 URI ownership

[IanJ]
CL: Add that domain owner can delegate URIs (i.e., domain owner is not necessarily URI owner).
RF: Add "For example, " before ". When the HTTP protocol is used to provide representations, the HTTP origin server (defined in [RFC2616]) is the software agent acting on behalf of the URI owner." Move "When the HTTP protocol is used to provide representations, the HTTP origin server (defined in [RFC2616]) is the software agent acting on behalf of the URI owner." to section on interactions
[Roy]
editorial: 2.4, remove "normative" in first para (redundant)

Section 2.7.2. Assertion that Two URIs Identify the Same Resource

[Norm]
s/syntactically different/that are not character-for-character the same/
[IanJ]
[IJ recalls that earlier RF talked about sem web not relying on consistency among URIs]
RF: "One consequence of this direction is that URIs syntactically different can be used to identify the same resource." This was true before.
IJ: Yes.
[On third para of 2.7.2]
TBL: Make the second URI in the paragraph also contain a "#".
RF: The limits of ownership between two organizations is the same as within one organization over time.
Resolved: Delete third paragraph; doesn't add anything.

2.5 httpRange-14

[IanJ]
TBL: There are two questions:
  1. What is the Web Architecture?
  2. Fix things. Find a transition strategy.
TBL: I agree with NW about the practical issue of dereferencing a stem URI and getting back a huge representation in order to understand the secondary reference. One suggestion to address this has been to let people know through HTTP headers. E.g., people have suggested generating "virtual URIs" for data in a database. You could say in protocols "To get a document about this thing, use this other URI"
CL: Why is this better architecturally?
TBL: It's a way to get out of a corner case.
[Chris]
not clear - the uri taken stand-alone does not convey that
problem is that we have no granularity of resources. one uri gives one resource gives one message body ....
[IanJ]
NW: I can assert that a particular resource is a lake and I can make a different assertion that the thing is a building. How is that conflict different from the one you are describing (# v. /).
TBL: An OWL system would put up red flag saying "inconsistent data". RDF processor might let you go on.
RF: It was my understanding from Pat Hayes' discussion that RDF doesn't have any such reliance on URIs with or without a "#" (i.e., no special meaning for # in the RDF spec).
TBL: Right, the RDF spec doesn't tell you about URIs; but the URI and HTTP specs do.
NW: Is it sufficient to say "Using a URI to identify a property and a Web page is likely to hurt."
TBL: The problem is related to the network effect - when you share a URI and they are used differently, you encounter problems.
RF: The identifier doesn't identify the information. It identifies the change in information over time. Not observable by a single observer in the general case.
TBL: The hope, on dereferencing, is that you get the same information back when you do a GET. You are not concerned with consistency of information about something, but consistency of the information itself.
RF: That would be true if the world did not have methods.

TBL draws this diagram on the white board (full-size image):

Whiteboard diagram showing
relationships among classes of URIs.

TBL later refined his diagram with a drawing tool (SVG version, full-size PNG version):

Refined diagram showing relationships among classes of URIs.

[Disagreement between RF and TBL about usage of POST]
TBL: There are some hairy things that are just post-able and that don't quite fit in the architecture. I think it's important to explain the architecture in a way that makes the hypertext web and the semantic web work and that holds together. Don't let the possible corner case disturb the depiction of the REST-ful information space.
RF: A lot of people working with the sem web will want to be able to deal with post operations. TBL doesn't need this entire solution, IMO. If you use a URI consistently, you don't need more.
TBL: This is where RF and I were in Irvine --- the thing that you identify with an HTTP_URI is a picture of a car, and you expect to get back a picture of the car in each case. What you are interested in is the photo, not the car in the photo.
RF: I said: "All identifiers refer to one resource." The URI doesn't tell you what the representations you'll get; what is important is that the representations are consistent.
TBL: What do you mean about "consistency"?
[DanC_]
(I can't follow this because neither timbl nor roy has been able to finish a sentence.)
[IanJ]
RF: I think that TBL's solution doesn't solve anything.
TBL: The "about" relationship is not part of the architecture. It's not important that a document is "about" a car; what is important is the information that the document conveys.
[Chris]
(Dan - nor can we...)
[IanJ]
[impasse]
RF: As far as I can tell, the document would be consistent if we were to talk about individual information systems. E.g., browsers present a particular view of the universe. Browsers provide representations that are consistent with the state of the resources. There are other possible systems that can make use of a Web of resources.
TBL: I think that the hypertext web and semantic web overlap and that we can build the sem web on the hypertext web.
RF: My intention is that there is one web. Think of it as a common data source and various ways of operating on that data source.
[More discussion about consistency of information content across representations.]
TBL: This is about expectations when using http uris - people pass them around and expect to plug into their browsers and get back web pages (not "Mark Baker")
[Secondary issue of content negotiation across the hypertext/sem web boundary. TBL thinks it could work in some cases, but that's a secondary question. TBL would be happy if the answer were "you can't conneg across the boundary", more inclined to answer the primary question.]
RF: Another point is, in the absence of representations to communicate meaning of resource, there's no way to gauge consistency, and thus one could say that the URI identifies anything.
[DanC_]
ndw, if you're looking for compromises, there's one: if URI U has a representation, then it refers to an information resource.
[Norm]
hmmm...
[IanJ]
And if it doesn't, you don't know.
[DanC_]
... and http://markbaker.ca/ actually doesn't have any representations; only http://markbaker.ca/index.html does, as you can tell from the Content-Location: header on the reply.
there are 2 axioms: { ?X log:uri [ str:notContains "#" ] } => { ?X a :InformationResource }. # the strong position on httpRange-14
[IanJ]
TBL: not thrilled with DC's approach; would prefer to know from the http scheme that those URIs identify info resources (and representations might be available).l
[DanC_]
and: { ?X webarch:representation ?R } => { ?X a :InformationResource } # a less-strong position on httpRange-14
[Norm]
What is webarch:representation?
[DanC_]
er... representation as we've defined it in webarch; e.g. the body of a 200 reply represents the resource you were doing a GET about.
[Norm]
ok. that didn't seem to help with tbl
[DanC_]
bummer; he was more receptive to it on other days.
[IanJ]
RF: The two things I have a problem with:
  1. The gross violation of orthogonality principle.
  2. Existing examples which violate the modeled separation.
TBL: I'm not telling people to change their robotic system, just how they describe what their system does (e.g., "identifies a robot control panel page" v. "identifies a robot")
NW: Suppose I set up a taxi service and you click on the URI and a taxi arrives at my door. Does the URI identify a taxi?
TBL: That's the taxi architecture getting you a taxi. I'd like two terms in the document (e.g., thing and information resource). Information space consists of information resources, interconnected by links. The web also consists of other resources, and those are not interconnected by links. This document focuses on information resources. Also need to say what we mean by "representation" if we do this.
RF: A Web browser takes any URI and treats it as though it identified an information resource, whether or not it is. The browser thinks of all URIs as being dereferenceable.
IJ: I'm not sure how one defines an information resource.
TBL: Hypertext web and sem web share some concepts: (1) identity (2) reference (3) uri ownership But a connection like ":bob :brother :joe" is not a "link". (Web link)
DC: Roy, do you think it would be useful to build on people's intuitions about Web as set of information resources?
TBL: It's useful to say what the hypertext web and sem web have in common.

See also continuation of discussion of httpRange-14

3. 13 May Minutes

Paul joins the meeting today.

IanJ
Five issues identified yesterday for later discussion.
  1. URI ambiguity
  2. Definition of "on the Web"
  3. httpRange-14
  4. Namespace documents or Namespace representations
  5. Fragment identifiers

3.1 Review of TBL's critique of .mobi TLD proposal

"New top level domains considered harmful" from TBL

IanJ
TBL: I still have some comments to fold in. Half of my comments are generic; about any new TLD. I am wondering whether I should send the generic comments to other lists. There are TLDs that I think would be a good idea , e.g., a TLD for telephone numbers. I think it's ok to create a new TLD that offers NEW functionality; e.g., a system with a particular persistence policy and new social arrangements. But, e.g., ".biz" and ".info" were not really worth it. Domain namespace is flat - people remember brand names (i.e., tree structure doesn't work). Users don't find it useful to have acme.net, acme.info, etc.
PC: Similar comments have been made on our own list.
[TBL continues to summarize his proposal]
TBL: One of the costs of change involves FUD about ownership - people are forced to pay more money to keep their existing position. The addition of new TLDs reduces the value of existing investments. Like printing money. The rush to get new names creates uncertainty in the market; also not a good thing. There's also chicanery during the rush. A number of people also commented "Why is this a for-profit venture?" That raises a lot of suspicion in the community. (The above comments were generic.)
TBL: Specific problems with ".mobi":
  1. If it's for mobile content, that's unsound.
  2. If it's for mobile carries, that's unsound.
TBL: People use numbers for their home lines and their mobile lines. There are many situations when the fact that something is mobile is unimportant (or almost "in the way"). Hard to define what it means to be "mobile" in fact. There are aspects of "moving around" but also "small screen", etc. Creating content for a particular device profile is an assault on the mission of W3C to create a universal Web.
[See the proposal for more details; scribe pauses]
TBL: Example of bookmarking and device variation according to context.
[Other examples of ghetto-ization through TLDs]
CL: See my additional comments.
[Discussion on how to do content tailoring with existing technologies such as CSS and CC/PP]
CL: Authoring tools should increase support by providing support for @media and allowing previewing in different modes.
RF: I agree with comments from others about the line regarding semantics of names. More confusing than helpful. Other than that I completely support TBL's position.
DanC
(I wrote an analogy between semantics in names and which/that, i.e. restrictive and non-restrictive clauses...)
IanJ
Action TBL: Update TLD essay for consideration this afternoon. (Done; see below)

See also continuation of discussion of TBL's critique of .mobi TLD proposals.

3.2 httpRange-14 (continued)

Two topics:
  1. Different uses of the term "resource"
  2. Not requiring a name to have one meaning.
DanC
Re: Attributes should only be there if part of the name/address space http://lists.w3.org/Archives/Public/uri/1997Feb/0025.html
Chris
+1 to ensuring we get to other issues
Norm
TBL: The http: space was designed specifically to create an information space. The hooks into ftp space were designed to leverage existing content. The hook into mailto was really to hook into a different architecture. The REST thesis describes the architecture and the properties that make it interesting. Those principles are what we're trying to explain. The thesis also tries to push the REST idea as far as you can. This allows you to look at anything through "REST-colored" spectacles. There is an abstraction in which everything is viewed through representations. Those REST colored spectacles give you the property that you don't distinguish between the car and the picture of the car. It's silly to argue about the resource because all you see is the representations. In the TAG architecture we need to talk about more things. Within the dotted-line of "gettable" things REST describes all the shoulds. When you go outside the dotted-line (mailto URIs, for example) the REST architecture does not apply.
RF: I wouldn't say it that way, the resources don't change...
TBL: That's not how email works
RF: We are talking about different information systems, not different resources.
TBL: I'd like you to only use the REST-colored spectacles for the REST-based system.
RF: There is no boundary between resources that are viewable .........
TBL: Please, can we put a boundary. Insisting on using the REST spectacles for everything means we can't make use of the other systems. We need to ack that there are other things in the Web arch that are not REST-like. In the sem web we distinguish between the photo of someone and the person. We distinguish the person and their email address. There are URIs where there is no GET, just POST -- mailto URIs. I think we can look at the URI scheme and tell some things about the identified resources. E.g., when you see a "http" URI, it's reasonable to use GET with it. And the value of repeated GETs is sameness of information across the GETs.
IanJ
RF: what is the commonality among resources identified by http URIs? They all have the same http interface? They all are document-like?
TBL: The essential identity is consistency of information content retrieved through GET. mailto is a store-and-forward system. It's one-directional (with SMTP).
[TBL and RF talk about pushing the REST model as far as they can take it...and limits that TBL encounters]
RF: I hear TBL saying he is convinced he needs a class to outline resources identified by HTTP URIs. My experience suggests that this is unnecessary. But it's ok to me if you do draw a line, as long as the interface is not more constrained than it is today. If you say that HTTP resources are all "document-like", that is problematic. E.g., what about CGI gateways.
[TBL goes to white board]
TBL: When you look through the REST interface, you don't have to (or can't) see the difference between the file on disk, the photo, or the car. The interface hides the difference from you. But other systems do separate these entities from you (e.g., photo of a car, car)
[Diagram represents constancy of information content across GET]
TBL: E.g., info could come from a different disk; what's useful is the same information content.
RF: What about when you have an HTTP server running on the car radio.
TBL: When two people do GET with a URI associated with the car radio, they don't want an audio stream one time and a photo back the next time. In this case, they are interested in consistency of information content.
RF: In reality, there are also gateway HTTP URIs.
TBL: I agree - HTTP URI space is big enough to identify resources that are not information objects. But the vast majority of resources identified by HTTP URIs are information resources.
TBL Proposal:
  1. In the document, introduce the concept of resources with consistency of information content (using http GET as an example, used most of the time). [That section talks about value of HTTP GET]
  2. HTTP has other methods (e.g., PUT, DELETE are connected to GET). POST allows very different behaviors.
  3. Thus, there's an important class of objects that have info resources that have http URIs. But there are other resources with HTTP URIs where other methods dominate (e.g., Web services, html forms using POST).
timbl
Proposal - HTTP range is the set of things which potentially support the HTTP interface.
IanJ
TBL: Propose that the answer to httpRange-14 is "Those things that potentially support the HTTP interface." Furthermore, those things that support GET (regardless of protocol) are special and we will talk about interesting properties. I'd like to call those things information resources.
timbl
In document, specify the set of things which in HTTP potentially support GET, or in other protocols an equivalence operation.
IanJ
TBL: When something supports GET, it has certain properties. When it supports POST, it has other properties. I'm happy to raise a separate issue about what relationships, if any, there are among those properties (or what relationships there should be, etc.).
IanJ
NW summarizing:
  1. Range of HTTP URIs is the set of resources that may support the HTTP interface.
  2. Those resources that support GET have the following interesting properties ....; TBL calls these resources "information resources"
timbl
http://lists.w3.org/Archives/Public/public-webarch-comments/2004JanMar/1057.html
http://www.w3.org/2004/delta#insertion
http://www.example.com/foo
Norm
PC: 1. what changes to the doc, 2. what impact does it have on httpRange-14, 3. what impact does it have on new issues, and 4. what impact does it have on public comments
IanJ
IJ: I would start with TBL/RF writing resolution to httpRange-14. I can edit document given a clear statement of how to proceed.
timbl
Paul has been much more strict: Documentation of its effect on issues and documents and comments, all to be reviewed and okayed by the group
DanC
PROPOSED: to address httpRange-14 by introducing "information resource" for those resources that support GET, i.e. have representations, contingent on doc changes agreed by RF, TBL, IJ
DanC
(paul's suggestion is fine by me too.)
IanJ
NW: I prefer PC's proposal. I'd like, for this particularly thorny issue, that RF and TBL do all the parts of PC's request.
RF: I won't get any work done on that until mid-July.
IanJ
Action TBL/RF: Write up a summary position to close httpRange-14, text for document.

See also continuation of discussion of httpRange-14.

3.3 Architecture Document Review (continued)

Section 3.3.1 Using a URI to Access a Resource

IanJ
Added "necessarily" per TAG resolution.
DanC
norm, I wasn't asking Ian to point out the editorial stuff; I was asking you to recruit reviewers for this draft

Section 3.2. Messages and Representations

IanJ
d/electronic

Section 3.3.1 Media Types and Fragment Identifier Semantics

Zakim
DanC_, you wanted to ask why the Greek letters (U and F) are still there
IanJ
RF: Delete para after story in 3.3.1
Norm
If we want to keep the normative text in the current para 1, I suggest: "Per [URI], given a URI with a fragment identifier, and a representation retrieved by dereferencing that URI (which is authoritative), the (secondary) resource identified by the URI with the fragment identifier is determined by interpreting the fragment identifier according to the specification associated with the Internet media type of the representation data. Thus, in the case of Dirk and Nadia, the authoritative interpretation of the fragment identifier is given"
Norm
by the SVG specification, not the XHTML specification (i.e., the context where the URI appears).
DanC
(oops; looks like it's my fault: Action DC: Write up some replacement text for text at beginning of 3.3.1, from 15 Nov 2003 teleconf)
IanJ
I propose to remove this story, which is not that useful.
DanC
no, keep the story; but take "authoritative interpretation" out; yeah, what PC said
IanJ
CL: I think the story is saying: frag id occurs in an xhtml content; fetching happens without fragid; interpretation occurs in the SVG context. I.e., separate occurrence and interpretation.
DanC
yes, elaborate the story to have the link found in XHTML and pointing to part of an SVG thingy.
IanJ
Resolved: Close DC's action.
Action CL: Rewrite story at beginning of 3.3.1. Consider deleting para that follows last sentence third para after story in 3.3.1. "Note also that since dereferencing a URI (e.g., using HTTP) does not involve sending a fragment identifier to a server or other agent, certain access methods (e.g., HTTP PUT, POST, and DELETE) cannot be used to interact with secondary resources."
RF: Delete "e.g." since people may interpret it as "i.e.".
IJ: How about "for example" instead.
RF: Ok with "for example"
[para ok]
DanC
A#X refers to b
IanJ
IJ: so add "using that URI" (i.e., the one with the frag id)
TBL: Say instead "You can't do POST using a URI that identifies a secondary resource."
DanC
that would work. "You can't do POST using a URI that identifies a secondary resource."
Chris
yes, agreed
timbl
+1
IanJ
What about: "Note also that since dereferencing a URI (e.g., using HTTP) does not involve sending a fragment identifier to a server or other agent, you can't do POST using a URI that identifies a secondary resource."
Chris
suggest in last paragraph s/such interpretations are not authoritative./such interpretations are not authoritative.because the media type is not known.
IanJ
3.1: "Agents may use a URI to access the referenced resource; this is called dereferencing the URI"
DanC
http://www.w3.org/2001/tag/2004/webarch-20040510/#uri-dereference
IanJ
RF: delete "or other agent"
DanC
NW too
IanJ
What about: "Note also that since dereferencing a URI (e.g., using HTTP) does not involve sending a fragment identifier to a server, you can't do POST using a URI that identifies a secondary resource."
IanJ
What about: "Note also that since dereferencing a URI does not involve sending a fragment identifier to a server, you can't do POST using a URI that identifies a secondary resource."
Norm
How's this: Note also that since dereferencing a URI does not involve sending a fragment identifier to a server, certain access methods (e.g., HTTP PUT, POST, and DELETE) cannot be used with a URI that identifies a secondary resource.
IanJ
Note also that since dereferencing a URI does not involve sending a fragment identifier to a server, certain access methods (for example, HTTP PUT, POST, and DELETE) cannot be used with a URI that identifies a secondary resource.
DanC
Norm's version seems good.
IanJ
TBL: Say "HTTP allows interaction with a primary resource." Take out "certain access methods"
DanC
hmm... "by design" has some appeal; it would be nice to say why this is the design...
Norm
Drop "Note also that..." to the end of the paragraph
IanJ
IJ: I note that RF has good text on why it is by design. I propose to delete "Note also that since dereferencing a URI (e.g., using HTTP) does not involve sending a fragment identifier to a server or other agent, certain access methods (e.g., HTTP PUT, POST, and DELETE) cannot be used to interact with secondary resources."
DanC
aye, s/during a retrieval action//
IanJ
RF: Proposed to remove "during a retrieval action".

Resolved:

  1. Remove "during a retrieval action"
  2. Delete from "Note..." to end of paragraph.
Zakim
DanC_, you wanted to note that secondary resources might be information resources; hmm... and to ask if this "Note also..." has already been proposed to the reviewer and if it's clear who's going to get back to the reviewer
IanJ
IJ: Regarding previous edit, I will go back to reviewer and say that we think that edits to penultimate paragraph answer the question.
Now looking at: "Parties that draw conclusions about the interpretation of a fragment identifier without retrieving a representation do so at their own risk; such interpretations are not authoritative."
CL: Note that because you can always get more representations, you never know "for good" the full set of representations ....
RF: Removing "Parties that draw conclusions about the interpretation of a fragment identifier without retrieving a representation do so at their own risk; such interpretations are not authoritative." will satisfy stickler8 and schema4.
Chris
but not really a good solution
IanJ
IJ: Recall that this sentence was to respond to DO question - can I assume a particular media type for particular URIs? The answer was "Yes you can, but if you don't get a representation, you do so at your own risk."
Chris
Better to say in a positive sense rather than just not discussing it. Point is that when you retrieve *a* representation, you still don't know if the fragment is consistent for the resource as a whole
DanC
I'm quite happy with "Parties that ... do so at their own risk". why remove it?
IanJ
IJ: Parties that draw conclusions about what a URI with a fragment identifier identifies without retrieving a representation do so at their own risk."
[discussion of issue schema4]
Chris
there are no guarantees - png, jpeg, text all have no bare names: "This is not a viable best practice recommendation, except as a bandaid, as it tightly couples URIs to representations, and constrains the ability of representations to evolve in untenable ways. This appears to highlight a weakness in the Web architecture that should be explicitly addressed."
IanJ
RF: I don't consider this a weakness.
TBL: There is a weakness - you can get a failure. But I think that the last paragraph of schema 4 is unwarranted.
Chris
we need to address it
IanJ
RF: I think that the other paragraphs added to 3.3.1 do help address their comments
Chris
saying 'any rich addressing is contrary to web architecture' is untenable, so I agree with their point
Norm
Many users have an absolute requirement to be able to point to parts of XML documents (that they cannot edit) that do not have IDs. To say that XPointer should be discarded is inflammatory.
DanC
I'm not familiar with such users, Norm. I wonder if there's a documented use case somewhere.
IanJ
Three proposals on the table:
  1. Delete "Parties..."
  2. Edit to "Parties that draw conclusions about what a URI with a fragment identifier identifies without retrieving a representation do so at their own risk.""
  3. Append to existing sentence: "because you don't know the media type"
timbl
Yes, but it is worth pointing out, as they ask us to point out, that there is a detriment in using XPointer that you are very specific to XML. If you can use a base local identifier it is better - also from resilience to document editing too of course,. Norm.
Chris
dan, you are saying that there are no uses who consider structure to be useful. i can see plenty of counter examples and find it hard to believe that you assert this
timbl
I think Annotea uses xpointer. (But are there users of Annotea?)
Norm
Yeah, I'm not suggesting that xml:id won't make it better.
Chris
annotea would need to
IanJ
RF: I don't think (c) addresses stickler8.
DanC
re a) Delete... is it said elsewhere, RF?
IanJ
Might be said above: "The fragment's format and resolution is therefore dependent on the media type [RFC2046] of a potentially retrieved representation, even though such a retrieval is only performed if the URI is dereferenced. If no such representation exists, then the semantics of the fragment are considered unknown and, effectively, unconstrained. "
DC: You can draw conclusions that the media type spec allows.
CL: But you don't know what the other possible representations are.
TBL: Recall that it's an error if the representations are inconsistent. You have to be able to trust the system.
Chris
ok so the onus is on the person minting the URO to deal with the consistency across representations - the user of the uri has to trust that
DanC
If I have doc#sec1 and I fetch doc and get HTML with <a id="sec1">, I can conclude, on the basis of the HTML spec, that doc#sec1 refers to that <a>.
Chris
And thus, after retrieving any of the representations, they have license to interpret the fragment And thus, suggested add on to the end of the sentence *is* correct. QED
IanJ
RF seems to agree with " If I have doc#sec1 and I fetch doc and get HTML with <a id="sec1">, I can conclude, on the basis of the HTML spec, that doc#sec1 refers to that <a>." But you can't conclude that there aren't other representations.
DanC
RF didn't agree, it seems to me.
timbl
Actually, specifically, you can hold the responsibility on the publisher is just to agree that the meaning of the document is that the secondary resource has the properties described in the representation.
IanJ
CL: Provider is responsible for consistency. Retriever has to trust that consistency has happened.
Roy
That assumes that the reader can fully comprehend the owner's intentions in providing that reference.
IanJ
RF: Your perception of consistency may not be the same as the URI owner's. The more precise you can be, the less risk there is.
Chris
oh ... so now its 'which consistency operator are we using' and 'why do you want to know'
Norm
Does this help: Parties that draw conclusions about the interpretation of a fragment identifier without retrieving a representation do so at their own risk; such conclusions are not authoritative.
DanC
(ugh; writing text about semantics of fragments is like coding, in that I'd need a quiet room for at least an hour or two to draft text.)
IanJs
This is from RFC2396 bis:"The semantics of a fragment identifier are defined by the set of representations that might result from a retrieval action on the primary resource. "
TBL: Nothing says that you can't get information about a secondary resource from other places.
DanC
(representations _rarely_ determine identity of #foo; they tell you things about #foo, but rarely narrow it down to exactly one thing.)
IanJ
TBL: Therefore it's wrong to say "Don't use other information." "People shouldn't draw conclusions about the identity of the secondary resource from a syntactic analysis of the frag id alone; someone could, for example, create a new media type which uses a similar syntax." If it's obvious in the referring document, you may know what type of object the URI is identifying. E.g., ":ian :drives :foo" I can conclude that :foo is a car because of my ontology.
DanC
(but info in the referring document has less status than info from the server; kinda like the a/@type attribute in HTML)
IanJ
Action IJ: Edit "Parties that draw...." to be about drawing conclusions from syntactic analysis.
IanJ
RF: In story, section 3, delete "by /satimage/oaxaca".
Chris
noise:signal ratio is infinite

Section 3.3.2 Fragment Identifiers and Multiple Representations, issue msm13

Issue msm13

IanJ
First para after story from RFC2396bis
This was deleted: "On the other hand, it is considered an error if the semantics of the fragment identifiers used in two representations of a secondary resource are inconsistent. "
TBL: I think you have to answer the question about what "consistent" means differently for the sem web and the hypertext web.
Chris
And delete the base element. So the internal links work
IanJ
IJ: I will make CL's change.
TBL: In "A resource owner who creates a URI with a fragment identifier and who uses content negotiation to serve multiple representations of the identified resource SHOULD NOT serve representations with inconsistent fragment identifier semantics", change SHOULD NOT to MUST NOT.
CL: PNG is tough since PNG defines no frag ids. More interesting would be CGM and SVG, both of which define frag ids. You can't conneg SVG and CGM since they use different syntaxes.
timbl
I suggest 3 examples: The fragid working in both, the fragid giving incorrect results; the fragid not being defined for one format.
DanC
3 examples sound good.
Roy
The sentence "It is not considered an error that only one of the data formats specifies semantics for the fragment identifier. " is confusing.
IanJ
From "Common User Agent Problems" (CUAP): 1.2 If the user attempts to follow a link that is broken because it designates a missing anchor, let the user know it is broken.
RF: I agree with CUAP, this is a known bug. It's a user interface error to not indicate that the reference was not found.
Chris
this is seen as a grave error in.for example, Web-based IETMs.
DanC
+1 s/SHOULD NOT/MUST NOT/
IanJ
RF: The paragraph does not say what a user agent should do when an agent doesn't find an anchor
IJ: I suggest we don't say anything. Or we can just point to CUAP.
Chris
3.3.2 should offer two possibilities and discuss the wins and losses in each
IanJ
RF: I find this confusing: "It is not considered an error that only one of the data formats specifies semantics for the fragment identifier."
IJ: Why?
DanC
yes, the good practice note is redundant with the para "For a given resource, an agent may have ...". Meanwhile, they both say 'should not' and I'm sympathetic to 'must not'.
Chris
one, a single object element with a conneg resource
two, a nested set of objects pointing to different resources
IanJ
RF: Even though it's not considered an error for the resource provider to provide representations that don't support frag ids, it is an error from the perspective of the user agent.
Zakim
DanC_, you wanted to endorse the 3 examples timbl sketched: "The fragid working in both, the fragid giving incorrect results; the fragid not being defined for one format."
IanJ
DC: I like three examples:
  1. Two representations where frag id works in both
  2. Two representations where they conflict
  3. Two representations where frag ids not defined in one representation.
Chris
so webcgm vs svg would be good for the second example
DanC
yes, note that the typical browser behavior of ignoring missing fragments is an example of silent recovery from errors, and is bad.
IanJ
TBL: I think we should point out that (c) is an error; can cross-ref to agent recovery from error.
[Support that (c) is an error]
And that silent recover is also an error. Proposed: Change "Should not" to "Must not" in GPN.
[Support from DC, NW, TBL]
RF: I don't think it's reasonable to have "MUSTs" in GPNs.
Chris
reference for WebCGM fragid syntax:
http://www.w3.org/TR/2001/REC-WebCGM-20011217/REC-03-CGM-IC.html#webcgm_3_1_1
Ditto for SVG:
http://www.w3.org/TR/SVG11/linking.html#SVGFragmentIdentifiers
Chris
CL volunteers to write these examples
DanC
+1 ndw
IanJ
TBL: I'd like to talk about "inconsistent" in two contexts - sem web/hypertext web MSM13 - how is "consistent" used? For sem web, if you use conneg, the two documents should contain the same information (same semantics). But I can see where two formats would not be identical, e.g., one more expressive than the other. And there's a subset relationship between the two. In the hypertext web, more about the publisher being sufficiently satisfied that the retriever is getting sufficiently similar information. The publisher takes responsibility for the degree of consistency. For the sem web, we can talk about consistency mathematically.
DanC
(I note that the math/logic/kr world already has an established definition of consistency)
Roy
overloading is already defined in terms of consistency ;-)
DanC
is it Roy? I guess I better go check...
IanJ
"URI overloading refers to the use, in the context of Web protocols and formats, of one URI to refer to more than one resource."
Roy
[I was joking]
IanJ
[But that text will change based on yesterday's discussion...]
DanC
(to my mind, RDF _does_ have resolution, btw. cf cyc paper on context.)
IanJ
From msm13: "How do the responsible parties determine whether a given plan of using fragment identifiers is or is not compliant with this rule?"
IanJ
PC: They read the specs for the N representations and determine whether the frag ids have the same semantics.
timbl
q+ to re-ask paul the question
Zakim
DanC_, you wanted to suggest that we can define "consistent" in terms of overloading (nee ambiguity)
IanJ
NW: It seems to me that a given community developing media types need to describe what "consistency" means for a given media type.
Chris
inconsistency if they produce URI overloading .... yes
IanJ
DC: I think it suffices to define consistency in terms of overloading. Frag ids are inconsistent if they result in URI overloading.
TBL: How would we know when two representations are inconsistent? What about the case of a photo of a hat and another person gets a photo of a pair of gloves? I think that we would call those inconsistent.
Norm
PC steps out for 10 minutes
Zakim
timbl, you wanted to re-ask paul the question
IanJ
RF: If the reference is to "Photo of what's in my closet?" then it could be considered consistent here.
TBL: RF is saying that the publisher determines what is consistent.
CL: Consistency depends on what the resource is advertised as doing.
RF: The target of the consistency GPN is representation providers, so I don't think the comment from MSM13 is irrelevant, as the comment was about receiving agents.)
Chris
my part two was - the information provider defines which formats they provide. no-one else can provide a different format at the same URI
IanJ
TBL: User expectations about consistency also exist.
DanC
(hmm... telling each other stories is something I'd expect to do early in the game. responding to LC comments is more of an end-game thing. But we don't seem to be able to achieve end-game mode, i.e. reviewing text sent in advance of the meeting.)
IanJ
Proposal:
  1. Include three examples about coneg as proposed by TBL.
  2. State clearly that it's an error for representation providers to provide representations with inconsistent frag id semantics.
  3. Talk about consistency as being in the eye of the representation provider (not forgetting that users also have expectations).
  4. On the Semantic Web, formats allow a more mathematical determination of consistency when doing coneg.
DanC
in order to avoid 'it is an error', perhaps: "conflicting representations result in URI overloading, which, as discussed in section [earlier], is EVIL.
IanJ
RF: I think the last point is somewhat speculative.
TBL: I'm ok to drop last point.
Resolved to accept these changes to 3.3.2:
  1. Include three examples about coneg as proposed by TBL.
  2. State clearly that it's an error for representation providers to provide representations with inconsistent frag id semantics.
  3. Talk about consistency as being in the eye of the representation provider (not forgetting that users also have expectations).
Action CL: Provide three examples.
Action IJ: Revise text for other parts of proposal. And destiny of GPN in that section uncertain.

Section 3.4 Authoritative Representation Metadata

IanJ
Reviewing first paragraph
DanC
(I can't parse the sentence including "the design choice for the Web is,")
Chris
the design choice for the Web is, in general, that .... is unnecessarily verbose
DanC
(I thought we had pretty good wording in the finding... reviewing http://www.w3.org/2001/tag/doc/mime-respect.html ...)
IanJ
TBL: Two problems in 3.4 first para:
  1. Media type is special; inseparable from representation data.
  2. Other metadata relates to things that relate to the resource; need an authoritative source in that case.
IanJ
q+ to say that document definition of representation includes more metadata than content type
Zakim
DanC_, you wanted to clarify by counter example: Bob says doc23 is text/plain; Joe says doc23 is text/html; Bob gives URI doc23 to Fred, then Joe gives doc23 to Fred; they've communicated different things using the same URI. global naming breaks.
IanJ
[DC gives a counter-example]
Zakim
IanJ, you wanted to say that document definition of representation includes more metadata than content type
IanJ
In counter-example, people who say "Check out this URI" provide metadata.
marioj
thinks we should provide a definition for "Representation metadata"
IanJ
RF: We should be talking about the proper interpretation of messages here. Within a message you have different types of metadata. You want to talk about how to obtain authoritative metadata.
Chris
CL: so who is the authority when i post to a resource I don't own?
Chris
(people): we don't know
DanC: yes
IanJ
TBL: The semantics of the message is a function of the data and the metadata.
IJ: Why is it an error to ignore the media type?
DanC
A: for the same reason any other deviation from a protocol is an error
IanJ
IJ: What should title of 3.4 be?
IanJ
DC: What about "3.4 Message semantics"
IJ: I suggest moving bits about which representations Nadia gets back to somewhere else. Propose to elevate 3.4.1 to 3.4 and shrink rationale to message semantics.
Action IJ: Edit 3.4 and 3.4.1 to this effect.
IanJ
"Agents MUST NOT ignore authoritative metadata without the consent of the user."
RF: Should say agents must not override metadata.
IJ: Having had our previous discussion, should this GPN be about deviating from protocols?
RF: That should be ok.

Section 3.5. Safe Interactions

IanJ
New note: "Note: In this context, the word "unsafe" does not mean "dangerous"; the term "safe" is used in section 9.1.1 of [ RFC2616 ] and "unsafe" is the natural opposite."
MJ: Perhaps an additional sentence about "safe" here means "side-effect free".
TBL: I think we arrived at "does not incur an obligation" rather than talking about side effects due to things going on under the hood.
Chris
it also means that 'the TAG says filling in a form is unsafe' which is true in our context but not a message we want to send
IanJ
RF: "Safe" does not *necessarily* mean dangerous. For a long discussion on this topic, cf. Nancy Leveson.
IJ: I may edit the note further along the same lines.
Review of "It is a breakdown of the Web architecture if agents cannot use URIs to reconstruct a "paper trail" of transactions, transaction results, i.e., to refer to receipts and other evidence of accepting an obligation."
RF: Should be about transaction requests.

Section 3.5.1. Unsafe Interactions and Accountability

IanJ
RF: Say that the results of the transaction are a valuable resource, and like all valuable resources, it would be nice to have an identifier for it. It would be nice if one of the effects of interactions were to populate the Web with new useful resources.
TBL: In the story, s/commitment/commitment to fly her/
CL: are we sure that it's "not widely deployed"?
TBL: Two separate issues - message out and message back.
Action IJ: Edit 3.5.1 to say:
  1. Useful to have a URI for requests and results
  2. Here are ways to do that in HTTP.
RF: Note that a reference to a transaction result/request is not useful for legal purposes; you need a copy.
IJ: You could do a checksum test...
CL: Why talk about "not widely deployed' in the story? It's worth saying that content location will work in deployed user agents.
Chris
so servers should provide it
IanJ
MJ: Right, talk about the representation provider (not server software).

Section 3.6. Representation Management

IanJ
What about "resource is unpredictable"?
[Discussion of unreliable v. unpredictable.]
CL: Use both terms: unreliable or unpredictable. (CL one could provide false information daily and be very predictable about it)

Section 3.6.1. Representation availability

IanJ
TBL: Say in 3.6.1 that "there is a benefit to the community in providing representations."

Section 3.6.2, issue klyne15

Issue klyne15

IanJ
s/authorities servicing URI/URI owner/
Persistence of a URI is a matter of policy and commitment on the part of the URI owner.
We look at text in 2.2: "This document does not address how the benefits and responsibilities of URI ownership may be delegated to other parties (e.g., to individuals managing an HTTP server)."
TBL: This isn't sufficient to answer klyne15. In 2.2, change to "(for example, to a server manager or someone who has been delegated part of the URI space on a given Web server).
IJ: I'll do that.
Resolved: Close klyne15 as follows:
  1. Include TBL's text in section 2.2
  2. s/authorities servicing URIs/URI owners

Section 3.7. Future Directions for Interaction

IanJ
Proposed: Delete freenet and mldonkey from 3.7, as well as their references. Leave in NNTP.
DanC
prolly ok... lemme look at 3.7.
Norm
I think you wanted those references, DanC_
DanC
any particular reason to delete them?
IanJ
I don't hear much support for them. Tim Bray wanted them gone, as I recall.
DanC
I don't think NNTP is representative of the sort of p2p developments that are coming down the pike
IanJ
I think we haven't discussed peer-to-peer at all. To include them here seems very premature.
DanC
we have discussed them...
Norm
DanC_, do you object to removing the parenthetical list of peer-to-peer systems?
IanJ
Or just leaving "peer-to-peer systems"
DanC
Can you do a quick straw poll, norm? am I the only one who wants them in there? I.e. would the TAG prefer to appear ignorant of p2p developments? Why leave "peer-to-peer systems" without a reference?
IanJ
Proposed: Delete "(including Freenet, MLdonkey, and NNTP [RFC977])"
Chris
this tag member is insufficiently informed to make an informed decision
IanJ
RF: Not sure that there are even future directions of p2p systems.
[Discussion here of whether NNTP is p2p]
Chris
RF says NNTP is not peer to peer, but client server
IanJ
TBL: Servers talk to each other.
DanC
No, I don't object, norm. I abstain. I'm a little surprised, given our earlier discussions, but oh well.
IanJ
TBL: NNTP is peer-to-peer since no centralization.
IanJ
CL: I support acking p2p, but don't know enough about specific examples of systems.
[Nobody seems to care much]
RF: It just adds noise.
PC: I tend to agree that it's noise.
DanC
(reviewing records, my action to produce these citations came from Action DC 2003/07/21 ...)
IanJ
Proposed: peer-to-peer systems "(including NNTP [RFC977])"
Proposed: "peer-to-peer systems" [Delete parenthetical]
mario
+1
IanJ
Dan, no objections here to: "peer-to-peer systems" [Delete parenthetical]"
DanC
hmm... the 2003/07/21 discussion was about future directions for identification. mldonkey has a URI scheme that doesn't depend on dns.
IanJ
NW: Any objections to "peer-to-peer systems" [Delete parenthetical]"
DanC
I intended the freenet/mldonkey citations to go under 2.7. Future Directions for Identifiers... do we acknowledge the weakness around DNS?
IanJ
Yes, we do, in section 2.3: "One consequence of this approach is the Web's heavy reliance on the central DNS registry."
DanC
no objection here, nw
IanJ
Resolved: change text in 2.7 to "peer-to-peer systems"

Section 4

Chris
http://lists.w3.org/Archives/Public/www-tag/2004May/0024.html
mario
Change order to "XHTML, RDF/XML, SMIL, XLink, CSS and PNG". This would be more consistent with the following section.

Last Call issues (continued)

Issues discussed here:

See also continuation of discussion of Last Call issues.

Issue kopecky3

Issue kopecky3

IanJ
"Thus, before inventing a new data format, designers should carefully consider re-using one that is already available."
TBL: You could dispatch on namespace, for example.
From 4.5.8: When the media type assigned to representation data is "application/xml", there are no semantics defined for fragment identifiers, and authors should not make use of fragment identifiers in such data.
RF: There's a visibility argument.
Chris
Visibility is higher if the media type is different (visible in the message metadata) as opposed to having to parse and understand the content, looking for xmlns declarations
IanJ
PC: Does RFC 3023 answer kopecky3?

We review RFC 3023 section A.1: "Why not just use text/xml or application/xml and let the XML processor dispatch to the correct application based on the referenced DTD?".

TBL: At the moment, there's lots of software out there that dispatches on mime types, so you need specific mime types. When systems start dispatching on namespaces, that situation may change. Remember that we think that xml docs should have namespaces.
MJ: Change title of http://www.w3.org/2001/tag/2004/webarch-20040510/#no-text-xml (since this is bad practice).
Proposed::
  1. We prefer to restate what RFC 3023 says in A.1: Because in many cases deployed software dispatches processing based on media type, it is important to provide a specific media type rather than a general in order to dispatch to a specific application.
TBL: In a context where content is intended for presentation for the user, and accompanied by sufficient style information, it is appropriate to serve it as application/xml.
timbl
Where a resource is deigned for presentation to a human being, and sufficient style information is referenced to allow that presentation, then it is appropriate to serve the content as application/xml.
IanJ
CL: Say instead - when the content is document-like and when it's suitable for box-model presentation...
IJ: What about blind users?
timbl
Where a resource is deigned for presentation to a small pink mouse this does not apply.
IanJ
TBL: I'd like to fall back on "sufficient style information". Box model is limiting. Given content and style sheets, there's enough available to make the presentation usable.
RF: What about if you want someone to be able to process some xml content as xml (and it includes links to style sheets).
Action IJ/CL: Draft a proposal to address kopecky3.

Issue schema13

Issue schema13

IanJ
CL: Summarizing - reviewer feels text does not communicate why extensibility may not be appropriate in some cases. E.g., high-security environment. I agree with reviewer; need to indicate that there are also drawbacks.
IanJ
NW: Add a para to 4.2.2 about trade-offs.
Chris
http://lists.w3.org/Archives/Public/www-tag/2004May/0024.html
IanJ
TAG agrees with reviewer for that part of schema13.
Action NW: Propose text on tradeoffs for section 4.2.2.
IanJ
"For computer-to-computer mission-critical applications, "fallback behavior" is semantically equivalent to "silently handling errors" and the Web architecture document is thus self-contradictory."
TBL: In general, useful to look at modularity on the inside but also on the outside (i.e., where one's work is itself a module).
Relevant text: "Additional strategies include prompting the user for more input, automatically retrieving data from available links, and falling back to default behavior."
IJ: On their third point (extensibility in protocols as well), we do talk about it in 1.2.2.
Chris
1.2.2. covers it
IanJ
Proposed: Delete "falling back to default behavior"
Chris
agree with proposal
IanJ
PC: Does finding talk about fallback behavior as 4.2.3 states?
IJ: Delete "and falling back to default behavior". This was a mis-transcription of what was in the finding and doesn't apply here.
Resolved: Delete "falling back to default behavior"
IJ: Should we talk about extensibility in protocols?

RF: Mandatory extensions was added to HTTP. Turns out nobody wanted to use due to complexity. The desire is still there; there just needs to be a better trade-off.

TBL: Tell the following story:
  1. HTTP as designed had optional extensions but no mandatory extensions.
  2. Attempts to add mandatory extensions to HTTP were not successful.
Second story:
  1. A protocol defines a series of messages.
  2. New protocols can be defined through new messages.
  3. When defining new protocols, message extensibility techniques may be applicable.
  4. Thus, when you develop a SOAP system, better to use documents than RPCs.
Action IJ: Insert a story in the section on extensibility about protocol extensibility.

4. 14 May Minutes

See also open actions by owner and open issues.

Present: NW, MJ, IJ, CL, PC, TBL.

4.1 Last Call Issues (continued)

See also continuation of discussion of Last Call issues.

Issue diwg4

Issue diwg4

[IanJ]

Resuming review of section 4
http://lists.w3.org/Archives/Public/www-tag/2004May/0024.html
DIWG suggests more discussion of limitations of using media type mechanism (e.g., text/html doesn't tell you whether HTML 3.2 or HTML 4)
CL: In qs factors, when you are deciding which format to send, difficult to adjust on a per-resource basis. We should probably say that the media type mechanism doesn't solve all problems.
PC: Where would such text go? Section on versioning? Section on data formats? Want to perhaps say something about this, but without creating new issues for ourselves.
[Chris]
suggest point about versioning in versioning; point on qs values in interactions section
[IanJ]
TBL: I think we should lead the way, even if we can't implement today. At the moment, people don't put version information in media types. We could say:
  1. Use parameters in new media type definitions.
  2. Servers need to support this.
  3. In the section about representation management, tell server managers to allow fine granularity (per-representation).
[per-representation specification of parameterized media type]
TBL: Should we encourage people to use quality factors?
[Discussion of usability issue of specifying per-representation media types and parameters; authoring tool support; server support...]
CL: Many users have no idea about what version of, say, CGM they are using. Need tool support.
[Chris]
there is no standardization so the authoring tools cannot help, so in practice the simpler the better
[IanJ]
PC: Should we open an issue on this and postpone changes to v1?
[DanC_]
DanC_ has changed the topic to: 14May http://www.w3.org/2001/tag/ going over updates to section 4
[IanJ]
Resolved: accept new issue mediaTypeManagement-45: "What is the appropriate level of granularity of the media type mechanism."
[DanC_]
mime type versioning is hairy. I'm very wary of "leading the way", which is really: waving our hands without sufficient implementation experience.
ah; yes, good to make it an issue. I concur.
[DanC_]
connect it to uriMediaType-9, pls, ian.
[IanJ]
Action CL: Write draft finding on this issue.
Action IJ: Add a bit of text to the document about media type limitations, versioning, and link to issue.

Issue msm14

Issue msm14

[Chris]
Issue msm14: WD-webarch-20031209, Section 4.2.2, Story: Allowing extra
attributes does change the conformance of existing data
Not really, given that in the story they deliberately chose that
extensibility policy. The point is valid in general but not for this
particular story.
[IanJ]
CL: I don't think the document needs to change; the claim is true in the document because we are explaining our story.
Resolved: Agree with the general case but it doesn't apply to this specific instance. The TAG does not feel changes are required.

Issue manola29

Issue manola29

[Chris]
Question: What's a language instance?

instance - see xml spec for meaning of instance

[IanJ]
NW: Proposed to revise language to avoid the word "instance".
IJ: Proposed "in content".
Currently reads: "A format specification SHOULD provide for version information in language instances."
PC Proposed: "A format specification SHOULD provide for version information."
IJ Proposed: "A format specification SHOULD allow content authors to identify format version."
PC Proposed: "A format specification SHOULD allow content authors to identify the format version."
[Norm]
A format spec SHOULD allow content authors to identify the version of format being used"?
[IanJ]
Resolved to accept this change: A format specification SHOULD provide for version information.
[Thus closeth manola29]

Issue schema14

Stuart joins by telephone. Dan joins by iChat and IRC.

Issue schema14

[IanJ]
PC: Note that some of these changes in the spec might lead to some updates in findings.
CL: A format specification SHOULD allow content authors to indicate the extensibility policy used This is one step further than the reviewer requested.
PC: Your change is stronger than what the reviewer requests.
[DanC_imac]
(I conclude this is pretty complete access to the meeting. this could save a lot of plane trips ;-)
[IanJ]
CL: Reviewer asked for indication of trade-offs; I proposed one way.
PC: HTML ("liberal in what you accept") had policy in the documentation, not a schema for example.
CL: We may have oversimplified when moving bits of finding to arch doc; we could respond by saying "see finding for more details".
[DanC_imac]
(are we still going over IJ's 10May changes? how did we get to schema14? I don't see it in webArchWithProposals)

Issue manola30

Issue manola30

[Chris]
issue manola30: Difference between "setting expectations" and "specifying"?
Yes, its wooly language. All instances of "set expectations" could be changed to "specify", I think.
[IanJ]
Action NW: Write some text to address schema14 (e.g., by reviewing finding).
CL suggests to s/set expectations/specify
IJ Proposal: Delete: "As part of defining an extensibility mechanism, a specification should set expectations about agent behavior in the face of unrecognized extensions." since redundant.
CL: Look for other instances of "set expectations"; consider changing them.
[DanC_imac]
manola30 is editorial, for my money.
[IanJ]
Resolved: Delete "As part of defining ..." sentence.

Issue schema15

Issue schema15

Roy joins meeting. DanC leaves meeting.

[Chris]
Yes; regardless of what MIFFY/MTOM might do, a quick s/JPEG/SVG would make the statement true.
[IanJ]
schema15 already closed as suggested by reviewer in 10 May draft.

Issue klyne19

Issue klyne19

[Chris]
issue klyne19: Unclear statement about mixing RDF vocabularies "RDF allows well-defined mixing of vocabularies, and allows text and
XML to be used as a data type values within a statement having clearly defined semantics. I couldn't figure precisely what this was trying to say."
Me either. (Either 'as data type values' or 'as a data type value', too). But see the next issue. "used as a data type values" is grammatically incorrect anyhow.
[IanJ]
[We review in context of 4.2.4]
In particular " and allows text and XML to be used as a data type values within a statement having clearly defined semantics."
TBL: Two points here:
  1. You can mix RDF ontologies; mixing is well-defined for RDF.
  2. You can remove RDF statements and preserve the truth value of the set of assertions.
[timbl]
You can add stuff in nd somebody else can strip it out. You can remove statements *because* the semantics of the RDF graph is defined to be the conjunction of the semantics of the statements.
[IanJ]
PC: Add a reference to an RDF specification?
[timbl]
RDF allows the mixing of information in the vocabularies from different application, with well-defined semantics.
RDF allows the mixing of information in the vocabularies from different application, with well-defined semantics. [ See RDFMS1.0]
[IanJ]
The semantics of combining RDF documents (i.e., sets of assertions) is well-defined.
The semantics of combining RDF assertions from multiple vocabularies is well-defined.
The semantics of combining RDF statements from multiple vocabularies is well-defined.
The semantics of combining RDF documents with multiple vocabularies is well-defined.
Resolved: Change third bullet to: "The semantics of combining RDF documents with multiple vocabularies is well-defined."
Resolved: Delete "and allows text and XML to be used as a data type values within a statement having clearly defined semantics."

Issue manola31

Issue manola31

Resolved: Closed by virtue of text being deleted.

Issue klyne20

Issue klyne20

[Chris]
issue klyne20: "Say something about relationship between Hypertext Web and Semantic Web?"

CL: So the SW uses the 'Hypertext Web' or parts thereof to make statements about. But it also makes a statement about cars, and things, so that boiled down to 'the semantic Web makes statements' or 'the Semantic Web is about semantics' (tautology) or 'the non-SW has no semantics' (not true either). So okay say something, but what? SW is at a different level of abstraction, layered on top?

[IanJ]
TBL: The simple answer to klyne 20 is "Yes."
IJ: Yesterday we came close to a definition of "information space" (related to information resources).
[Chris]
information space is the network effect applied to information resources
[TBL goes to white board; see image from yesterday]
TBL: There are at least two subsets of the information space: hypertext web and semantic web. There can be others.
RF: How does a transforming proxy fit into TBL's model? The transforming proxy is taking the universe of information and transforming it.
TBL: In case of gateway, the gateway provides an annotated view of resources (including those outside the gateway). There are services that, in effect, provide mirrors of the hypertext space.
NW: Maybe we need a new section at top of document?
IJ: What about a section on information space?
PC: I thought we had agreed that we were going to push discussion of the Semantic Web to another version of the arch doc.
TBL: We don't have to "go into the SW bubble" in v1 of the arch doc. We can put one liners about both sem web and web services to state where they are (in the diagram). In short, explain how they relate, but don't describe their architecture.
[Chris]
CL reiterates what the reviewer said originally: So okay say something, but what? SW is at a different level of abstraction, layered on top?
[Stuart]
Hmmm... does application/xml+svg create a new application? I don't see at an architectural level you distinguish media-types that do/do-not give rise to new 'applications'. Also... are applications an 'architectural' concept that we talk about anywhere else?
[IanJ]
NW: I think the way to address klyne20 is to say that this document does not attempt to describe the relationship between these systems in detail.
TBL: 3.7 is currently wrong and misleading. Remove the semantic Web because that's not about interaction. Sem web uses the same interaction mode (e.g., GET).

Resolved to accept this proposal (SW abstaining):

  1. Create a new section 4.6 about use of media types to build new applications that make use of the information space. Explain that you create new "applications" such as the semantic web through new media type definitions.

    "Above we describe a global hypertext application which has been defined using specific Media types. The creation of other media types allows new applications to be build in the same space, using the same information space infrastructure. The semantic web is one such application, which is not described in detail in this version of this document. We expect future versions of the Arch doc to describe the semantic web in more detail."

  2. Remove sem web from discussion in 3.7
  3. Respond to reviewer that we expect future versions of the arch doc to go into more detail about the relationships among various systems.

Issue clark6

Issue clark6

[Chris]
issue clark6: Separating Presentation From Content. "This is often harder than the AWWW lets on, and sometimes it's simply not possible at all. I think the language should be modulated to reflect that reality; see proposed text..."
[IanJ]
CL: This is another instance of oversimplification; needs more discourse on trade-offs. I agree with reviewer.
[Chris]
Absolutely. Again the finding conveys this trade off but WebArch over simplifies and presents as a single choice without trade-offs.
[IanJ]
Action CL: Proposed additional text that includes more about tradeoffs.

Issue gilman1

Issue gilman1

[Chris]
issue gilman1: 'legal requirement' as justification for 'particular presentation' misses 'leading Web to highest' mark This is another occurrence of meta-issue-1 "its a trade off, not clear-cut".
[IanJ]
IJ: The new text is "Designers should consider appropriate technologies for limiting the audience. For instance digital signature technology, access control, and other technologies are appropriate for controlling access to content."
Resolved: We believe that 10 May draft addresses gilman1.

Issue klyne21

Issue klyne21
[Chris]
issue klyne21: "Add statement about scalability concerns." Good point, scalability in for example client side vs server side image maps. Pushing computation to the client increases server throughput.
[IanJ]
PC: Stored procedures on databases - scalability by pushing computations to server. This is about light v. heavy clients. This is clearly a time when one considers tradeoffs. Largest increases in scalability of database applications has involved putting triggers on the server; stored procedures (which trigger a bunch of actions), etc. Scalability of SQL databases is about moving computation to the server.
[Chris]
and cutting down communication
[IanJ]
PC: Security architecture scales better as well when procedures stored on server.
CL: The example I was thinking of was moving server-side images to client-side images. Improved load on server and network lag.
PC: Counter-example - there are cases where you want to filter a view of data, and you should filter on the server side in order to ship less data.
Action CL: Draft a little text to explain that there's a tradeoff in this situation.

Issue clark4a

Issue clark4a

[Chris]
issue clark4a: "Hypertext Good Practice Redundancies". No, these are not equivalent. There have been hypertext systems that allowed only linking within a small set of documents, and SGML allowed within-document linking (id/idref) without actually being a hypertext system. The second good practice note is a scalability note, and is needed. However, it should be broadened not only to hypertext links but other sorts of links as well.
[IanJ]
CL: I think the two are not redundant. First motivates hyperlinks, second motivates global hyperlinking. idref should be changed to anyuri.
NW: idref should be changed to anyuri
Resolved: We believe that the statements say different things and both are justified.

4.2 Review of TBL's critique of .mobi TLD proposal (continued)

"New top level domains considered harmful" from TBL

[IanJ]
Proposed: That the TAG send TBL's revised TLD text to the ICANN comments list.
SW summarizing what he thinks: There are some potential benefits and potential harms to the proposal. I do not think there is an essential need to create a TLD to provide the potential benefits. The way to give clients with small devices more tailored content is not through TLDs. I have not seen technical arguments that support the need to create a TLD.
[Roy]
I have read and support Tim's TLD document.
[timbl]
DRAFT: This document comes with the support of the W3C TAG with Norm Walsh of Sun Microsystems abstaining, and Paul Cotton of Microsoft recusing himself.
[DanC_]
(I haven't looked at /TLD.html very carefully, but I'm satisfied that it's gotten enough review and endorsement that I should have spoken up if there was stuff I objected to. If the TAG records "concur" votes, then that's what I'd like the record to show. Otherwise, show me as not objecting nor abstaining, i.e. in support)
[IanJ]
Resolved: The TAG supports sending TBL's revised TLD text to the ICANN comments list. Support from CL, MJ, RF, TBL, SW. DC concurs. NW abstains. PC recuses himself (citing charter).
Action TBL: Send his text to the ICANN comments list today.
Action TBL: Send the link to his posting to www-tag.

4.3 TAG rechartering and patent policy

Two questions re: patent policy:

  1. Should TAG Recs be covered by the policy?
  2. What should the licensing obligation of participants be?

See draft charter.

[DanC_]
hmm... "TAG findings have no normative status and must not be referenced normatively from a W3C Recommendation Track document."
[IanJ]
PC: IJ sent out proposed text; IJ responded to them and sent refinements.
TBL: I propose that some folks from the AB, the TAG, and DJW hold a discussion on the TAG and the patent policy.
Would prefer to be counted as an invited expert: SW, PC, RF, MJ, NW.
(RF: I think PC should be an invited expert)
PC: Note that if Day Software were to join, RF's obligation would change.
IJ: I note the same is true for any WG.
PC: The difference with the TAG is the election process for participation.
[Discussion postponed to another meeting]

4.4 Review of TAG's general issues list (continued)

List of actions by owner

Issues with actions discussed here:

4.4.1 Issue uriMediaType-9

Issue uriMediaType-9

Propose CL's three changes to registration process to Ned Freed.
CL: Ned has not responded.
[Chris]
My action is completed, though a new action could be assigned for followup
[IanJ]
Proposed to close CL's action on uriMediaType-9
CL: My charmod review actions are all done.
Resolved: Close charmod review actions for CL.

4.4.2 Issue xlinkScope-23

Issue xlinkScope-23
Resolved: Close CL's action regarding xlinkScope-23.

4.4.3 Issue contentPresentation-26

Issue contentPresentation-26

Action CL: Talk with others about aspects of this finding and revise it.

Resolved: Close CL's action for contentPresentation-26

4.4.4 namespaceDocument-8

Issue namespaceDocument-8
Action PC: Prepare finding to answer this issue, pointing to the RDDL Note. See comments from Paul regarding TB theses. Per 23 Feb 2004 teleconf, modified into an action to produce a bulleted list of points.
PC: Still pending

4.4.6 URIGoodPractice-40

Issue URIGoodPractice-40
Action RF: Draft a finding for this issue.
RF: No progress other than the URI spec.

4.4.7 URIEquivalence-15

Issue URIEquivalence-15

Action SW: Track RFC2396bis where Tim Bray text has been integrated. Comment within the IETF process.

SW: I am reviewing this within the week.

4.4.8 metadataInURI-31

Issue metadataInURI-31

Action SW: Produce a revision of this finding based on Vancouver ftf meeting discussion.
SW: I hope to have by mid-June.

4.4.9 IRIEverywhere-27

Issue IRIEverywhere-27

Action TBL: Explain how existing specifications that handle IRIs are inconsistent. TBL draft not yet available on www-tag.
TBL: Please continue.

4.4.10 RDFinXHTML-35

Issue RDFinXHTML-35
Action TBL State the issue with a reference to XML Core work. See email from TimBL capturing some of the issues.
TBL: Lots has happened since then. I think that GRDDL changes that landscape.

See email from DC on Syntax and Semantics for embedding RDF in XHTML.

Action IJ: Add GRDDL to background list in issues list.
[IanJ]
IJ: Who is working on this?
[timbl]
http://www.w3.org/2002/04/htmlrdf
[IanJ]
[Stuart leaves]
Action IJ: Add this to background materials - http://www.w3.org/2002/04/htmlrdf

Resolved to drop TBL action regarding stating this issue w.r.t. XML Core work.

PC: When we report to the AC we should explicitly tell them that we are preferring to close LC issues in preference to closing general issues.

4.4.11 XMLVersioning-41

Issue XMLVersioning-41

Action TBL: Suggest changes to section about extensibility related to "when to tunnel".
Resolved to drop this action.

4.5 Next draft, meeting planning

4.5.1 Next Editor's Draft, Next WD

[IanJ]
DC: Likely to go back to WD status. I'd like to publish a new TR page WD soon.
[Zakim]
DanC_, you wanted to note we owe a /TR/ publication. it's been >5 months.
[IanJ]
DC: IJ has feedback queued up on 10 May draft. I'd like to see a draft in the next 3-4 weeks that at least IJ and two others have looked at.
[Discussion of next Editor's draft publication date - first week of June]
IJ: I'd shoot for 8 June.
Regrets for 14 June: IJ, TBL, MJ. At risk: RF
Support for:
  1. Next editor's draft 8 June
  2. At 14 June teleconf, if at least 2 thumbs up and no thumbs down, IJ will request publication thereafter.

Commit to reading the 8 June Editor's Draft: NW, CL, PC

4.5.2 Meeting planning

Next scheduled ftf meeting: 9-11 August in Ottawa. IJ unlikely to be in Ottawa but might be able to attend part-time by phone or IRC.

[IanJ]
NW: I heard people say "do LC comments until we're done"
[DanC_]
PC: looks like a LC will be Sep at earliest, following our Aug ftf. DC: sigh. true.
[DanC_]
rather: (eek! minute taking and issue tracking!)

4.6 Last Call Issues (continued)

Issues discussed here:

See also continuation of discussion of Last Call issues.

Issue schema17

Issue schema17

[IanJ]
CL: I think their point was that a ns URI is not a globally unique name.
DC: I disagree.
[Norm]
http://lists.w3.org/Archives/Public/public-webarch-comments/2004JanMar/0005.html
[Roy]
Suggest: a mechanism that may be used to establish globally-unique names.
[IanJ]
CL: I think the argument is "You can pick http://normwalsh.org/foo" as a ns URI, and so can I.
[Zakim]
DanC_, you wanted to recall a comment that {uri, lname} may refer to more than one thing
[IanJ]
NW: It's important to point out that it's true that the ns spec lets you to create globally unique names, no matter how they are used.
DC: The comment I heard was that "globally unique" suggested to someone that it could only be bound to one thing. There is an architecture of global uniqueness and then there are bugs. The architecture assumes that the domain owner is competent. Sometimes they are and sometimes they are not.
[Roy]
instead: "a mechanism that reduces the likelihood of name collisions."
[DanC_]
I think s/that can be understood in any context// is responsive to the commenter. let's try that and see if they're satisfied.
[IanJ]
or: "reduces the risk of URI overloading"
[DanC_]
there should be a link to 2.2. URI Ownership from this section somehow...
[IanJ]
RF: Namespaces in xml use URIs in order to obtain the properties of a global namespace.
[DanC_]
perhaps "... perhaps of a global namespace discussed in 2.2. URI Ownership"
[IanJ]

Resolved:

  1. Continue to delete "that can be understood in any context"
  2. Modify remaining sentence to say: "Namespaces in xml use URIs in order to obtain the properties of a global namespace."
  3. Include a reference to 2.2 URI ownership.

Issue schema18

Issue schema18

About 4.5.3...
"The xsi:type attribute..."
CL: I prefer "type attribute from XML Schema"
MJ: Also, state the namespace that it's in.
[Norm]
The type attribute from the W3C XML Schema instance namespace, "http://...".
[DanC_]
[[
"The type attribute from W3C XML Schema namespace is an example of a global
attribute."
]] is one of their suggestions
... and is OK by me
[IanJ]
IJ: And a ref to the schema spec.
[IanJ]
Resolved: IJ to make changes to point to XML Schema spec (possibly with ns URI in text).

Issue schema19

Resolved: The TAG agreed with the reviewer about issue schema19

Issue booth3

Resolved: The TAG believes that with the 10 May draft it has satisfied the reviewer for issue booth3

Editorial in section 2.7.2

[DanC_]
"Nadia is requesting the namespace document." -- http://www.w3.org/2001/tag/2003/lc1209/webarchWithProposals.html
[IanJ]
2.7.2. Assertion that Two URIs Identify the Same Resource
[DanC_]
editorial: "owl:sameAs" vs sameAs. Pick one.
[IanJ]
NW: Replace ns prefixes with clear references.
DC: Pick one of the above tokens: "owl:sameAs" vs sameAs

Issue kopecky6

IJ: I added one sentence.
DC: Satisfied

Resolved: The TAG accepted the change in the 10 May 2004 Editor's Draft for issue kopecky6.

Issue schema20

Issue schema20

Added as well: "In practice, applications may have independent means of specifying ID-ness as provided for and specified in the XPointer specification."
NW: Change to "such as the means provide for..." I'm concerned that bullet 4 is incorrect. It does not establish ID-ness; it allows you to point to things that are not identified by ID.
[DanC_]
[[
In practice, applications may have independent means of specifying IDness as
provided for and specified in XPointer. XPointer carefully
discusses these options at
http://www.w3.org/TR/2003/REC-xptr-framework-20030325/#shorthand
]]
[Norm]
In practice, applications may have independent means of locating identifiers inside a document such as provided for and specified in the XPointer specification.
[IanJ]
TBL: Do we have any examples of this?
[Chris]
http://www.w3.org/TR/2003/REC-xptr-framework-20030325/#shorthand
[DanC_]
(I'm writing them to ask where their implementation experience for this feature is recorded: see email externally-determined ID implementation experience?)
[Chris]
its a pointer cursor not a hand cursor
[IanJ]
DC: +1 to NW's text; please refer to section 3.2 of XPointer.
Resolved: For the fourth bullet, adopt NW's proposal as amended to include ref to 3.2. The TAG believes it has closed schema20 with these changes.

Issue stickler3

Issue stickler3

[marioj]
originally occurred @ 3.1. Using a URI to Access a Resource
[IanJ]
Current definition: "Dereference a URI: Access the resource identified by the URI."
Patrick recommends: "Indirectly access the resource identified by the URI via a representation of that resource."

Resolved: Keep the text in the glossary.

Action IJ: Ask the commenter if the definition in context (in section 3.1) explains the way we use the terms to his satisfaction.

[Roy]
rfc2396bis: URI "resolution" is the process of determining an access mechanism and the appropriate parameters necessary to dereference a URI; such resolution may require several iterations. To use that access mechanism to perform an action on the URI's resource is to "dereference" the URI.

Issue diwg1

Issue diwg1

[IanJ]
RF: IMO, the commenter misunderstands the concept of "URI persistence".
Action IJ: Ask the reviewer to clarify the question.
TBL: I think this is more about .html v. .asp (for example).

Issues clark1a and clark1b

Issue clark1a, Issue clark1b

[IanJ]

RF: One indirection always suffices.

TBL: Reviewer only thinking about hypertext; no constraint to just hypertext. I do note that "secondary resource" is not a subclass of "resource". "Secondary" is a relationship ("something referred to by the primary resource"). In fact it is a "referenced resource".

RF: Rather, it's "defined by". The secondary resource is "defined by" the primary resource.
TBL: Note that "a#b" and "c" can identify the same resource; "b" is just a local name within the context of the representation of a.
RF: Within the global set of resources, there are subsets (primary/secondary) which may be the same sets or overlap. I don't know where to place resources not yet assigned a URI.]
IJ Proposed: "resource" and "resource with a local name"
[That is an attempt to move towards naming and to avoid thinking about subclassing resources]
TBL: You can never establish whether something is a secondary resource because you don't know whether someone has created a URI that designates it.
RF: I do not expect to change the terms in RFC2396.
DC: Best to describe secondary resources as a relationship between the two resources.
[Norm]
Can someone post the URI for the most recent 2396bis?
[IanJ]
PC: Explain to the reviewer that the definitions of primary/secondary are in RFC2396. Include a reference to the RFC2396bis for info about the definitions.
[DanC_]
(just because a term is defined elsewhere doesn't make our reviewers happy about it. to wit: Hayes and 'resource')
[Norm]
(point taken)
[IanJ]
TBL: The response to the reviewer can say (at least) that we mean "Just a relationship."
Resolved for issue clark1a:
  1. No change to document other than inclusion of a ref. to RFC2396.
  2. Respond to reviewer indicating this and answering "(1)".
[timbl]
From clark1b: "Let's assume however, that these are mutually incompatible representational semantics. The question arises then, which of the several, mutually incompatible secondary resources is identified by a primary resource with a fragment identifier? In other words, the answer to the claim that "the secondary resource identified by a URI with a fragment identifier is expected to be the same across all representations" is yes, but which one?"
[IanJ]
[Sentiment that new text will address clark1b as well.]
PC: Ask commenter at next draft to review the revised text.

More on clark1b.

TBL: In the case of, say, SVG and HTML. You can create named anchors that correspond - and consistency from the point of view of the publisher.

[IanJ]
TBL: I propose a new issue.
DC: I think the long discussion we had yesterday suffices.
TBL: There's a discrepancy between what you can do with an html anchor and an rdf anchor. There is a real need to be able to write a resource and to provide both html and rdf representations.
[DanC_]
this is issue http://www.w3.org/2001/tag/issues.html#fragmentInXML-28 , no?
[IanJ]
TBL: There are cases where, e.g., HTML is generated from the RDF. In that case, if someone asks for a representation and gets back html instead of RDF, and for every concept named in the RDF representation, there is an anchor in the HTML document, it's reasonable for coneg to work (from RDF-> HTML). There are pitfalls but also a need. The answer to "Can RDF be consistent with HTML?" is "That's a long story." Is the move from "reality" to a "description of reality" a reasonable downgrading of information that a publisher could be happy with sufficient to do coneg?
[Zakim]
DanC_, you wanted to ask if fragmentInXML-28 is close enough
[IanJ]
RF: How is an RDF description of a car different from an HTML description?
DC: See fragmentInXML-28.
[IanJ]
PC: Why not have a parameter on a fragment to provide different text on a per-media type basis?
[DanC_]
e.g. foo#if-HTML(sec3);if-RDF(car12)
[IanJ]
RF: There is a reason - makes it easier for people to create implementation-names instead of implementation-independent names. People in general don't do this; they don't create names in different representations that are incompatible, and then use URIs to refer to them.
TBL: Ken Clark is operating in a space where he wants to be able to use both RDF and HTML representations.
[IanJ]
NW: I do this today - you can get either HTML or RDF files back; I assert they are consistent (see example from NW site)
IJ: I believe that I can incorporate what Kendall and TBL want into text that we agreed to yesterday: namely that the architecture licenses publishers to serve both RDF and HTML representations when the publisher believes they are suitably consistent (despite switching semantic "levels"). However, need to proceed with caution when doing this (nothing new!).
[NW's URI is an example of doing this.]

4.7 HTTPRange-14 (continued): TBL text and image

See TBL's electronic version of diagram as well (SVG version, PNG version)

[timbl]
See my proposed text regarding HTTPRange-14
[DanC_]
which part of the document does that text go in, timbl?
[timbl]
I'll leave that to Ian?
[DanC_]
I'd rather you didn't, timbl. I'd much rather you carefully considered where the text goes and how it flows.
[IanJ]
MJ: [For TBL's graphic] Why not enlarge the Web services box to overlap with the Semantic Web ("semantic web services")
CL: Also, not all web services work over http.

4.8 Review of slides for TAG presentation at AC meeting

[DanC_]
See my slides for AC meeting.
[IanJ]
CL: I agree to them.
NW: Please add a Note that the TAG will be focussing on LC comments. Who will present these slides?
DC and PC will split slides.
Resolved: TAG happy to have DC and PC present slides at AC meeting.

4.9 Last Call issues (continued)

Issues discussed here:

Issue clark3

Issue clark3.

Resolvedto refer reviewer to 3.6.2, no change to document.

Issue clark4b

Issue clark4b

[DanC_]
[[ The real problem here is that there is no real formalization of "hypertext link" in the AWWW. ]] er... isn't hypertext link discussed in 4.x?
ah... he's aware of that... [[ But if, as seems likely from Section 4.5.2. ]]
[IanJ]
IJ: I understand the question to mean "why not allow hypertext links in the format even when there is a different paradigm (non hypertext UI)
NW: But this section is explicitly about hypertext.
DC: I think that the reviewer means "Use hypertext links even if you expect machines to be using the data."
NW: We've said "Identify things with URIs". We could also say "Format designers should allow people to create references in formats using URIs."
DC: I think that this is more about: Create a network using URIs.
NW: I suggest pointing reviewer to "A specification SHOULD provide mechanisms that allow Web-wide linking, not just internal document linking."
[Zakim]
DanC_, you wanted to suggest tweaking "Formats that allow content authors to use URIs instead of local identifiers foster the "network effect": the value of these formats grows with the size of the deployed Web."
[timbl]
Chris, what do Web Services use when they don't use http:?
[Chris]
sockets?
[DanC_]
not only do Formats foster the network effects, but also collections of representations that refer to each other.
[IanJ]

Action IJ: Incorporate DC suggested tweak for section 2.

[DanC_]
they use SMTP or other message-passing infrastructures. (MQseries?)
[IanJ]
TBL: We should include some references to materials that define "hypertext".
[timbl]
http://portal.acm.org/citation.cfm?id=175237&coll=portal&dl=ACM&CFID=21545401&CFTOKEN=39367008
[IanJ]
CL: I'd like to nail down a bit more a definition of hypertext link.
[timbl]
The Dexter hypertext reference model
[IanJ]
NW: I'd like to see if we can address the reviewer's comment by virtue of IJ's action to incorporate DC's proposal.
[timbl]
RF: [sotto voce] The web isn't based on dexter

The TAG did not discuss the topics below this line.

5. Other action items


Ian Jacobs for Stuart Williams and TimBL
Last modified: $Date: 2004/05/27 14:59:46 $