Technical Architecture Group Teleconference

03 May 2012

See also: IRC log



Jeni, Larry, Peter, Ashok, Jonathan, Noah
Robin, Yves, Tim
Noah Mendelsohn
Jeni Tennison


<trackbot> Date: 03 May 2012

<scribe> Scribe: Jeni Tennison

<scribe> ScribeNick: JeniT


noah: I am probably not available next week

Ashok: could you create the agenda?

<masinter> i volunteer to chair if Noah can't make it

Ashok: I could then chair if you weren't available

noah: the meeting will be going ahead 10th May

Approve minutes of prior meeting(s)

<noah> http://www.w3.org/2001/tag/2012/04/26-minutes

noah: these are noted as being a draft

masinter: I was using the minutes as a privacy example

noah: Larry, please look at actual text if you want it to be changed

masinter: that text is fine

<masinter> did anyone else care?

<noah> ACTION: Noah to put health warning in "Booth Script" for formatting minutes [recorded in http://www.w3.org/2001/tag/2012/05/03-minutes#action01]

<trackbot> Created ACTION-703 - Put health warning in "Booth Script" for formatting minutes [on Noah Mendelsohn - due 2012-05-10].

RESOLUTION: Minutes of 2012-04-26 -- http://www.w3.org/2001/tag/2012/04/26-minutes -- are approved

<jar> as a use case, the interesting thing is what the health warning should say, to set an example for other health warnings

<jar> based on what I know of TAMI etc. I think it could say a different thing; will follow up in email

Administrative items

noah: we should review the DANE work

<masinter> i've been talking to Philip Hallem-Baker (sp?)

noah: Larry had suggested getting invited experts

masinter: I've been talking to Philip, who has a paper on an alternative method, and is critical of DANE

noah: we'd need someone in addition to him, to be an advocate for DANE, right?
... Larry, would you sort out who should come and talk to us?

masinter: yes

Ashok: I've been following this, and it's still ongoing
... it looks like there's hardly any agreement on direction
... I'm wondering if it's too early to look at this

<masinter> push date of action-697 out a couple more weeks

<jar> action-697 due 2012-05-16

<trackbot> ACTION-697 prepare overview of DANE for TAG consideration due date now 2012-05-16

masinter: it is early to come to conclusions
... but even so, if it's chaos it's not unreasonable for us to see if we could contribute to resolving it

Ashok: then you would ask a couple of people?

masinter: yes, I think so, I haven't yet understood enough to know how to bring it to the TAG
... so let's (Ashok and I) meet this week and see if we can make sense of it together

Ashok: ok

noah: ok, Larry will take next step with ACTION-697

Close pending Actions without discussion

noah: ACTION-636

<masinter> I did move the document I was working on to the WIki

noah: updating product page which closes out Mime and the Web work for now

no objections

agreed noah can close ACTION-636

<noah> ACTION-670?

<trackbot> ACTION-670 -- Noah Mendelsohn to update product priority list to mark MIMEWeb completed after final product page available -- due 2012-05-08 -- OPEN

<trackbot> http://www.w3.org/2001/tag/group/track/actions/670

ACTION-680: review of http://tools.ietf.org/html/draft-ietf-appsawg-media-type-regs-04 (note: should be the -06 version)

<noah> http://tools.ietf.org/html/draft-ietf-appsawg-media-type-regs-06

<masinter> i think -06 made changes in response to our comments

masinter: I think they have responded to our comments
... the changes between -05 and -06 were the edits in response to our suggestions

<noah> Could we briefly list the things they did in response to our requests?

masinter: we need to publish Jeni's document as an RFC or an Architectural Recommendation on best practices for media type registrations

noah: this (media type regs) is a IETF document that will become a RFC or BCP, right?

masinter: yes, right

noah: do we need to do something else?

masinter: we started on best practices to fragment identifiers
... we wanted them to refer to that normatively, but there wasn't time
... but we still need to finish that document

<noah> ACTION-690?

<trackbot> ACTION-690 -- Jeni Tennison to sort next steps on Fragment Identifiers and Mime Types -- due 2012-04-17 -- OPEN

<trackbot> http://www.w3.org/2001/tag/group/track/actions/690

noah: right
... Jeni has the action on what we should do

masinter: my opinion is we should finish it

<masinter> and get it published as an architectural recommendation or else an IETF BCP

<masinter> i also had two other items with regard to media types which aren't currently covered

<masinter> a) file extensions, which *are* a part of web architecture, mentioned in MIME reigistrations, but the registration process isnt currently useful

<masinter> b) 'magic numbers', since sniffing is part of web architecture, but the MIME registry of magic numbers isn't

JeniT: The discussion for item 6 on the agenda is about how to take that forward, and what its scope should be.
... I think at present it's got description, but the "best practices" aren't pitched right.
... I was going to propose to refocus on what should be put into mime type registrations relating especially to fragment identifiers

<Zakim> ht, you wanted to say yes, but don't lose back-story

noah: ok, let's talk about TAG work, and then jump back to mime draft

ht: yes, I think what Jeni volunteered to do would be great
... but don't lose the back story, put it in another document and reference it from the Best Practices document

noah: I agree, keeping the back story is good

<masinter> i would be glad to help

noah: Do we need to set priorities relative your other work, Jeni?

JeniT: How to prioritize relative to copyright and linking

<jar> copyright + linking more important

masinter: they're both important, I'm willing to help on mime draft, but I'll want direction

ht: since Ned won't reference the fragment identifier document, it takes the time pressure off

noah: he's reluctant to reference it because it's not there

<masinter> what's the shortest time we could publish this?

noah: the more we do things slowly, the more we're not good partners
... this feels like something we could get something concrete out rapidly

<masinter> I think the fragment identifier document is nearly done

<ht> I'm OK either way

JeniT: feels to me like we could do something moderately rapidly, good timing with media type reg. document

noah: any objections?

<Ashok> +1

noah: none heard, so do fragid / media type higher priority
... please bring product pages up to date
... especially if I can put it in the report

JeniT: I will try to do that in next couple of days

masinter: do we need to push back on media fragments group?
... at the web conference, I talked to Thomas, and they were saying it was a mistake when the group was chartered
... we have a draft from a WG that the folks from the IETF thought was laughable
... there's a disconnect there

noah: what's the concern with what they've done?

masinter: they propose that fragids have a meaning that applies to all registrations of a top-level type
... and that idea doesn't jibe with the folks who run the media type registries, and what they could enforce

<masinter> it relates to item 7, in that if we expect all future registrations of image/* to use the media fragments fragmentID scheme, the mime registration document should say so

<masinter> at one point, the rtsp scheme wanted to define fragment identifiers for URIs for that scheme

<Zakim> noah, you wanted to talk about XML vs. image

noah: We might have something to say here, which is analogous to the generic processing debate; specifically we could say that there's a choice for the media type registrations:

My point is not to argue strongly for either one, but rather that it would be good if the IETF registration procedures allowed for both, and if appropriate, pointed out the tradeoff (and yes, the example of +xml is at a different level than image/*, but I think the analogy applies, at least up to a point. If the subtypes buy into the base type, generic processing works; if not, not.

ht: Yves' story as Jeni described it is not consistent with the remarks in the current version of Jeni's document
... about the possibility of inheritance from the top
... the fact that Yves doesn't think it means that, and it's clear that it caused an aversion reaction from the IETF community
... I think it's worth keeping in view the possibility of inheritance 'from the top' as in principle a coherent position

noah: it's interesting to talk about this in terms of the major type and the +ext syntax
... certainly there's generic processing you could do on text for example

<noah> http://lists.w3.org/Archives/Public/www-tag/2012May/0015.html

noah: the media fragment spec made some statements about web architecture that I found surprising
... sent in email earlier

<masinter> fragments don't have media types

noah: they said fragments have the same media type as the original representation
... I'd welcome getting comments back on this

masinter: fragments don't have media types

<masinter> fragments aren't "retrieved"

noah: that spec says they do!

<noah> Quoting media types draft: A further requirement put on a URI fragment is that the media type of the

<noah> retrieved fragment should be the same as the media type of the primary

<noah> resource.

<masinter> "retrieved fragment" doesn't have a media type

<ht> Fragments _could_ have media types, if the parent media type registration specifies that they do

noah: I agree with Larry that they don't have a media type
... it might be worth going through the spec to look at these

<masinter> we went through this with http Range

<masinter> byte range retrieval... (need to look up)

JeniT: I could have a look as part of the fragid & media type work

<noah> ACTION-698?

<trackbot> ACTION-698 -- Noah Mendelsohn to schedule discussion of how to take forward the TAG concerns with respect to managing fragment identifer schemes, inheritance and overlap -- due 2012-05-01 -- PENDINGREVIEW

<trackbot> http://www.w3.org/2001/tag/group/track/actions/698

<noah> close ACTION-698

<trackbot> ACTION-698 Schedule discussion of how to take forward the TAG concerns with respect to managing fragment identifer schemes, inheritance and overlap closed

<noah> ACTION-690?

<trackbot> ACTION-690 -- Jeni Tennison to sort next steps on Fragment Identifiers and Mime Types -- due 2012-04-17 -- OPEN

<trackbot> http://www.w3.org/2001/tag/group/track/actions/690

<noah> ACTION-690 Due 2012-05-05

<trackbot> ACTION-690 sort next steps on Fragment Identifiers and Mime Types due date now 2012-05-05

<masinter> http://tools.ietf.org/html/draft-ietf-httpbis-p5-range-19

noah: anything else on the media type registration draft?

masinter: there are two things in my original analysis that aren't within this
... one is file extensions, which aren't part of the web, they're mentioned in several specifications

<noah> What do you think we/they should say about file extensions?

masinter: a recent thing in HTML5 in file upload that talks about file extensions
... the media type registration lists extensions, but the lists don't correspond to what people use
... this is a gap that neither W3C or IETF is working on

<jar> 4.12. Additional Information

<jar> SHOULD File name extension(s) commonly used on one or more platforms to

<jar> indicate that some file contains a given media type.

<jar> (quoting the draft)

noah: do you have a concrete suggestion about what should be done?

masinter: I've been worried about this for years, and I'm trying to see if anyone else on the TAG is interested in pursuing this?

<masinter> that the MIME registry for file extensions is incomplete, not useful, etc. but W3C specs use file extensions, and they're getting *more* used in specs

noah: it's part of the work we just closed out on Mime and the Web

<masinter> another thing in the same category are the 'magic numbers'

jar: is there a registry of file extensions?

masinter: there's no official registry of file extensions in IETF or W3C
... the packaging spec includes file extensions for use within packaged documents

<noah> (and of course, different OS's have at least somewhat different constraints on extensions, e.g. 3 char vs unbounded.)

masinter: we have a best practice that HTTP URIs with file extensions shouldn't mean anything
... the media type registrations should include extensions

jar: would this be an IANA considerations thing? a request to make a registry?

masinter: we could make a registry

ht: there already is a registry, a hugely complex registry
... in linux

<masinter> there are multiple registries, apache filetypes

noah: that's not cross-OS

ht: there's an enormous degree of overlap, except with Windows
... all the unix variants buy into it, including the Macs

noah: Windows has its own, and people hack it

ht: there is a registry, it's just not an IANA one

<masinter> http://filext.com/

<masinter> is a web site dedicated to registering file extension

noah: the IETF has a loose form, in that media type registrations can list extensions
... I just wonder if they might be checked for consistency with others

jar: that's not going to happen

noah: you mean if I gave my new type a suggested file extension of 'jpeg' then they wouldn't object?

masinter: there's a website http://filext.com/

<noah> FILExt is a database of file extensions and the various programs that use them. I

<masinter> i'm just pointing out that this is a part of "how the web works" that isn't standardized, and that disagreements about cause problems

<masinter> and that use from W3C specs is increasing

jar: the way I see these registraties is that the truth comes from IETF and IANA just does administration
... we could under IANA considerations ask them to start keeping a registry
... I'm not saying it's a good idea, just it might logically make sense
... but if the information is in other places, maybe it doesn't

noah: what should we do about this as the TAG?

<masinter> put it on our list of "things we should do later" ?

masinter: maybe we need a list of things that webarch doesn't talk about

noah: what was the other one you had, Larry?

masinter: magic numbers and sniffing
... I'm bringing these up because they are both in the media type reg document as something you could say

jar: they have a SHOULD, you have to justify not saying them

masinter: but it doesn't tell you why you would say it, or what the best practice is

noah: does the TAG need to engage on this?
... couldn't an individual make comments on the draft?
... what do you want them to do, and does it involve the TAG to get there?

masinter: I see people building things that don't work well because of inconsistencies in these areas

noah: but what should we do?

masinter: I'm ok with moving on

<masinter> I don't know what else to do with those

<masinter> At one point in an AC meeting I argued for more "fixing the potholes" vs. "building bridges to nowhere" and that file extensions and magic numbers are potholes in the web architecture

<ht> For the record, http://mx.gw.com/mailman/listinfo/file is the mailing list which manages the magicnumbers/extensions 'registry'---there is no website for this, the closest there is is the README file from the 'file' package, e.g. http://hpux.connect.org.uk/hppd/hpux/Editors/file-5.11/readme.html

<noah> ACTION-680?

<trackbot> ACTION-680 -- Jeni Tennison to lead TAG telcon review (rescheduled) of http://tools.ietf.org/html/draft-ietf-appsawg-media-type-regs-04 -- due 2012-04-24 -- PENDINGREVIEW

<trackbot> http://www.w3.org/2001/tag/group/track/actions/680

<noah> close ACTION-680

<trackbot> ACTION-680 Lead TAG telcon review (rescheduled) of http://tools.ietf.org/html/draft-ietf-appsawg-media-type-regs-04 closed

ISSUE-57 (HttpRedirections-57), ISSUE-63 (metadataArchitecture-63) and ISSUE-14 (HttpRange-14): httpRange-14 — URI Documentation Discovery

<noah> http://www.w3.org/wiki/HTTPURIUseCaseMatrix

<noah> http://www.w3.org/wiki/HTTPURIUseCases

Ashok: Jonathan, I have a small quibble. I'm tempted to look at 3rd row, see 1 fail, and say "ah ah". Then I go to the proposals and see nothing matching the left hand column entry "New status code"

jar: Good point.
... That's an omission. Wasn't formatted as a change proposal, but we did get the proposal. I'll put on my "to do list" to document. The proposal is for something like an HTTP 209 status code meaning "what I'm giving you is a description, not a representation"

<masinter> I'm still talking to jonathan about the scenarios and the underlying assumptions that they seem to make (to me)

noah: have you got feedback on the matrix from the community?

jar: I only really asked the TAG, and haven't really got feedback except from a couple of TAG members

Ashok: could you put in the left column a reference indicating the section in the other document?

jar: yes, I will try
... if there are questions that people on this call have about the matrix, this might be a time to surface them, but email would also work
... I'm not sure how to review this

noah: ideally this should lead to making a proposal
... it might be that not all failures are equally bad for the community
... so you might push forward to narrow discussion to just two or three

jar: we talked about forming a task group, and one has emerged
... Henry, Jeni and I have been exchanging email for the past couple of weeks
... we're working on a narrative for the issue behind the scenes
... I have strong opinions about how to go at this
... I'm trying to get support from Jeni and Henry on these ideas
... they have strong opinions too, so we're going to try to get something between us

noah: the other thing I missed in the matrix, I wondered if there were things that it doesn't cover
... for example, aside from use case J, I was looking for something that would break if there was a new status code
... are all these deployable to some degree?

<masinter> "deployable" has a scope.

jar: the question is who do we give veto power to

noah: what do proxies do if they see a status code they haven't seen before?

jar: I think it's specified in the HTTP spec

noah: doing some due diligence on what running code actually does would be a good idea
... to see if the options are actually viable
... many of these have fallen down when we look at what ISPs can do

jar: I don't know how to do that in the matrix

noah: I'm not saying how to format it, I'm saying do analysis early enough

jar: almost no one will accept a new status code

noah: but why? we need to capture that
... it makes architectural sense to me, you say it fails only one use case

Ashok: browsers would have to implement it, right?

jar: they might treat it the same as a 200
... one of the big complaints about the 303 is you have to do a song and dance around your server configuration, and that's what use case J is saying
... maybe we should talk about the other goal
... we have to think about how to generalise this
... if this is an RDF-specific problem, we need to hand it off to people who care about RDF
... draw some line to say what we think is good practice
... there are some things that are TAG business, and others that are RDF business, and a lot of the proposals look like RDF business
... we've been talking about use case M, HTTP consistency, which seems like a TAG issue
... I have a proposal for how to do the hand off, but I want to wait for Henry and Jeni to agree or disagree about that
... unless we want to talk about this on the call, maybe we should put it out to committee

Ashok: is this really a RDF problem?

jar: it depends on what you think the problem is
... to me, it's a serious issue whether HTTP and RDF diverge
... maybe it's ok if they diverge, but that suggests they should be careful in the language they use
... so there isn't confusion about how web architecture works

Ashok: suppose we took RDF off the table, here
... what would we still have?

masinter: I don't think there would be things left

<masinter> i think this is a 'use of RDF' problem

<masinter> RDF doesn't have a problem, URIs don't have a problem

<masinter> it's people who want to turn "statements in RDF using URIs" into "statements about the real world" who want to take a URI and have it identify something in the real world

<jar> absolutely

<masinter> and the problem is that URIs were never intended as a general mechanism for identifying things in the real world

ht: there definitely would
... there are definitely things left that would matter
... the language that the TAG and TAG members such as TimBL have used for years
... is self-contradictory
... if we care about clearing up a mess that we created, we have to fix that
... and that would be true even if RDF didn't exist

noah: I think on general principles, and given the TAG had this resolution years ago
... I think it would be helpful to explain why it's not our business, if it isn't
... for example, Larry and I didnt immedately agree about what HTTP status codes are used for
... explaining what that answer is is very much in TAG's scope
... especially as our last solution was to use a status code
... I'd like to explain what the limits are of web architecture and their applicability to this problem

<masinter> people building ontologies think an ontology is a mechanism for turning idnetifiers for web resources into identifiers of concepts

<ht> I also don't think we could just profer a new status code as a solution without a story behind it

jar: it would be nice to get any bit of consensus we can on what's TAG business and what isn't

noah: if there are useful back-channel discussions happening, maybe we should hang loose for a while

ht: I think we have to work a bit longer towards convergence
... I'm the major bottleneck, probably, but I'd like another week

<noah> We're hearing that Jonathan, Henry and Jeni are working in the background to get some more clarity. Should we just hold off a couple of weeks and await progress?

Ashok: Henry, you've spoken about the language around content negotiation and so on being self-contradictory
... do you have an idea how to fix that?

ht: no, I don't
... I mentioned it because it's something I use to test proposed solutions
... because I think 'it better solve this problem too'
... the cardinality problem: how many resources are there?

<Zakim> masinter, you wanted to talk about another approach

masinter: I've been trying to pursue a different direction, where we have a mechanism for making statements in RDF
... and an implicit assumption where you can use URIs to ground a statement in the real world
... and the problem is the nature of the grounding is ambiguous
... and we have given conflicting advice, using status codes and so on
... the problems that we really want to talk about wrt publishing, linking, copyright, security, trust
... are not easy to explain in a world where we presume that there is a common understanding about how a URI identifies something
... and if you move to a different model, where you don't think about the truth of statement or their meaning
... as something that's disembodied from the agent that is saying it or reading it...
... and look at a speech act model
... then you stop asking what the triple means, you look at intention
... a speaker utters an RDF assertion using URIs, a receiver reads an interprets it, as a speech act
... this is a different model, which is better suited to talking about the difference between copying something and linking to it
... that you don't try to interpret statements independently of the context in which they are stated
... this leaves httpRange-14 behind, because we're taking on a more complicated, less trusting model for communication
... I've been trying to think about how to express the copyright & linking document we've been working on
... if we were stating these things in RDF
... and that gets you beyond the status code

<jar> This is the "choose based on context" proposal in the chart

noah: how much in RDF would have to be deprecated or turned off to use that new model?

masinter: we have systems that assume trust between communicators
... you can't do induction unless you assume that everything is true

noah: so all the statements from DBPedia remain equally useful if we go down this path?

masinter: you have to put them into context that DBPedia is using, where maybe they intended you to look at 303 codes and maybe used something else
... it's part of a larger framework

noah: I'd be interested to see how far you could get on getting Tim to agree to this

masinter: I think I'm making some progress on him

jar: we have to figure out what problem we're trying to solve
... we've had a hard time getting focus
... these are all fine things to work on, but it's not what we started with
... this isn't an issue for the linked data community

masinter: I think when you get to linked data in the real world
... with government agencies publishing data
... we will need a better model

jar: but is that the TAG's problem?
... that's a problem for linked data, and they will figure it out for themselves
... right now we're doing design for nobody if we take this on

noah: maybe if you change the ground rules, the solutions are different

masinter: I think the question being asked isn't interesting

jar: this is why I want to get it out of the TAG

masinter: we can say we're working on something else which is more important

<masinter> i think we should drop 'findings' more than 2 years old if we can't bring them to community consensus

noah: we have to be careful about how we do this after 8 or 2 years, depending how you count
... we could look really stupid, beating our heads on this for ages, and finally to say it's not our problem, without making clear why new information caused us to change our minds
... I would have a problem just punting without such an explanation.
... Why did this appear to be a TAG problem for so long?
... FWIW, I think the TAG needs to involve itself in clarifying what's appropriate and what's not when using HTTP. Maybe or maybe not we should stop there.

timbl: sorry for arriving late
... having discussed this with jar just yesterday
... I think the hope has been, for me, that the TAG would be able to describe an architecture that included linked data
... the original architecture didn't: it was a hypertext system, it didn't include RDF
... the problem has been partly that there's a small overlap between the TAG and the RDF community
... a relatively small overlap in general between linked data and people doing things with HTTP
... I think RDF should use HTTP properly
... it may be that the niceties of how RDF works can't be discussed with non-RDF people
... so we end up with an architecture that is consistent with HTTP, but is more detailed and refined

noah: one of the proposals that seems to have a low number of downsides is to use a new status code
... I would have thought that the TAG should explain whether that's architecturally appropriate
... the RDF community might separately settle on whether they want to do that
... do you think that's more than what the TAG should be doing?

timbl: if the TAG can solve this, great, write it up
... I regard httpRange-14 and the 303 header as a compromise

noah: we're discussing jar's matrix
... one of the proposals is to use a new status code

<jar> 207, 208, 209

<jar> refresh the page

noah: I'm just probing because you seemed to be saying the TAG should back off because the communities were separate
... I would say the TAG has a role to play to say whether it's an appropriate to use a status code for this

timbl: there are a lot of issues here where it's only the TAG who are chartered to do this
... but I think it would also be reasonable to just make this work for linked data
... if we had a new status code, we would have to explain how to do so, maybe push towards hash more strongly
... still have to deal with simple OGP as well
... there's valuable work to be done here

noah: we're at time
... Jonathan, Henry and Jeni have been talking behind the scenes to move towards more clarity
... I'm inclined to wait a week or two and see what happens

jar: agreed

timbl: I'd be happy to be brought up to date offline

masinter: when they come to a conclusion, they should bring it to us

<noah> ACTION-691?

<trackbot> ACTION-691 -- Jonathan Rees to prepare table as described in 2012-04-04 minutes, for TAG review -- due 2012-04-24 -- PENDINGREVIEW

<trackbot> http://www.w3.org/2001/tag/group/track/actions/691

<noah> close ACTION-691

<trackbot> ACTION-691 Prepare table as described in 2012-04-04 minutes, for TAG review closed

masinter: eliminate from consideration all but three rows

jar: I'm not sure we have standing to do that

noah: you could do it as a proposal

<noah> ACTION: Jonathan with help from Jeni and Henry to try to identify next steps for moving forward on httpRange-14 - Due 2012-05-15 [recorded in http://www.w3.org/2001/tag/2012/05/03-minutes#action02]

<trackbot> Created ACTION-704 - with help from Jeni and Henry to try to identify next steps for moving forward on httpRange-14 [on Jonathan Rees - due 2012-05-15].

masinter: come back with at most the number of rows as the number of people in the subgroup

<masinter> I don't like any of the proposals, because I don't like the question

<masinter> but at least if you're going to continue to ask the same question, come back with at most N rows, where N is the size of the subgroup

<noah> We are ADJOURNED. Informal discussion can continue.

Summary of Action Items

[NEW] ACTION: Jonathan with help from Jeni and Henry to try to identify next steps for moving forward on httpRange-14 - Due 2012-05-15 [recorded in http://www.w3.org/2001/tag/2012/05/03-minutes#action02]
[NEW] ACTION: Noah to put health warning in "Booth Script" for formatting minutes [recorded in http://www.w3.org/2001/tag/2012/05/03-minutes#action01]
[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.136 (CVS log)
$Date: 2012/05/18 12:54:53 $