WS Description WG telcon
5 Oct 2006

<scribe> SCRIBE: alewis

approval of minutes

minutes approved 28 September 2006 teleconference

Action items

[Interop]
?         2006-07-06: [interop] Jonathan - create validation-report
                      stylesheet.
?         2006-07-20: [interop] Jonathan to add timestamps to result
                      stylesheet.
?         2006-09-21: [interop] Philippe to transform HTTP headers in
                      the logs to XML format.
WG
?         2005-07-21: Pauld to write a proposal for a working group
                      report for requirements for schema evolution
                      following closure of LC124
?         2006-07-06: Glen to contribute some extension test cases.
?         2006-09-21: Jonathan to check periodically that SPARQL has added
                      schemaLocation.
DONE [.3] 2006-09-21: Arthur to create a un-namespaced test case.
?         2006-09-28: Youenn to propose an alternative syntax for MTOM
                      if F&P is removed
?         2006-09-28: Marsh to suggest some generic conformance text
?         2006-09-28: Jacek to follow up with Karl Dubost to get info
                      about RDF issue 297.


Current Editorial Action Items
?         2006-09-28: Editors to factor the "extra" MEPS out of the
                      specification (Part 2) and make a new NOTE

Note: Editorial AIs associated with LC issues recorded at [.2].

[.1] http://www.w3.org/2002/ws/desc/#actions
[.2] http://www.w3.org/2002/ws/desc/5/cr-issues/actions_owner.html
[.3] http://lists.w3.org/Archives/Public/www-ws-desc/2006Oct/0004.html

(quickly reviewed)

Jonathan: no longer affiliated with Microsoft, now with WSO2, will continue on working group, as co-chair and WSO2 representative
... Asir continues to represent Microsoft (currently on paternity leave)

(chaffer targeting Jonathan and Asir and Arthur)

<Jonathan> Jonathan's new email address: jonathan@wso2.com

Administrivia

Interop event, November 14-18

scribe: Canon to host.
... location of interop event is Rennes, France
... more administrivia: should we review the change from x-www-url-encoded to www-url-encoded, approaching submission in IETF.

Jonathan: do we need to review this, or make specification changes?

little interest or concern expressed.

features at risk

UNKNOWN_SPEAKER: features and properties awaiting submission from Youenn on MTOM (CR62). Discussion awaits fulfillment of action item.
... is there anything other than Youenn's action that we're waiting for to make the decision on Features and Properties?

Philippe: SOAP modules may also be linked to removal of features and properties, since based on that.

Jonathan: Glen modified that statement, to say that the functionality is less than ideal, but still usable.
... Arthur is suggesting that we need to make the Features and Properties decision next week, when we've seen Youenn's proposal.

CR80, canonical component designators.

Jonathan: is the purpose of component designators furthered by having canonical designators?

Jacek: yes, these can be useful to RDF and web arch generally, but ought to be "SHOULD" rather than "MUST".

Jonathan: someone needs to write up the canonical form of XPointer URIs for component designators
... might be restricted to namespaces and such.
... canonical XPointer format might have utility outside of this group.
... fragment identifiers are too restrictive; component designators should be more sharply targeted. Should be able to use a component designator as a fragment identifier, but not necessarily the other way around.

Jacek: anyone using fragment identifiers knows how to use them, as a rule.

Arthur: but in RDF, these things need to be opaque, comparable URIs

Jacek: yes, but it's unlikely that two different sources will generate fragment identifiers from the same WSDL and then need to compare them.
... it would be nice to have canonical designators, but is not critical.

Jonathan: add a paragraph to spec. "For easy lexical comparison of fragment identifiers ..." (he needs to put the rest of this in an email, probably)

discussion of how to generate namespace prefixes, as applied to the problem of canonicalization.

Arthur: does this also require that the base URI be canonicalized?

<Jonathan> - only 0 or more xmlns() and exactly 1 wsdl.*() pointer parts may appear

Jonathan: we start from a target URI, which is lexically defined.

<Jonathan> - prefixes must be named ns1, ns2

<Jonathan> - no whitespace appears in the XPointer

<Jonathan> - no duplicate URIs in xmlns()

Arthur: true for WSDL, but we also have schema URIs.

Jonathan, Jacek: same thing.

<Jonathan> - multiple xmlns() parts appear in the same order as in the wsdl.* part

Arthur: if namespace recurs, reuse prefix.

Jonathan: are these rules sufficient for the case of an extension?

Arthur: yes, the frag id is a string, so has an inherent left-to-right order. name them in order of occurrence.

<Jonathan> "For ease of comparison, Component Designators SHOULD conform to the following canonicalization rules."

<scribe> ACTION: Arthur to revise appendix to include this text. [recorded in http://www.w3.org/2006/10/05-ws-desc-minutes.html#action01]

Arthur: WS-Policy will also use these fragment ids.

RESOLUTION: CR80 closed with the addition of a statement recommending canonical construction of URLs representing fragment ids.

<Jonathan> note that the use of must within SHOULD is not very nice

Arthur: maybe also revise the examples in the specification using the canonical form?

Jonathan: one more issue on list, from 2004, note from TAG on safety extension. TAG wants to track usage through deployment.
... suggest that at some point he can report implementation to the TAG.
... Woden and Canon currently implement.

Arthur: can WSDL claim that something is safe when it isn't?

Jacek: planning to use this information in research; although no implementation, it's very useful in its present form.

Tom: reads this as looking for clarification on what "safe" means.

Arthur: maybe we need to say that "read-only" is not enough, it needs to have no side effects.

Jonathan: but we don't want to define it.
... point to canonical definition.

Tom: can't we just add a couple sentences?

Jonathan: no, they want something in the test suite.

Tom: "we called launchNuclearWeapons twice, and it wasn't safe even though it was marked that way!"

question: are there examples in the primer? Is this the best place to do it?

Arthur: there's discussion in the primer, maybe Jonathan should point the TAG at the primer on interface operation.

Jonathan: yes, this text looks pretty good. Reluctant to do more, except eventually ping the TAG again to say "still there, seems to be working, no internet-scale deployment at this time, but no reason to expect misuse"

Arthur: is it generally an error if an operation is bound to HTTP DELETE and it's marked as safe?
... can we check that in validation?

Philippe: no.

Arthur: sure we can, if we know the binding.

Philippe: only GET is safe.

Jacek: what is guaranteed to be *unsafe*?

Amy: is there an assertion about usage of HTTP methods in the WSDL HTTP binding?

Joanathan, Arthur: well, we should be able to work it out from the combination of the HTTP spec and the definition of safety.

Arthur: yes, but we could put the assertion in, in order to have a test case that clearly violates an assertion (and is therefore invalid).

<pauld> DELETE is unsafe

discussion of whether a WSDL can be called invalid if it does not violate a direct assertion. can validity be based on implicit violation of the safety attribute?

discussion of which methods are guaranteed unsafe.

paul says delete is; jonathan suggests not (if there's a nonexistent resource, for instance).

Arthur: and there is no guarantee that an application will implement the HTTP verbs properly.

Jacek, Arthur: we can point them to the primer, and point out that it's not in the test suite.

Jonathan: the issue will be resolved via a report to the TAG, the longer we wait, the more implementation experience we can report on.
... substantially completed with work. we will have a call next week to talk about features and properties.

<scribe> NEW ISSUE: per email from Ramkumar Menon, question of fault binding raised in discussion of synchronous versus asynchronous.

Jonathan: will add issue 81, and close it immediately by requiring fault binding.

Jacek: faults outside operations don't need to have bindings.

<pauld> notes DELETE on a non-existent resource should return 404, not 200, 202, 204 and invalid implementations sounds bogus

Jacek: uncomfortable with this MUST.
... maybe say it in the qualified version?

Jonathan: maybe leave issue open for this week?

<scribe> ACTION: Jacek to send response email noting his concern. [recorded in http://www.w3.org/2006/10/05-ws-desc-minutes.html#action02]

Arthur: in practice, you won't have a problem until you use a fault.
... if something goes wrong, and you need to send a fault, you need to know how to bind it.
... so there can be unused faults.

Jacek: but what about faults that are defined, but not referenced from any operation?

Jonathan: let's look at it; it doesn't sound controversial, just needs a little more thought.

<Jonathan> ACTION: Jonathan to add issue 81 [recorded in http://www.w3.org/2006/10/05-ws-desc-minutes.html#action03]

Arthur: yeah, probably happened when we pulled fault out of operation.

Meeting closes.

<Arthur> assertion coverage report: http://dev.w3.org/cvsweb/~checkout~/2002/ws/desc/test-suite/assertions-report.html?content-type=text/html;%20charset=utf-8

implementors' report.

226 assertions, only 54 covered by test cases.

Arthur: also invites people to contribute to Woden.

<Arthur> Woden project needs help: http://incubator.apache.org/woden/

<JacekK> the minutes are public, is that a problem?

bye, gang.