September 2004 RDF Data Access Working Group Meeting in Bristol, United Kingdom (original) (raw)

Minutes

Convene, take roll, review record and agenda

Roll call: see participants above.

We reviewed the agenda, which was updated several times during the meeting.

We RESOLVED to accept 7Sep telcon minutes, ammended to show Farrukh present.

Introduction and abstract (name, marketing)

We discussed the naming issue along with the abstract and introduction, keeping in mind the importance of outreach. We discussed candidate names such as:

We discussed the costs and benefits of changing names more than once or twice.

ACTION EricP: set up a naming poll using WBS

In many contexts, readers will only see the title or abstract and decide whether to read more based on that alone; we discussed the attractiveness of terms and phrases such as

Kendall suggested that we take the intro of the use cases document and turn it around to be positive. Janne suggested borrowing from a description of SQL.

ACTION KendallC: work on an abstract for the spec

Example Data Access Services

The group discussed example RDF data access services. Additional pointers and technical details are in the irc log starting at 08:41:55Z.

DanC pointed out that while the mail archive search engine can do fielded searches and not just fulltext, it doesn't do arbitrary graph matching; DanC suggested that many interesting data access services won't do arbitrary graph matches, but our RDF data access protocol should allow these services to describe what sorts of queries they do support; several members agreed.

DanC noted the annotea project which provides annotation services, and suggested enhancing them to interoperate with bloggin services such as trackback, and investigate standardized RDF data access interfaces to such services, as a way to realize the media use case.

Dirk encouraged development of negative test cases to mitigate the risk that vendors will essentially do what big customers expect; and make things as close to what their 100.000 ODBC programmers expect, regardless of what our spec says

DanC noted plans to attend XML 2004 in November in Washington D.C. and the possibility of presenting "these things are RDF services from DAWG participants... will interoperate soon (if not already)" aslate-breaking news.

Other outreach venues include ISWC 04, W3C LifeSci workshop, publish something on an O'Reilly site

section 2, Making Simple Queries, core tests

DanC asked the group to confirm that the test results in A few simple tests were in fact specified by the Making Simple Queries section of the spec (v1.55).

This brought up discussion of thetest organization document. DaveB pointed out that test data format information is encoded in URIs, which is "just wrong."

AlbertoR asked about the relationship to previous work already done for SquishQL/RDQL testing.

The question of N3/turtle vs RDF/XML for test data came up; the test editors were advised to make standard RDF/XML available to the extent that it's feasible.

Returning to the test data, some WG members asked whether agreeing to these tests meant agreeing to the specific syntax used in the test. DanC clarified that yes, once a test is agreed by the WG, like any WG decision, new information will be required to revisit it.

AlbertoR noted the interaction of ?foo variable syntax with database APIs and suggested the group consider fooaswellorinstead.ThechairadvisedWGmemberstodiscussthiswiththeeditorsinpreparationforadecisiononthesetests,asitseemedtobeamatteroftasteratherthananoveltechnicalissue.∗∗ACTIONDirk:providedetailsaboutDBinterfaces,re?andfoo as well or instead. The chair advised WG members to discuss this with the editors in preparation for a decision on these tests, as it seemed to be a matter of taste rather than a novel technical issue.ACTION Dirk: provide details about DB interfaces, re ? and fooaswellorinstead.ThechairadvisedWGmemberstodiscussthiswiththeeditorsinpreparationforadecisiononthesetests,asitseemedtobeamatteroftasteratherthananoveltechnicalissue.∗∗ACTIONDirk:providedetailsaboutDBinterfaces,re?and ("usual suspects which are troublesome; postgress, oracle, dbi, jdbc, odbc and the connections to java, perl, c++ et.al.")

JFB asked: why the distinction between query variable and rdf variables? AndyS ansswerd that query variables quantify over terms in a graph, while bNodes quantify over objects in the domain of discourse.

The group discussed the core definitions such as RDF termand binding and substitution. KendallC found:

Definition: A substitution is a finite set (possibly empty) of pairs of the form Xi=ti, where Xi is a variable and ti is a term, and Xi\neq Xj for every i \neq j, and Xi does not occur in tj, for any i and j.

src

ACTION AndyS: fix def'ns in sect 2 up to 2.2

The group talked about which genre this specification is in or should be in: tutorial? reference? math text? DanC said he had asked that the editors introduce the material by successive elaboration, introducing mathematical definitions along the way, and that he was largely satisfied with the editors' progress in this direction.

Yoshio asked, re section 2.3, if one can join triples with ";" like in N3? A straw poll showed 5ish for, 4ish against. DaveB argued that the query syntax should be regular; no abbreviations ala N3 ;nor ,.

SteveH asked about the { s p o. s p o. } syntax used in BRQL v1.55; suggested the ( s p o ) ( s p o ) syntax is more mature and proposed returning to it. AlbertoR noted this is the convention in the jena-dev list. JosD noted that using (...)for triples conflicted with the potential to use it for lists. AndyS explained that the BRQL 1.55 design was based on:

  1. comments that there was no difference between constraints and triple patterns, so they should share syntax.
  2. including nesting syntax for disjuction etc.

JanneS asked that the spec and test editors ensure that the full range of RDF literal constructs are supported: lang, datatype, etc. AndyS agreed that more examples would help.

Then we broke for lunch.

Technical Issues

DanC presented his recent (version 1.13 2004/09/09 15:29:55) work on the issues list, and reiterated some milestones from the WG schedule:

There was some discussion of time-to-market vs expressivity.

Those who had offered to own issues were asked what order they wanted them discussed.

SOURCE

DaveB introduced the the SOURCE design in BRQL 1.55 and related positive implementation experience. The group discussed various technical details. DanC asked if the definitions cover SOURCE; DaveB said no, he hadn't gotten to that yet. AlbertoR pointed out an identity management use case relevant to this issue.

DanC argued to postpone this issue to allow more time for experimentation in the community. Others argued they had seen enough experimentation. DanC asked whether the implementations in the community have converged on any particular design and suggested that test cases would shed some light. ACTION SteveH: test cases (10 or so) re: current SOURCE design. ACTION AlbertoR: real test cases re: current SOURCE design. ACTION AndyS: make test case available as it distinguishes between two designs of implementing source.

ACTION AlbertoR: (with SteveH) edit the examples in this msg into test cases (either positive or negative tests)

ACTION DaveB: Update the source section 9, add more formal links, update the examples, try to think about extra constraints as EricP proposed (SOURCE ?s and ?s onlin in SELECT). Look at various ppl's source test cases.

PREFIX

KendallC updated the group on his progress toward a proposal: he has surveyd namespace management in addition to rdf query languages; would prefer declarations of namespace before (strawpoll indicates support for before instead of after).

other issues

ACTION SteveH: owns issue 'nested optionals'

ACTION AlbertoR: owns issue 'DESCRIBE'

ACTION DanC: add issue 'design of disjunction' (DONE in 1.14)

ACTION DanC: owner of issue 'yes or no questions'

ACTION DanC: add new issue 'eddie number of bindings vs. number of graphs'. (DONE in 1.16)

Requirements from Web Services Constraints and Capabilities

DanC noted the upcoming W3C Workshop on Constraints and Capabilities for Web Services and asked if there was overlap between that work and the work of this group. Brief discussion uncovered some possible overlap and some interest in investigating further. ACTION KendallC: investigate overlap between RDF query and web service constraints.

UDDI Use Case

We noted the lack of progress toward a UDDI use case.WITHDRAWN: Hendler draft UDDI use case. ACTION DanC: tell semweb CG that the UDDI action was dropped

Device Independence WG Use Case

DanC summarized discussion of use case: content selection based on client profile). The use case appealed to a number of WG members.ACTION KendallC: add DI use case.

AndyS noted possible complications regarding querying bags, containers, and the like. ACTION DanC: add ISSUE: collection or container member accessor. (DONE in 1.17)

SteveH mentioned some related work. ACTION SteveH: send mail about the signage project.

XQuery integration

The group discussed two proposals for integrating RDF data in an XQuery environment:

Making RDF Data Available for XML processing

AlbertoR noted that the RDFLet mechanism uses this query return ResultSet format to pass the information from the RDF query engine to an XSLT processor. This allows sorting etc. to be handled in XSLT if it's not handled in RDF query.

SteveH reported on a similar effort with a slightly different XML syntax for results. One of the difference was terseness, which, for this application, made a noticeable difference. DanC suggested using markup like HTML tables to avoid repeating the variable names, which led to questions of how to represent unbound variables.

When asked how datatypes are represented, AndyS gave an example:

mbox 10

RobS noted that many XML tools look for their datatype information only in schema.

KendallC suggested that consistency between the element names in this format and concept names in the spec would help people learn the technology.

KendallC noted some related in-progress work. ACTION KendallC: put xml format up somewhere public, mail a pointer

see also later discussion of XML format for results

XQuery syntax for BRQL semantics

EricP reported that he had started implementing the XQuery syntax for BRQL in order to find out if the details work out; he had a parser generated from a grammar but hadn't hooked it up to a query engine.

DanC recalled a question from AndyS about how the extension functions are declared; exploration of the XQuery specifications suggested the relevant extensiblity was there.

RobS re-iterated Network Inference's observations that customers are tooling up on XQuery and will expect RDF query to look similar. He also re-stated that a different query language for every layer of the Semantic Web architecture would be poor design. DaveB noted that a normative dependency on XQuery would be incompatible with the current DAWG schedule. Some WG members (JanneS, Dirk) noted their customers use XSLT and XPath but have not yet adopted XQuery. JosD re-iterated arguments based on the principle of least power. Straw polls showed little support for further exploration of the XQuery syntax for BRQL. The possibility of a WG NOTE on this syntax was discussed without much enthusiasm.

We then called it a day.

issue: disjunction

The group discussed the interaction between disjunction designs and implementation/optimization strategies. JosD noted requirement 3.13, but SteveH offered that some OPTIONAL designs might meet that requirement. Some examples/test cases of disjunction and optionals were discussed. A straw poll showed 5ish in favor of dropping disjunction, 4ish against. ACTION SteveH: own (i.e. propose resolution to) disjunction issue

f2f, telcon schedule

DanC noted that the 2005 Tech Plenary date is settled (28 February - 4 March 2005) and invited discussion of meeting dates before then, especially offers to host. Discussion of availability and constraints in Jan/Feb 2005 showed 19-20 Jan 2005 as likely meeting dates.

After brief discussion of teleconference times, we RESOLVED to meet at 14:30Z on tuesdays until further notice (i.e. one hour earlier in Boston after the EDT changes back to EST). RESOLVED to cancel the 21Sep telcon; next meeting is 28Sep; JanneS to scribe. f2f proposals expected before that meeting.

Query Language Syntax

AndyS and EricP, BRQL editors, collected advice on query language syntax. While much of the focus was surface syntax, issues of expressivity (e.g. nested optionals) were also discussed. Discussion concluded with some support for a syntax exemplified by:

PREFIX foaf: http://xmlns.com/foaf/0.1/

SELECT ?name1 ?name2 ?hpage2 WHERE (?x foaf:knows ?y) SOURCE http:D2 (?x foaf:age ?n) (?x foaf:age ?n) AND ?n < 18 (?x foaf:name ?name1 )

SOURCE http:D2 (?x foaf:status ?status)

SOURCE ?src (?x foaf:status1 ?status1) SOURCE ?src (?x foaf:status2 ?status2)

[ ( ?y foaf:name ?name2 ) ( ?y foaf:age ?m) AND ?m < 18 ]

After lunch, JanneS and JosD were excused to catch their planes.

Publication plan

AndyS noted limited availability 20-24Sep, followed by a more normal week 27Sep-1Oct. ACTION AndyS: (with EricP) update QL spec draft by 1 Oct. The WG will be interested in a status report in its 28Sep meeting. ACTION SteveH: review updated spec and advise WG on whether to publish as 1st WD. ACTION DaveB: review updated spec and advise WG on whether to publish as 1st WD. Note Well: WG members should be prepared for a 5 Oct decision on whether to publish the 1st WD of the design. EricP indicated availability to act as team contact for the publication in the event the WG approves publication.

ACTION DanC: talk with Kendall about issues list maintenance

ACTION DanC: add a pointer to the issues list to the DAWG home page. (DONE in 1.95)

Protocol: XML format for query results, HTTP Binding

Noting considerable interest in an XML format for query results, DanC asked for volunteers to continue that work. DaveB offered to make a schema and perhaps rename the elements to match the BRQL spec terms.ACTION DaveB: draft XML query results format spec. DaveB collected advice on other details.

The group then discussed the use of this results format in HTTP-based protocols, noting RDF Net API, Joseki, and the like, as well as related work such as the Annotea protocol and the Atom publishing API. ACTION DanC: add issue: protocol URIs are for services or for document/graph/models? (DONE in 1.17) ACTION EricP: follow up on " if we send a queyr to the document rather than service, we loose some of the document compositibility possibilities" in email. ACTION KendallC: write a protocol document draft (est delivery in 1 month).

Protocol: RDFS query, server capabilities

DanC presented source of a triple, which discusses swap/cwm/n3 design for interaction of source and inference, keeping the URI for a document/graph separate from its RDFS closure.

Adjournment

We RESOLVED to thank the host, HP, especially AndyS and Julie Lanfear, for hosting the meeting, with applause. The meeting was then adjourned.