RDF Data Access Working Group Face to Face Meeting, 2 - 3 March 2006

<_ed> scribenick: SteveH

Convene ...

<AndyS> Face to Face Meeting Page

Agenda discussion

<AndyS> minutes of last meeing

PROPOSED to approve minutes of last meeting

APPROVED, no objections or abstentions

ericp: want to ask the community at large what they want from SPARQL
... try and echo opinions while everyone is in the same place. WBS form to ask for features

<LeeF> SPARQL Next Steps form

filling in of the WBS form...

<ericP> earlier summary of feature requests

SerT: we can not go into sleep state, as the spec. explictly rules out SPARQL over OWL engines

JosDeRoo: you can query over OWL

SerT: there is no clear semnatics

Andys: its more an issue of scoping [?] eg. should come from SWIG

pfps: there has not been enough thought into what it means to query and OWL Full graph

ericP: can you sketch out what it would look like to do that

pfps: I think it would be insane to query OWL Full syntax
... we would need some way to distinguish between the phantom syntax stuff, and the real data

ericP: surely it's the OWL WGs job to distinguish between the useful and useless stuff

JosDeRoo: you can do it with the comprehension principle
... we have addressed this in the OWL testcases, re. positive entailment

pfps: OWL may not be finitely describable, its possible you can come up with a cutoff algorithm, but maybe not

<AndyS> DAWG Issues list

<SerT> for undecidability results for OWL-full see e.g. http://www.springerlink.com/openurl.asp?genre=article&issn=0302-9743&volume=3729&spage=548

SPARQL Query/Protocol pub issues

ericm: we need to define an exit criteria
... eg. 2 or 3 implementations each testcase

LC -> CR -> PR -> Rec

ericm: only allowed to party after Rec
... need to identify codebases that are implementations of SPARQL
... engage with non-WG members, looks better if you have a lot more than the minimum number of impls.
... work with communities, bet them to beat against the testcases
... press issues, once it looks like were going to make it, its a good idea to think about contacting organisations, to get testimonials "we think this technology is cool because..."
... think about the kind of things that you organisation might be able to say, will do press release with testmonials to get the idea of why different orgs. think the spec is a good thing
... lots of dawg shows, most done by individuals in this room

dannyw: its fine to have overlap, but there better be a reason for it

ericm: need to explain why, and need to respond to objections
... build a matrix of testcases and implementations

ericp: broadly, functionality falls into SELECT/ASK, CONSTRUCT and DESCRIBE

<LeeF> SerT: we don't have any test cases for CONSTRUCT and DESCRIBE

SerT: groups who want DESCRIBE should come up with testcases

ericm: arcs in, arcs out os sufficent for explorations

AnydS has a DESCRIBE impl. no-one can rember if they do or not

ericm: matrix could/should be a feature matrix, not just testcases

<EliasT> ericm: We need to be thinking whether our exit criteria matrix will be test case, feature or application driven

ericm: need testcases to features mapping

andys: should map testcases to UC&R

<AndyS> I haves 432 sripted tests - maybe 50% overlap with DAWG that I contribute to DAWG

<jeen> our sparql engine has 45 unit tests, the serql engine has a couple more (including a whole set of construct-tests), which might be useful as well.

<AndyS> AndyS will help with the mapping between UC&R and testcases (but can't offer to own the action)

<ericP> ACTION: SteveH and AndyS to map approved tests to UC&R features [recorded in http://www.w3.org/2006/03/02-dawg-irc]

<ericP> RRSAgent this is DAWG

ericm: need to think about the CR duration

<EliasT> http://www.w3.org/2001/sw/DataAccess/crq349

<AndyS> http://esw.w3.org/topic/SparqlImplementations

<ericP> ACTION: EricP to follow up on mime type requests [recorded in http://www.w3.org/2006/03/02-dawg-irc]

<ericP> ACTION: SteveH and AndyS to map approved tests to UC&R features [recorded in http://www.w3.org/2006/03/02-dawg-irc]

ericm: ask folks on /SparqlImplementations if they can provide assertions about which test cases they pass
... ask them if they can get this data by the end of CR

start CR then add more testcases, or add testcases, freeze then go into CR

Test Cases for CR

<_ed> Scribenick: jeen

<AndyS> SPARQL Tests

ericP describes test case schema/vocab

SerT: SPARQL has more features than are in UC&R
... do we have to justify those features as well?

Use Cases and Requirements

AndyS: no. UC&R is what the WG is chartered to address

<LeeF> ScribeNick: jeen

<SerT> test cases for CONSTRUCT operator (Section 3.4 of UC&R): http://lists.w3.org/Archives/Public/public-rdf-dawg/2006JanMar/0262

<ericP> http://www.w3.org/TR/2004/WD-rdf-dawg-uc-20040802/

<ericP> http://www.w3.org/TR/2004/WD-rdf-dawg-uc-20040802/#d4.6

ericP: identify requirements by url+fragid
... how much do we want our implementations to say that we do/don't match these tests?

steveh: let's stick to feature-compliance and not go the conformance-tests route

ericP worries about this being too trivial

AndyS: we don't want to create a burden on the process

ericP: do we then want additional requirements (next to UC&R)?

(everyone says no)

SerT: compliance is defined in the protocol doc

AndyS: for every requirement, there must be two or more implementations.

SteveH: we can add tests for describe and then just assign it to a feature

<AndyS> Or not assign it and regard it as "helpful" to implementers

<ericP> PROPOSED: all exit criteria are mapped to terms in the UC&R (ala http://www.w3.org/TR/rdf-dawg-uc/#r3.1)

<_ed> APPROVED -- noting that this is advisory to the director

<AndyS> Draft text: "All functional requirements from the UC&R will be covered by at least one test case"

<AndyS> OWL had pass, fail, N/A and no data

<ericP> PROPOSED: All functional requirements from the UC&R will be covered by at least one test case. All of the exit criteria tests come from these UC&R requirements.

APPROVED, no objections

Next Steps

<_ed> ScribeNick: _ed

Sorting feature requests by popularity. Notation: feature - requests

<ericP> protocol for graph update - 5

<LeeF> ScribeNick: EliasT

AndyS: has already done "accessingCollections" using built-in predicates ?x list:member <cool>
... JosD has also done this on euler. It might just be a matter of writing a note.

<ericP> querying collections - 0

ericP: believes that it's still confusing since euler might do it differently than cwn or jena.

<ericP> count(*) - 5

<AndyS> It's probably easier to do it (count() (agreed amongst impls - not WG track) than keep answering emails requests.

ericP: thinks that relational algebra could help us with group by

<ericP> aggregate functions - 1

Jeen: already working on aggregate functions on Sesame

Sven: interested on subqueries (nested)

<ericP> subqueries (i think that means nested queries) - 1

<AndyS> AndyS interested in designs for aggregate (so impls do similar) but not a priority at the moment

<ericP> told bNodes - 2

SerT, Jeen, AndyS

<ericP> 2.5

<ericP> "fourth slot", (provenance) queries ala RDQL

<AndyS> Confession: I didn't take the support for << >> out of ARQ :-)

<ericP> "fourth slot", (provenance) queries ala RDQL - 0

AndyS: believes that fourth slot is a semantic web issue not a querying issue.

<AndyS> RDQL does not have a fourth slot :-)

<ericP> but it is a common RDQL extension

<ericP> higher level semantics issues (OWL) - 1

SerT: transitive properties have different meaning, it's either regular expressions for graphs or OWL transitive properties

<ericP> transitive properties expressivity in the QL -

<ericP> transitive properties expressivity in the QL - 1

Federated Query

<ericP> Federated query -

SteveH: there are two things: query of a join of two datasets, or a union of the datasets

AndyS: sees two things: you ask a node and it goes around and asks the web, the other web, it's the application writer needs control of how it's happenning.

SteveH: was having a conversation with Bijan and he thought Bijan wanted multiple service endpoints in the query.
... distributed join and union

<ericP> Federated query (distributed join) - 0

<ericP> Federated query (union over results) - 0

<ericP> cascadedQueries

JosD: would like to have SELECT WHERE SELECT WHERE ...

<ericP> cascadedQueries (distributed join) - 0

SteveH: maybe just syntactic sugar like adding UNION at the top.

<ericP> cascadedQueries (union over results) - 0

<AndyS> <insert text="usual stuff on server state management" from="AndyS"/>

service description

<ericP> service description -

<ericP> ericP: addresses complaints that we don't have a way to identify logics

<ericP> service description - 0

<ericP> XML abstract syntax

-- http://www.w3.org/TR/rdf-sparql-protocol/ SPARQL Protocol

-- http://www.w3.org/TR/rdf-sparql-protocol/#select-post-xml SPARQL in XML example in section 2.2.1.13

<ericP> XML abstract syntax - speculate 1

Query By Reference:

EliasT: adding another queryRef parameter to the protocol to a .rq file.

AndyS: There a couple of issues such as in the SOAP case mixing different protocols, base for relative url handling

http://www.w3.org/2002/09/wbs/1/DAWGnextSteps/results

ericp: it seems that the clear winner is update

SerT: update gives hint that query is attached to an RDF store
... for example external data sources

<LeeF> +1 to just one sparql-dev list

<jeen> +1

SerT: If we do update, we also need to worry about transactions

JosD: I'm afraid of doing update right now.

SteveH: We could do update without transactions.

The group has agreed that we would have to recharter to handle update due the complexities and nature of the problem. It would also be good to wait to see what the RIF defines as its goals.

JosD: believes that having a note would help, especially in the case of accessing collections

<AndyS> (member submission)

<ericP> 2006-03-02T13:13:08Z <ericP> querying collections - 0

<AndyS> . ACTION EricP: email the semweb CG about a sparql-dev mailing list

<scribe> ACTION: EricP to email the semweb CG about a sparql-dev mailing list [recorded in http://www.w3.org/2006/03/02-dawg-irc]

BijanP: suggested sessions

<ericP> SESSIONS!!!! - 0

sessions in the protocol

<ericP> transactions

<ericP> transactions - 0

<AndyS> Sessions (transaction particularly) will be a request soon after update

<SerT> http://en.wikipedia.org/wiki/ACID

<ericP> protocol for graph update - 5

<ericP> count(*) - 5

<ericP> concensus is to have interested parties work on UPDATE and COUNT(*) as member submissions

<ericP> concensus is to finish with the rec-track docs and adjourn indefinitely

<AndyS> After the break: test cases : 16:00

<LeeF> ACTION: LeeF to find email thread about wording of restricting pattern solution domain to variables in BGP; follow up on that to get text back into rq23 [recorded in http://www.w3.org/2006/03/02-dawg-irc]

<AndyS> not touching sparlq.org until AFTER any demos :-)

<LeeF> i thought you were going to say until after you got your beer =)

<AndyS> Is that before the demos?

<ericP> i bet you're not drunk enough to rewrite your kernel

<LeeF> Could be ;-) (actually, demo is against my local joseki, since that's the only one I have with ex:substring :-)

<AndyS> local != sweb : can anything be shown on the web?

<LeeF> I disagree with that assertion

<LeeF> It runs on both (local or web URLs), in any case, as long we use a joseki which supports substring

<AndyS> I could reploy but the args and name are different to your demo. Uses java:

<LeeF> too late, now

<AndyS> Local joseki OK - if the data is remote.

<LeeF> local data is OK, too

<AndyS> Absolutely!

<LeeF> :)

<LeeF> I promise to show you it working remotely at the end of the demo =)

<AndyS> Local data + local processor < MS office / open office :-)

<LeeF> But it's Web 2.0 and has shiny colors!!

<AndyS> Shiny thing that way ===>

<libby> very neat stuff

<LeeF> like most such things, it's far prettier on the surface than under the hood

<libby> heh

<libby> that's the right way around I reckon

<jeen> :P

<AndyS> Shiny Web

<LeeF> "it may be butt ugly, but boy is it engineered well!"

<libby> hehe

<libby> mine is ugly inside and out!

<jeen> yeah, that never sells I guess :)

<AndyS> SPARQL contractors

<libby> where am I gonna find them?

<LeeF> "interns"

<libby> ahh

<LeeF> "unpaid interns"

<jeen> i haven't yet been contacted about that from thailand or korea, but I'm expecting a flood of them after we go CR

<libby> unpaid++

<AndyS> We're looking for cntractors as well - tool builders and app builders - need a contractor site on sparql.org

<AndyS> :-)

<libby> there are none left!

<AndyS> I seem to have 0.4 slaving away at the moment

<AndyS> but seriously - we could do a pass-around if first contact not interested.

<AndyS> Or maybe we are really stepping into training people up as a cost of getting contractors. Yuk - but that is where we were 3 years ago

<ericP> Scribe: LeeF

Tests

<ericP> requests for test cases

<ericP> schema

(discussion of test case schema vocabulary)

SerT: do we have any cases in which we need multiple test cases to cover a requirement?

Group: we think we can avoid that

<EliasT> http://www.w3.org/2001/sw/DataAccess/tests/test-dawg.n3

SteveH: To start to clean up the tests, we could reject (or delete) the tests that use the old syntax

<EliasT> http://www.w3.org/2001/sw/DataAccess/tests/data/simple2/

<EliasT> http://www.w3.org/2001/sw/DataAccess/tests/data/simple2/manifest.n3

old syntax tests

ericP: as per W3c persistence policy, keep everything in current test directories (mark with big red Xes if they are misleading)
... also provide a tarball of approved tests

AndyS: target audience is SPARQL implementors who will provide experience reports during CR, so we should have a download which is only the approved tests

SteveH: why are the tests in an rdf:List

AndyS: ordering is nice when running tests - if a test fails, the failure always occurs at the same place in the testing procedure

ericP: other WGs produce and publish tests. do we want to publish our approved tests as a Note, a subset of which are the ones that support requirements which apply to CR exit criteria

SerT: Test cases should not be used to show the conformance of implementations
... keep implementation/conformance requirements distinguished from requirements of defining the query language

AndyS: w3c doesn't have authority to draw conclusions from implementations that pass certain tests

SteveH: "approved" tests mean that the WG believes that the test accurately reflects the specification of the QL

ericP: 3 options here
... 1) keep everything in 2001/.../tests/data annotated appropriately to allow informed use of that data
... 2) sequester the tests that support requirements in a separate place to allow implementors to run those tests to support our CR exit criteria
... 3) that in addition to 2, or in a separate place, we have the approved tests that do not support requirements

SerT: -1 on other approved tests in a separate place due to concerns of maintaining consistency

(discussion of how to avoid lists in the test schema)

EliasT: we don't need tests in a list for the protocol tests

AndyS: To the projector!

(AndyS shows us the following:)

running a JUnit test cuite inside Eclipse

a green bar grows across the screen

1140 tests are run, displayed in groups

about 400 of the tests are scripted -- they come from manfiest files

ericP: one approach is to stay with lists -- people running the manifests in that case cannot use SPARQL--

SteveH: that's not true because you can query based on other predicates

<EliasT> http://lists.w3.org/Archives/Public/public-rdf-dawg/2005OctDec/0052.html

<ericP> SyntaxFull manifest

SerT: by using SPARQL to query for items which have an action, you can't find which group a particular test belongs to

SteveH: use GRAPH ?g to associate actions with manifest origin
... that way we can build everything with a single CONSTRUCT query

SerT: we can preprocess the data and shortcut the list structure with the rdf:member property

ericP: we still need pub;lication mechanism and (b) ?publication contents

AndyS: and who is the test case editor?

s/publication contents?/publication contents/

<AndyS> Sergio suggested adding mf:member property to directly connect the manifest to the action

<AndyS> AndyS: This could be as well as the list structure.

ericP: publish 1 directory structure and tarball thathas every manifest entry that is approved

<AndyS> ... The addition properties can be added mechanically.

ericP: people who are doing CR testing will do a query for supportsRequirements
... this leads to our matrix of test case coverage

SerT: only concern is that we maintain coherence in the entire tests structure

ericP: Manifest of manifests (a la AndyS) will do this

everyone agrees on publishing this as a Note

ericP: should we have unapproved tests in the structure that we publish as a Note?
... No.

err

SerT: No.

ericP: master will continue to be 2001/.../tests/data
... add fromManifest property pointing from manifest entries to the manifest itself

[ coffee break while ericP does homework ]

<AndyS> [ Eric starts homework ]

<AndyS> Issue: protocol examples use URIs (http://www.example/) that don't actually have an RDF graph behind them.

<AndyS> ... so we can't just publish the protocol examples and have implementers execute them directly.

<AndyS> [ WG works on designing the query test manifest ]

<ericP> PROPOSED: 1) create manifest manifest 2) add the includes, fromManifest and supportsRequirement per photo from meeting record 3) publish a copy of 2001/test/data/ that includes all approved tests

<ericP> PROPOSED: 1) create manifest manifest 2) add the includes, fromManifest and supportsRequirement per photo from meeting record 3) publish a WG test note that includes a copy of 2001/test/data/ that includes all approved tests

<AndyS> http://www.w3.org/2005/09/27-dawg-minutes.html#action16

<ericP> PROPOSED: 1) create manifest manifest 2) add the includes, fromManifest and supportsRequirement per photo from meeting record 3) publish a WG test note that extracts the ManifestManifest, manifests and all approved tests from 2001/test/data/ that

<ericP> PROPOSED: 1) create manifest manifest 2) add the includes, fromManifest and supportsRequirement per photo from meeting record 3) publish a WG test note that extracts the ManifestManifest, manifests and all approved tests from 2001/test/data/

<ericP> APPROVED, AndyS abstains

<ericP> LC status report (note open and pending threads)

<scribe> ScribeNick: ericP

<scribe> scribe: ericP

Pending Comments

LC status report (note open and pending threads)

<EliasT> that's ok jeen. Now you may come in to the room.

<DanC_lap> # 2005-09-08T01:26:59Z Peter F. Patel-Schneider

<scribe> ACTION: DanC to join the [2005-09-08T01:26:59Z Peter F. Patel-Schneider] thread with PPS's late Feb comments [recorded in http://www.w3.org/2006/03/03-dawg-irc]

resolution path on [2005-09-09T20:39:31Z Dan Connolly]: ask if Ian Horrocks is satisfied with DAWG resolution of owlDisjunction

<DanC_lap> which isn't novel from this meeting; that's just a matter of course

<DanC_lap> ACTION: patH to respond to PFPS comments [WITHDRAWN] [recorded in http://www.w3.org/2006/03/03-dawg-irc]

action -2

<DanC_lap> overtaken by pfpf's late feb comments

<scribe> ACTION: SergioT to ask Ian Horrocks if he is satisfied with DAWG resolution of owlDisjunction [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<DanC_lap> i.e. reply to http://lists.w3.org/Archives/Public/public-rdf-dawg-comments/2005Sep/0049

SerT, i've just bounced efb2f1492677c59ad7c4459e134425f6@cs.man.ac.uk to you

<DanC_lap> (steveh is excused for a bit)

<scribe> ACTION: DanC to fix the "inconsistent use of terminology" [2005-09-23T14:34:50Z Peter F. Patel-Schneider] thread [recorded in http://www.w3.org/2006/03/03-dawg-irc]

http://www.w3.org/2001/sw/DataAccess/rq23/#optionals

1.595 (aseaborn 19-Dec-05): <p>The <code>OPTIONAL</code> keyword is lef

t-associative: </p>

<DanC_lap> investigating ter Horst's comments. http://lists.w3.org/Archives/Public/public-rdf-dawg-comments/2005Oct/0068

<DanC_lap> 2006-02-22T14:46:35Z Arjohn Kampman

<scribe> ACTION: AndyS to respond to [2006-02-22T14:46:35Z Arjohn Kampman] [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<scribe> ACTION: LeeF to create test case resulting from 2006-03-03T11:15Z (before lunch, 2nd day) whiteboard discussion [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<scribe> ACTION: EricP to respond to [2006-02-23T09:38:05Z Arjohn Kampman] [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<DanC_lap> # SPARQL Dataset Graph Pattern

<DanC_lap> * 2006-02-23T16:02:59Z Olivier Corby

<scribe> ACTION: AndyS to respond to [2006-02-24T08:45:23Z Olivier Corby] [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<DanC_lap> # Typos in SPARQL

<DanC_lap> * 2006-02-24T08:45:23Z Olivier Corby

<scribe> ACTION: EricP to work on [2006-02-24T08:45:23Z Olivier Corby], potentially automating examples2tests [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<scribe> ACTION: SvenG to follow up with Reto Krummenacher on [2006-02-24T16:40:55Z Reto Krummenacher] [recorded in http://www.w3.org/2006/03/03-dawg-irc]

lc-status-report

<scribe> ACTION: SergioT to respond to [2005-09-18T15:02:48Z Peter F. Patel-Schneider] [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<scribe> ACTION: AndyS to change "this is" to "RDF graphs are" in response to both [2006-02-22T20:32:56Z Peter F. Patel-Schneider] and [2006-02-22T23:56:54Z Peter F. Patel-Schneider] [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<LeeF> ACTION: Jeen to figure out whether he can become test case editor and report back by telecon on Mar 14 [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<AndyS> Scribe: AndyS

<LeeF> ScribeNick: AndyS

Process matters being considered

No support for UC&R on rec track

<scribe> Agenda: JSON

<LeeF> --> http://www.mindswap.org/~kendall/sparql-results-json/

JSON defined at http://json.org/

Variable name used in hash tables for solutions

Discussion on whether it should a W3C WG Note

<scribe> ACTION: Elias When finialized, email the WG for review as possible WG Note [recorded in http://www.w3.org/2006/03/03-dawg-irc]

Review offered by: AndyS, Sergio, Steve

</json>

Back to tests:

Review testing actions in the agenda

* ACTION: JosD to make test case from Sergio's basic query patterns examples. progress

* ACTION: EricP to fix test schema to match manifest with negative tests [recorded in 09/27-dawg-minutes.html#action16]

* ACTION: DanC to follow up re optional test based on op:dateTime triple

* ACTION DaveB: add to test suite the temperature case from comment on truth tables in commentor's message

JosD action overtaken

<ericP> ACTION: JosD to make test case from Sergio's basic query patterns examples. progress [WITHDRAWN] [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<ericP> action -15

Sergio to email EricP with test case files

<SerT> ACTION: SerT to make test case for bnode scope in different BGPs <http://lists.w3.org/Archives/Public/public-rdf-dawg/2006JanMar/0021.html> [recorded in http://www.w3.org/2006/03/03-dawg-irc]

Test Schema

EricP outlines changes to testing format for bad syntax queries

Change incolves formalizing the idea that an action is a blank node always

Discussion on structure of testing vocabularies

There are two vocabs: one for manifests/actions , one for query tests

Discussion around negative syntax tests

and the modelling in the manifest

<ericP> http://www.w3.org/2001/sw/DataAccess/tests/data/SyntaxDev/Syntax-SPARQL3/manifest.ttl

SyntaxDev/* tests are a donation of the ARQ tests that test the parser generated from the javacc used to make the BNF in rq23

ls

<ericP> ACTION: EricP to fix test schema to match manifest with negative tests [WITHDRAWN] [recorded in http://www.w3.org/2006/03/03-dawg-irc]

EricP: Negative tests not on the critical path to CR

<ericP> action -17

<ericP> that was EricP echoing SteveH

<ericP> ACTION: DanC to follow up re optional test based on op:dateTime triple [CONTINUED] [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<ericP> ACTION: DaveB: add to test suite the temperature case from comment on truth tables in commentor's message [WITHDRAWN] [recorded in http://www.w3.org/2006/03/03-dawg-irc]

<ericP> action -19

</test actions review>

<ericP> ADJOURNED

<EliasT> http://www.json.org/java/index.html