Data Access Working Group

5 Jul 2005


DanC, HowardK, AndyS, EricP, Kendall_Clark, Souri, LeeF, SteveH, Elias, JosD
Jeen_Broekstra, Dave_Beckett, Yoshio_FUKUSHIGE (福重貴雄)
Eric Prud'hommeaux


  1. Convene, take roll, review records and agenda
  2. SPARQL QL publication
  3. blank node handling... new requirements?
  4. SPARQL protocol publication
  5. SPARQL results set publication

See also: proposed agenda, IRC log,

Convene, take roll, review records and agenda

PROPOSED accept 28 Jun minutes as a true record


Next meeting: 12-July

Next scribe: SteveH

continued without discussion:

SPARQL QL publication

<scribe> ACTION: EricP add a note that users should be aware of non-canonicalization of IRIs [DONE]

ericP: see warning about identical-looking IRIs

<DanC> ACTION: EricP to refine definitions extraction [CONTINUED] [recorded in http://www.w3.org/2005/07/05-dawg-irc]

<DanC> ACTION: EricP clarify which regex lang, new section ericp; have AndyS check it. [CONTINUED] [recorded in http://www.w3.org/2005/07/05-dawg-irc]

<DanC> ACTION: PatH to review new optionals defintions, if any [CONTINUED] [recorded in http://www.w3.org/2005/07/05-dawg-irc]

<DanC> ACTION: DanC to write SOTD; work with EricP to publish [CONTINUED] [recorded in http://www.w3.org/2005/07/05-dawg-irc]

Andy believes he can produce OPTIONALs text this week in time for PatH to review it

blank node handling... new requirements?

KendallC: we care about bNodes issue, but don't care that it is before last call

<DanC> example from DanC in bnodes

DanC: at first, I thought bNodes was an overspecification problem. Now see that a test case reveals that they want a different definition of matching.
... (bnode rich stuff can be hard to deal with, yes. http://esw.w3.org/topic/IdentifyEverything advises "don't do that". hmm.)

ericP: (multiple identities for everything causes FOAF database bloat)

<kendall> yes, i agree about "don't do that", but FOAF is kinda a big deal!

[KendallC describes UM's interest in stable bNode identifiers]

KendallC: I don't want to make a decision here that will make FOAF and OWL/DL queries harder in the future

PatH: we might be able to allow bNodes to pin down a match without requiring it

<DanC> ron's reply regarding the test case

KendallC: I believe that is what we want. This affects using SPARQL between portals

<DanC> input data: _:l23c14 foaf:mbox <mailto:connolly@w3.org>.

PatH: in the past, we tried bNodes not treated as variabls
... requires that you put a variable there instead.
... removes bNodes from the QL entirely
... backing off, you can send a query with a bNode but it might not match

Proplems with removing BNodes from SPARQL:

1. SELECT * gives more bindings

2. need to rename named bNodes and []s when translating turtle to SPARQL

<DanC> in _:l55c33 , it's an ell, not a one. line 55 character 33.

<AndyS> 3 ways round it

PatH: how about some syntax for "marked" bNodes?

<SteveH> yes

PatH proposes something that DanC notes appears to match Andy's _!:xyz proposal

AndyS: protocol solution is also interesting 'cause you're in a session context

<AndyS> We also need to relax XML results format (and RDF/XML??)

<kendall> but that's *one* implementation strategy among others; I don't see any reason to privilege it.

Elias: if we already have the capability, with OPTIONALs, is inventing this stuff necessary?

<EliasT> I meant: what else is going on outside our group that deals with bNodes across RDF documents...

<kendall> Elias: and my answer is foaf & owl dl :>

AndyS: I don't find the FOAF example so compelling because you can always use mbox or mbox_sha1 [FOAF IFPs]
... I find update more compelling

<EliasT> I know FOAF, OWLDL uses bNodes, but I don't know of toolkits that can tell you that two bNodes point to the same person... or for that fact take a bNodes as an input to their query.

<DanC> EliasT, there are a number of such toolkits. cwm has a "smushing" mode, for example.

PatH: I find asking the server to create URIs which it is willing to author more appealing

LeeF: that puts a slightly larger burden on the servers that *do* offer bNode stability

AndyS: can be done with a simple map

KendallC: URI label space solution makes me nervous as I don't know the implications on OWL/DL
... protocol solution is interesting. want to think about it.
... glad we have a record of discussing this.

<AndyS> Within rq23: FILTER ext:bnodeLabel(?x , "label") - it's mildly cheating

<DanC> DanC suggests we go with the design we have, with some flexibility about new information later.

<kendall> it's all well and good (seriously) to suggest that FOAF allows URIs instead of bnodes, as well as definining mbox as IFP, but it doesn't seem like OWL DL has that flexibility.

<SteveH> that doesnt mesh well with CONSTRUCT

<kendall> It should be possible to query RDF vocabularies that require heavy use of bnodes in a user-friendly manner. ?

<DanC> right, kendall, I think OWL DL does not (though I'm never quite sure without looking it up)

<kendall> eh, that sucks, but ??

PatH: my intuition is that mapping to URIs has the same implications as re-using bNode labels.
... but have to think about that hard

DanC: the effect of adopting Kendall's requirement is that the WG will spend weeks considering a technical solution

JosD: in my experience, I query billions of bNodes and lists and I haven't seen this problem come up.

<SteveH> variable length lists are tricky

<kendall> jos, i'd welcome you writing an email explaining yr experience in this regard

<LeeF> DanC, does that wording place a requirement on servers to support this, or only on the QL to allow clients to ask queries hoping that the server supports it?

<kendall> ericP, i thought we found language in the present spec that does *not* allow that presently

STRAW POLL: who is in favor of a requirement "it must be possible for a client to refer to a bnode provided by a server"

0, -1, -1, +.5, +.5, +.5, +.5, +.5, -1, +1

<kendall> eh the weakly = .5 thing should be non-canonical, IMO :>

<patH> Jos, consider a list of children. I query the common children of Mary and Bill, and I get back a bnode indicating the list, which tells me that some children exist. I want to ask what is in the list. How do I refer to it?

<DanC> do the same query, pat, and add more to it

<patH> I think the problem we ahve here is that several folk do not see that there is a real problem. Suggestion: if non-idiotic users (Maryland) say thery have a problem, there really is a problem.

<kendall> fwiw, i don't know what design instantiates that distinction! :>

<DanC> howardk is excused

<patH> Add what, Dan? Do I have to put the RDF list syntax into my query? (Yech)

<DanC> yes, yech, but it works, pat

<patH> OK, sorry, I shuld know better than to say "yech" in an RDF context.

<SteveH> always mapping bnodes into uris is messy

<SteveH> having it be an option would be more acceptable IMHO

<kendall> steveh: yes, i understand this as "you may do" instead of "you must do"

<AndyS> Would _!:xyz cover that?

<JosD> Pat, it is pretty convenient with the ( list ) notation

<kendall> Andy: splitting the bnode label space seems a variant of bnode->uri

<AndyS> Sort of - but there are not forced to be URIs by design

<SteveH> kendall, I meant optional at runtime, sorry wasnt clear

ericP:Of the +5s and 1s, who wants this in this version of SPARQL vs. making sure we don't make it difficult in the next version?

<scribe> ACTION: PatH to consider implications of answering bNode bindings with created URIs [recorded in http://www.w3.org/2005/07/05-dawg-irc]

<scribe> ACTION: KendallC to ask Bijan to consider implications of answering bNode bindings with created URIs [recorded in http://www.w3.org/2005/07/05-dawg-irc]

AndyS: ericP, What about the other designs?

ericP: AndyS, I only pushed on one of them, the one that I saw as most immediate

<SteveH> JosD, you cant use the ( list ) notation in that case because of the :nil URI

<DanC> editors still working on optionals

<AndyS> I'd like protocol considered because the requirment was for session usage

<scribe> ACTION: JosD to fix up the relevant tests [recorded in http://www.w3.org/2005/06/28-dawg-minutes.html#action01] [CONTINUED]

<JosD> SteveH, what is meant was e.g. query ... { ?X owl:interscetionOf (:a :b :c) ...}

<DanC> syntax-qname-08-rq and syntax-qname-14-rq

<SteveH> JosD, ah, sorry, I thought you were talking about the unknown list length case

<SteveH> JosD, eg. find all the classes that this class is the intersection of, but you dont knwo how many there are

<DanC> WHERE { :a. x.: : . } <- old or new bits?

<DanC> old

SPARQL protocol publication

KendallC: still have a short todo. spending time on query lang.
... no complaints apart from Mark Baker

SPARQL results set publication

DanC: what's standing in the way on Results Format?

EricP: we needed a namespace document. done. don't know what else is critical path.

<kendall> I don't. :>

<DanC> re results format, EricP notes an outstanding comment about xsi:type

<DanC> xsi:type on sparql:literal elements

<AndyS> Dave recommedned I remove xsi:schemaLocation=... from example in rq23

<SteveH> I'm happy to add some results test, but it will be at least a week before I can do it

<DanC> (ericp, in the minutes, please continue my 2 actions under the Comments item)

<scribe> ACTION: DanC to put a doc at the new results format namespace [DONE]

DanC: can one put spurious extra attributes without changing the meaning?


EricP and DanC
$Revision: 1.6 $ of $Date: 2005/07/08 16:55:52 $
formatted by David Booth's scribe.perl version 1.126 (CVS log)