RDB2RDF Working Group Teleconference -- 08 May 2012

<trackbot> Date: 08 May 2012

I can scribe

scribenick boris

<scribe> TOPIC : Aprove minutes of last meeting

http://www.w3.org/2012/05/01-RDB2RDF-minutes.html

RESOLUTION: Last telecon minutes approved

Implementability for tables w/o primary key - DM

Ashok: there are a few mails about this
... also proposal from Ivan
... please other members of the group please speak about this ....
... other possibility is to do nothing

cygri: summarizes this issue from his perspective
... issue1 : DM hard to implement in some situations
... issue2 : not compatible with R2RML
... Richard is happy with the solution of issue1

issue2: DM can be expressed in R2RML ...
... in tables without PK
... one option is do nothing
... other is document it
... another option just acknowledge
... other option is fix the incompability from R2RML or DM
... Richard suggest to do it from the R2RML
... this is brand new feature, no other rdb2rdf are able to this
... we do no't have time to this ....
... fixing from the DM
... in long run ...we have to do it in R2RML

Ashok: is the wording (of Ivan) ok to fix you

cygri: ... no

<juansequeda> It is not a corner case if we consider Views!

cygri: as R2RML implementor, we want to have a DM ...
... the purpose of DM is provide support to R2RML
... current DM behaves different in some corner cases
... people is asking if we are implementing DM and R2RML
... we cannot implement the current DM spec

Ashok: so, we have to add a feature to DM?

cygri: if we'll fix in R2RML is a new feature
... if we'll fix in DM we have to change the wording .... is not so much
... it's so not clean solution ...

Ashok: any comments from the group?

ivan: the way of looking to the DM is no the complete picture, Richard
... it's not 100% necessarily
... if we follow your proposal, there'll be several versions of the DM

<Ashok> Who is 2714444?

ivan: the only way to do that is going beyond to that propose ... DM is not using fresh bnode labels

<Ashok> Sorry, 2714@@@@ ?

ivan: other way is to introduce a fuzziness, but if we later fix R2RML, current DM will be invalid in the future
... this is dangerous

<juansequeda> I'm curious to hear Souri and Seema

<Marcelo> 2714444 is Marcelo

<ericP> mysql --user=genome --host=genome-mysql.cse.ucsc.edu -A

ericP: let's a use case

<ericP> show create table uniProt.gene;

ericP: showing the use case ...
... current R2RML does not capture that table ...
... not cardinality repeated rows
... DM captures it ...

<Zakim> ericP, you wanted to outline the use cases

ericP: start from DM and "changed" to R2RML

Souri: DM purpose is a kind of starting point for R2RML ....

<juansequeda> but not all databases have row ids

Souri: if a table doesn't a PK, if we are doing DM ... if we are able to capture the row number (well it depends on the DBMS) ... we capture the unique id

<Ashok> Juan, you can always join a column with an ascending number

Souri: why do we automatically convert that into a column
... it can work in Oracle, probably difficult in other dbms
... we can do a Bnode, or URI ... it's different issue

juansequeda: Souri's proposal is similar to Juan/richard proposal
... initial proposal

<Souri> please note: I am talking about ROWNUM in Oracle DB, not ROWID (b/c view does not have ROWID)

cygri: ROWNUM unless you materialize
... ROWNUM doesn't work without fetching the table
... ROWNUM changes every time, so it doesn't work in general case

<Zakim> cygri, you wanted to say we'll get multiple versions of the DM *anyways* if we don't fix it

cygri: other dbms do not have ROWID
... in Oracle is easy ...but in other dbms is not the case
... also postgress is also possible
... regarding ericP comments

<Souri> example=> select rownum, e.* from EMP e;

cygri: R2RML doesn't change the semantics of the db, is issue of the mapping not of the language
... DM reflects what is on the database, R2RML is what you get from it ...
... by allowing the alternative behaviour of the DM, there won't be use case for R2RML ... Richard is not ok with this

<ericP> =IOUs=

<ericP> Bob 20

cygri: table no pk no duplicate row, I can write an R2RML that captures the semantics of that and matches the DM

<ericP> Bob 10

cygri: there'll be a use case for this ...
... regarding the comment of having 2 versions of the DM

<ericP> an R2RML mapping which presumes cardinality of 1 could use -{name}-{amount} can capture all of the rows

<MacTed> can we strawpoll on the two proposals, as shown in the agenda?

cygri: it would be worth to have default mapping ... having to fork w3c spec in order to be compatible with the other w3c spec is worse

<MacTed> (or revised to capture all the conversation on the list, which I think did modify both...)

cygri: comment about how it is now in the spec

ericP: a table without a unique key ... but there is one ....
... there should be a warning if it's going to create duplicate row
... spec that remove some parts of the previous specs exist
... the solution is cheap ...

<Souri> Example of deterministic association between ROWNUM and row-content (in Oracle DB): select rownum erownum, e.* from EMP e order by e.empno, e.ename, e.job, e.mgr, e.hiredate, e.sal, e.comm, e.deptno;

Souri: showing an example how to do it in Oracle
... ROWNUM is always there, view, table
... for some users can go ahead and create a ROWNUM ...

<cygri> souri, ROWNUM does not work.

Souri: this is deterministic ...
... if we have duplicate rows there won't be no problem

Ashok: we have 2 positions

<Souri> Richard, please give an example where ROWNUM does not work

Ashok: position 1. Ivan wording ...
... position 2. to modify the DM

<cygri> souri, 2-row table, join the table onto itself. you end up with 4 rows, but only 2 blank nodes. ROWNUM will give you 4 values and no way to work out which are the same

Ashok: could you live with position 1.?

<ericP> +1

<Souri> ROWNUM corresponds to the result rows only

<ivan> +1 (of course…)

<cygri> -1

<MacTed> +1

<Marcelo> +1

<Souri> +1

<Seema> +1

juansequeda: please paste Ivan wording

<ivan>

ericP: explaining Ivan wording ...

<Souri> ROWNUM does not care about the nature of the query, only the order in which results are coming out

<MacTed> I'm sorry, I didn't catch that Agenda order was reverse of presentation now :-(

<juansequeda> -1

<MacTed> -1

ericP: option 1 <-> Agenda option B

<cygri> souri, i'll send an example via email

option 2. <-> Agenda option A

<Ashok> Nuno -1

Marcelo: I can live with the 2 options ...

Ashok, 4 people with -1

MacTed: rephrase the options

<MacTed> STRAWPOLL: who can live with A, B, or AorB, as in the agenda?

<MacTed> +A

<ericP> +B

<Ashok> Ashok +B

<Marcelo> AorB

<ivan> +B

<Seema> AorB

<cygri> +A

AorB

<nunolopes> +A

<juansequeda> +A

<Souri> +AB

MacTed: 6 for A (include AB)

Ashok: how are we going to do some progress on this?

ericP: let's figure out scenarios with the text of the specs ...

Ashok: ericP volunteer to do that

<Zakim> ericP, you wanted to propose using use cases to illustrate our choices

ivan: more pesimistic than ericP/Richard
... the only clean way forward it's to solve the technical issue in R2RML ...
... it can be a trivial change or half-year change
... we need a clear picture of what it means ...
... we are already in second round LC
... how much time we need to fix this in R2RML?

cygri: there's a proposal to change this on R2RML
... this is not implemented in the current rdb2rdf implementations

<Souri> Suggestion: use rr:rownum in R2RML as a pseudocolumn (how it is generated will vary from DB to DB)

cygri: I dunno what it means in terms of time ....

ivan: we can have feedback for the current R2RML implementations in the near future ...
... about this issue
... probably before we finish LC

Ashok: ask to R2RML editors to write in the spec the solution
... and implementors to report this later on

<Zakim> ericP, you wanted to say that the quickest way to implement that is with a materialized R2RML implementation. two of those good enough?

<Souri> I'd suggest rr:rownum as a pseudocolumn

Ashok: two action items: ericp create the uses cases, and R2RML editors say how the R2RML spec will change

ericP: proposing a quick way of implement this

juansequeda: the issue is beacuse of the rewritting

Souri: materializing is not going to help us ....
... in this particular issue

<cygri> The mapping language defined by the WG will facilitate the development of several types of products. […] It could be used to generate a virtual mapping that could be queried using SPARQL and the SPARQL translated to SQL queries on the underlying relational data.

cygri: materialize option doesn't work for users ...
... explaining what is in the charter
... propose a concrete change on the R2RML spec and implementors say what they think about that

Ashok: Richard will be involved in both actions

<Souri> consider a new pseudocolumn: rr:rownum

<Marcelo> I have to leave ...

AShok: this week for doing the actions

<nunolopes> :)

<Ashok> Boris, Thank you for scribing ... good mimutes

you are welcome Ashok, however the list of Attendees is empty

I dunno how to fix it

<Ashok> present, Ashok, Ivan, Richard, Nuno, Boris, Marcelo, Ted, Juan, Souro, Seema, Marcelo, David

<Ashok> I will take care of it

ok, thanks Ashok

<Ashok> regrets, Michael

<Ashok> rrsagent generate minutes

<Ashok> regrets Michael

- DRAFT -

RDB2RDF Working Group Teleconference

08 May 2012

Attendees

Contents

Implementability for tables w/o primary key - DM

Summary of Action Items

Scribe.perl diagnostic output