- From: Pat Hayes <phayes@ihmc.us>
- Date: Mon, 6 Dec 2004 22:52:30 -0600
- To: "Seaborne, Andy" <andy.seaborne@hp.com>, RDF Data Access Working Group <public-rdf-dawg@w3.org>
First, Im sorry Im so far behind, and missed the last telecon. (I thought I could attend through IRC, but Mark Greaves pulled the network plug during the DAML meeting. ) Second, I have to send regrets for tomorrow as well, unfortunately, and will make the next week only on a cell phone with no network connection (I should be back online by Wednesday of that week). I agreed to review the definitions in http://www.w3.org/TR/2004/WD-rdf-sparql-query-20041012/ but its taking a lot longer than I thought it would partly because I had a lot of trouble understanding the document. It seems to have a lot of internal contradictions (??). Anyway, some of my problems and comments and so on are visible in an annotated version at http://www.ihmc.us/users/phayes/SPARQL-PAT.html I started trying to edit the text (pink deletions and additions) but eventually I had to give up, as I couldnt figure out exactly what it meant. Hence the pink verdana-font comments. Sorry if these have a rather crotchety tone, they are surface-of-consciousness reactions to my frustration at trying to figure out the meaning. Having gone through the document twice I think I now have the general picture rather clearer; but if Im right, then I think the document needs a major re-write, as in its current form it keeps making definitions which it then almost immediately contradicts. Eg a pattern is not a set, in general. Anyway, Ive drafted a more positive summary of the way that the definitions and constructions might go. http://www.ihmc.us/users/phayes/SPARQLdefs.html This is bare definitions, minimal comments, no examples. Stuff in italics is comment rather than content. As per the verdana comments, there seems to be no sense in allowing patterns to have bnodes and distinguishing a subset of variables, since a pattern variable which is not selected plays exactly the same role as a bnode in the pattern would play. Im assuming therefore that patterns do not contain bnodes. If this is wrong its easy to tweak the definitions, but it only adds complexity and provides no extra expressiveness. General issues that emerged include: 1. Do we want to allow query variables to be placed on parts of literals, eg querying the type of a literal or asking for literals of a certain type? Or looking for languages: SELECT ?v WHERE (?x ?p "cat"@?v) (Why not?) 2. What is an answer to a query with a selected variable which does not occur in the pattern? 3. What are the rules for answering a limited-number-of-answers query? (do we get the NEXT n answers if we repeat the query?) 4. . Are constraints intended to be normative? The text in 3.1 seems to say not, but then why does the document spend so much time on them? 5. What does section 10.3 specify? As far as I can tell, it just says that the answer can be some arbitrary piece of RDF. OK, but who decides what is in that RDF? 6. There are several alternative forms for the answer to a SELECT query. Who decides what form the answer will take? Can the query specify this? How? If not, how can a querying app know how to process the answer? ------- Pat -- --------------------------------------------------------------------- IHMC (850)434 8903 or (650)494 3973 home 40 South Alcaniz St. (850)202 4416 office Pensacola (850)202 4440 fax FL 32502 (850)291 0667 cell phayes@ihmc.us http://www.ihmc.us/users/phayes
Received on Tuesday, 7 December 2004 04:53:56 UTC