- From: Steve Harris <steve.harris@garlik.com>
- Date: Tue, 2 Oct 2012 15:12:43 +0100
- To: Andy Seaborne <andy.seaborne@epimorphics.com>
- Cc: public-rdf-dawg@w3.org
On 2012-10-02, at 14:57, Andy Seaborne wrote: > > On 02/10/12 14:43, Steve Harris wrote: >> I would prefer that 3/ have two bNodes, having the label scope cross ;s strikes me as a bit strange (just visually/aesthetically). >> >> It would be nice to be able to dispatch INSERT DATA … ; INSERT DATA … in parallel, though this is not a particularly strong use-case. > > The other option is to ban use of the same label in different operations (carefully - DATA only). I have not checked the effect on the test suite. An other option? If bNode labels scope is stopped by ; (as in INSERT … WHERE) then there's no co-ordination needed. > But that would make parallel processing needing communication for the same reason that coordinating "same label means same bNode" needs communication. [*] It can be (pre-)parsed in one location, the operations can be carried out distributed. We do this with Keepalive PUTs, but not Updates currently. We have to pre-parse in any case to ensure that the operations are orthogonal, e.g. all INSERT DATA {} (or theoretically INSERT WHEREs that can't have cross-effect each other). > Andy > > [*] Actually, "same label - same bNode" does not need communication depending on the implementation of internal ids. Yes. - Steve > Error detection in parallel parsing does regardless. > >> >> I believe we sometimes glob small updates like this, and it would require us to be careful with bNode labelling in that case. >> >> I also find Olivier's argument about 3/ and 4/ being equivalent to be pretty persuasive. >> >> On the other hand PREFIXes span multiple ; blocks (to some extent), so there's a bit of precedent. That decision hasn't been wildly popular though. >> >> I wouldn't object to either decision, though I might reserve "told you so" rights til the next WG :-) >> >> - Steve >> >> On 2012-10-02, at 11:22, Andy Seaborne wrote: >> >>>> Concerning insert-data-same-bnode2.ru I prefer that the scope of bnode >>>> _:b be one insert data and hence that the generated bnodes be different. >>>> The reason is that if you execute the two insert data using one query or >>>> using two queries, the result is the same. Also it is more uniform with >>>> insert where. >>> >>> GRAPH is not a factor so let's simplify a bit: >>> >>> Consider >>> >>> 1/ >>> One operation >>> Two uses of a bNode label in one INSERT DATA >>> >>> INSERT DATA { _:b :p :o1 . >>> _:b :p :o2 . } >>> >>> 2/ >>> One operation >>> Only one syntactic mention of of _:b >>> >>> INSERT DATA { _:b :p :o1 ; :p :o2 . } >>> >>> 3/ Two operations, one request >>> INSERT DATA { _:b :p :o1 } ; >>> INSERT DATA { _:b :p :o2 } >>> >>> 4/ Two operations, two requests >>> >>> Request 1: >>> INSERT DATA { _:b :p :o1 } >>> >>> Request 2: >>> INSERT DATA { _:b :p :o2 } >>> >>> >>> Cases (1) and (2) have the same bNode and because of (2) they must be the same bNode. And it's that way in Turtle and SPARQL 1.0. >>> >>> Cases (1) and (4) must be different. >>> >>> We are left where the change happens - is (3) like (4) or (1)/(2)? >>> >>> We already distinguish requests as atomic ("SHOULD be"). So across two requests, other things can change. Requests are also the unit in the protocol and in the grammar. >>> >>> I see (3) as just a different syntax, like (2) is to (1) hence I prefer (3) to be be like (1). The rule is labels are scoped to the document (which is the request) rather than make an "operation" a significant unit. >>> >>> This is not a "must be" technical decision - it's about style. >>> >>> It also means ";" works like concatenating Turtle files in regards to label scoping. >>> >>> Andy >>> >>> >>> >> > -- Steve Harris, CTO Garlik, a part of Experian +44 7854 417 874 http://www.garlik.com/ Registered in England and Wales 653331 VAT # 887 1335 93 80 Victoria Street, London, SW1E 5JL
Received on Tuesday, 2 October 2012 14:13:17 UTC