- From: Cory Casanave <cory-c@modeldriven.com>
- Date: Thu, 22 Apr 2010 13:55:02 -0400
- To: "Erik Wilde" <dret@berkeley.edu>
- Cc: "Richard Cyganiak" <richard@cyganiak.de>, <public-egov-ig@w3.org>
Erik, I'm sorry but I find this highly bias, I'm not sure what you are trying to prove/disprove. [dret] You can use the GET/PUT/POST/DELETE primitives on an RDF graph just link an XML, HTML, or (some other) document. [cbc] Yes, there exists the SPARQL specification and I have written previously about the lack of connection between a SPARQL endpoint and the RDF graph. My suggestion is that all RDF resources should be able to respond to SPARQL at the same address, but that is not standard at this time. However, XML "has" xQuery with the same issues. But, we are talking about RDF - not every RDF related speciation. [dret] nope, the granularity is in the metamodel. XML has trees which are a first-class model structure. RDF has only triples [cbc] The graph IS the unit of granularity, and a graph can be huge or tiny just like an XML document can. There is nothing inherent in the RDF logical model or the XML model that specifies the granularity with which it will be used. You could have a graph that corresponds to a row in a DBMS or the entire DBMS. I am not, in any way, suggesting perfection of RDF. It should have a better connection between a URL and sparql and it should have more universal support for named, nested graphs. There should also be an established "post" interface for both sparql and graph update. These are all things for which there are existing solutions and ongoing work. But, every solution has issues it is just a matter of which one(s) we are going to vest in so that we don't have the information fragmentation problems we do now. Of course we can always invent new ones, but people are getting tired of that. -Cory -----Original Message----- From: Erik Wilde [mailto:dret@berkeley.edu] Sent: Thursday, April 22, 2010 1:37 PM To: Cory Casanave Cc: Richard Cyganiak; public-egov-ig@w3.org Subject: Re: [dcat] Tomorrow's dcat Agenda hello cory. > [cbc] Something can only be "self describing" if there is some > understanding of the structure that self describes which implies some > agreement on representation. Of course how much you agree on and the > level of variation can vary. No data "client" can understand any > logical data model unless there is describing meta-meta data, which is > then a data model. often, people are confused by the way in which "self-description" is meant in REST, which is not the same as in RDF. in REST, it just means that if i GET a representation, that thing will be labeled with a media type that will allow me to "understand" that representation. please no discussions around "understand" here, REST just tells you that you should communicate media types and those should allow clients (and servers) to somehow process instances of that type. > [cbc] Certainly RDF introduces constraints. I would classify it more as > a logical data model in that any meta model as any can be encoded into > RDF. you can always map something onto something else. you can map any RDF model into an XML model and vice versa. for some models this works nicely, for others, the mapping produces awkward results. what almost reliably produces awkward results are generic mapping schemes. we've written about this here: http://dret.net/netdret/docs/wilde-cacm2008-document-design-matters/#top ology > [cbc] I don't understand this assertion and don't see how RDF removes > any uniform interface constraint. REST/the web tells you to allow clients to interact with resources based on a fixed set of methods, which means if you have resources you expose them at URIs and can interact with those URIs using those methods. RDF currently has SPARQL, so the resources identified by URI do not expose the HTTP uniform interface *at the URI*. > [cbc] Agree that the "CRUD" standards for RDF are not done yet, but > there are CROD interfaces for RDF. You can use REST as much asa for any > web resource. It is not true that you cannot have non-read > interactions. However, there is no MORE or less CRUD capability for the > generic web interface. You can, however, use the normal REST style of > interactions and this works quite well. if you have a RESTful URI you can simply GET, PUT, POST, DELETE it. this works because URIs expose resources and allow interactions with resources. RDF has the problem that it only has triples, and it would take a substantial rewrite of the RDF metamodel (probably adding named graphs to the metamodel) to transition to something that would map well to REST. we've looked at that a little here: http://dret.net/netdret/publications#wil09i > because there is less granularity in the model, > [cbc] don't understand this either, the granularity is the choice of the > publisher. nope, the granularity is in the metamodel. XML has trees which are a first-class model structure. RDF has only triples, and while you can group them (doing the reification dance), that's not something that is supported in the metamodel or the interactions with it. > and there is no uniform interface. > [cbc] Think this is a false statement. not in the REST sense that there is a small set of well-defined interactions with URI-identified resources. cheers, dret.
Received on Thursday, 22 April 2010 17:55:35 UTC