WebDataInterfaceDesign

From W3C Wiki
Jump to: navigation, search

There is now a fair amount of RDF/XML data around, thanks to various SeedApplications and other Semantic Web efforts.

How do programmers deal with this in code? What is it like to be an RDF coder? How do RDF toolkits compare to the machinery familiar to XML developers, such as SAX, DOM, XPath, XSLT...?

hmm... is this an API design discusion, aiming at convergence, or just a catalog, something for? need better PPR:OpeningStatement

See also: RDFAccessProtocol for network protocol ... er... network protocol something... ugh... naming cop-out..

What RDF APIs can I use?

What OWL APIs can I use?

Tasks

Many RDF toolkits offer similar facilities. In-memory and persistent representations of RDF graphs (often but not always SQL-based in the latter case). Parsers for the RDF/XML syntax, often a serializer too. Often a graph-oriented API, sometimes with richer textual query interfaces.

This is an attempt (partial/incomplete) to list some common tasks that occur when building Semantic Web applications, so that we can walk through the way the different RDF tools address these needs.

When learning a new RDF package, we might ask ourselves:

"How do I...?"

  • load a graph from a URI
  • create a new SQL-backed RDF store
  • stash some RDF there
  • refresh a named portion of the DB with this graph
  • serialize to RDF/XML
  • get a graph that matches (some blanked out statement)
  • do fancier queries
  • create a... statement... graph... term...

Language integration

One pattern is to make APIs computer language independent so that the same libraies can be shared by users of many languages. Another is to make the interface use as many nice features of the language as possible. When using RDF seriously, one tends to need access to the properties of things just as often and just as easily as access to the slots in an object. This suggests a value for a specialized language-specific interface. Also, one needs a neat way of using namespaces.

For example, in python, which is object-oriented and allows the object model to be bent in all kinds of ways, namespaces can be used as though they were dictionaries (rdf['type']) or objects (rdf.type) and objects in an RDF store can be represented by python objects.

There have been various implementations of this in python, including Mark Nottingham's Sparta, Aaron Schwartz's Tramp, Andrew MK's thoughts and the Namespace class in cwm.

To take this to a logical extreme, one makes a new language, a mixture of a declarative RDF langauge (like N3) and a procedural language (like python). Adenine is an example.


genesis: Apr 2003 discussion among DanBri, DanC, eikeon.