RDF Framework
Dave Beckett

1. Overview

  XML Parser Factory - via pipe/filter, expat, ...
  XML Parser
  RDF Parser Factory - via pipe/filter, libwww version
  RDF Parser
  Digest Factory - MD5(*), SHA1(*), RIPEM160(*)
  Digest(*)
  Hash Factory - GDBM(*), berkeley DB v2, In Memory
  Hash(*)
  Node(*) (for Resource / Literals)
  Statement
  'Set of Statements' / DataSource? <- storage link here?
  'Set of Set of Statements' / Set of DataSources/ Model? <- querying here?
  'Storage Model'

(*)implemented

2. XML Parser Factory Class

A factory that returns XML Parser objects that implement the interface.

Implementations:

• via pipe/filter to external program
• expat
• ...

3. XML Parser

An object that implements the XML interface

4. RDF Parser Factory

Implementations:

• via pipe/filter to external program
• libwww version
• ...

5. RDF Parser

6. rdf_digest_factory - Digest Factory

Implementations:

• MD5(*)
• SHA1(*)
• RIPEM160(*)
• openssl digests (above)
• ...

7. Digest(*)

A digesting function is a mathematical function operating on a series of octets/bytes returning a digest (sequence of octets) that represents it with defined levels of confidence over how unique this is

8. rdf_hash_factory - Hash Factory

Implementations:

• rdf_hash_gdbm - GDBM(*)
• berkeley DB v2
• In Memory
• ...

9. rdf_hash - Hash(*)

A mapping of keys to values, also known as an associative array.

10. rdf_node - Node(*) (for Resource / Literals)

11. Statement

11.1. Issues

If two agents assert the same 'statement' (aka p/s/o triple) into an RDF datastore, do we model that as:

1. a single statement with two 'asserting contexts' associated with it
2. as two statements having the same content.

Former says Dan

The math language we used makes it seem wrong to attempt to have the same statement appear in S twice; however pragmatics of real world apps means we'll want two of _something_ to be stored.

An asserting context is what?

Dan: More than a count - we want to be able to retract everything from http://rdfwebring.org/ldab/rdfweb/webwho.xrdf and replace it with what we get from it on a later visit. So there's likely all sorts of things we'll want to track. I don't believe it's RDF's core business to specify these -- different apps will store different things. Over time some common cases will emerge; right now, we don't have enough experience to know how be

Maybe the context would be a URI refered to by a util:gotFrom relation. e.g. If it was got-from an HTTP transaction there's a whole bundle of information (date, time, auth, content-neg prefs etc)

12. 'Set of Statements' / DataSource? <- storage link here?

13. 'Set of Set of Statements' / Set of DataSources/ Model? <- querying here?

14. 'Storage Model'

20. Other Issues

• 1. Need to model all RDFS defined classes in order to implement M&S?
  

  Do you need need to model all the M&S concepts in order to support schemas? e.g. what if I want to query for all resources with an rdf:_23 property but have implemented a bag as a hash. Each content of the bag contains a property rdf:_1 to rdf:_30 which have to be accessible in the graph if people ask for them. If not, then the model has been changed.

• 2. Can you actually infer the number 14 from rdf:_14
• 3. Can bag/seq members have 'numbers' rdf:_14 from non-RDF namespace?
• 4. automatic renumbering of bags - is this allowed?

  Dan pronounced: "the semantics behind the assignment of numbered container-membership properties (ie. rdf:_n) to members of RDF containers (Seq, Alt, Bag) is application specific. Generic RDF processors should make no assumptions about continuity of these numbered properties (eg. rdf_1, rdf_2, rdf_34223 is a reasonable sequence). While some RDF application contexts may support or employ automatic renumbering of these container properties, other applications are likely to employ 'gappy' sequences of numbered container membership properties (eg. to maintain a correlation with another sequence elsewhere, such as street numbers). RDF processors should not make closed-world assumptions about container membership (ie. a description of a Bag/Seq/Alt might well provide only a partial unumeration of its members). RDF processors can make use of additional information provided in RDF to understand more accuraately the semantics associated with particular uses of the basic RDF container constructs."

• 5. track namespaces of URIs for use on output?
• 6. do we need an isReified AND an isAsserted flag or just one (per statement)?
• 7. do we need isReified and/or isAsserted booleans in statements
• 8. is it so important to be able to ask classing questions sub/superclass that have lots of depths, that the typing support should be delivered with the framework?