Re: ISSUE-33

On 11/02/13 18:37, Wilde, Erik wrote:
> it needs to be a stable graph-to-bytes function, so that the byte sequence
> is a safe thing to request parts of the graph. canonical xml
> (http://www.w3.org/TR/xml-c14n) has done that for xml, and there must be
> something for rdf somewhere, because without this, it's pretty much
> impossible to implement very basic utility functions such as hashes and
> signatures. but notive that this is a non-trivial thing to do: developing
> canonical xml took quite a bit of effort, because of the difference in
> abstraction layers between "just talking about trees", and the necessity
> to talk about resources you can use in basic crypto operations.

RDF is inherently unordered.  Systems make use of this e.g. storing 
using hash tables and it change to seemly unrelated changes.  Typically, 
the order does not change unless there is an update but that's small 
scale thinking.  Imagine a distributed store that pulls parts in parallel.

The way to sign in RDF is sign a document. There are other ways to sign 
at the abstract level but this structural analysis does not help paging.

What needs to be done is for paging to "freeze" as a series of the 
documents (freezing as one document is possible but less powerful 
without additional assumptions on, for example inline namespace 
declarations).

Then GET /resource returns Location: for the documents, which can be 
pulled as needed (or being very clever, generated on later access - 
implementation detail).  The documents can be transferred using any HTTP 
techniques.

It is important to "freeze" the document against updates and potential 
reordering.

	Andy

Received on Monday, 11 February 2013 19:53:37 UTC