Re: A certain difficulty

Walter Underwood wrote:

> [MCF] fails my basic architectural test (I think
> this is from Rob Pike): simple things must be simple, hard things
> must be possible, and you must be able to use part of it without
> understanding all of it. It's too complex for the simple things,
> too hard to explain how it helps the hard things, and non-modular.

It's a good thing you were talking about MCF.  I would disagree if you were
talking about RDF.

Simple things *are* simple.  There's just a lack of examples to demonstrate
this.  Optimally efficient?  No, but what general purpose interchange
standard is?

Hard things *are* possible, though often ugly as sin, or grossly inefficient
(Jeff Sussna's property ID case is a good example).

RDF may fail the using part without understanding all criteria -- lack of
full understanding of the model can get you into trouble.  However, I
believe it would be possible to define rules-of-thumb that people could
follow that would keep them out of trouble, but still get 80% of the utility
out of RDF.

When I help people who are having a hard time understanding the model, I use
this approach.  The interesting bits of the full directed graph can often be
transformed into ordinary trees without loss of information, each rooted at
a node that was originally identified with an rdf:about attribute.  Once the
model is recast into something people are more familiar with, its easier to
see how to do the simple stuff, and if needed, to tack on the extra stuff
(cross-cutting IDREFs, reification).

Perry

Received on Monday, 28 February 2000 16:48:38 UTC