Re: PSI hashing and relational databases

Hi Alex,

I am a bit surprised that someone is still interested in Topic Maps, 
but why not… some want your soul, some want your first born child, some want 
Topic Maps; I think we should accept strange desires. ;)

Anyway, I think the current preferred strategy is to canonicalize any input and
to map the c14n to an internal identifier. The internal identifier could be
a hash or an integer or whatever fits. You'll end up with a bunch of singletons,
it doesn't matter if a string is used as PSI or as occurrence IRI, each 
string is matched to the same (internal) data structure.

Several (all?) RDF engines use this strategy, and some Topic Maps engines
use it, too. So it depends on your (internal) model if (i.e. a MD5) hash 
works for your model (after c14n) or an integer or a string, but singletons
may make sense if you want fast query answering and/or tests for equality.

Best regards,
Lars
-- 
http://www.semagia.com/