[sc34wg3] Comments on latest TMRM draft

[Robert Barta]

On Thu, Jul 14, 2005 at 09:31:20PM +0200, Jan Algermissen wrote:
> >given that proxies will occur as values in other
> >proxies?
> 
> This is BTW a quite critical/interesting issue: Since values can  
> contain proxies[1], proxies that contain such values are tied to the  
> runtime environment that manages the proxies; they cannot have  
> standalone significance (as opposed to relational results (...or RDF  
> triples)). Will the proxies-in-values be object references? IDs? Will  
> such IDs be meaningful accross restarts of the system where they have  
> been obtained from? Are those IDs subject indicating properties then....

They cannot be references, at least not on the conceptual level. If a
proxy can be part of another proxy, then it is conceptually expanded.

Whether an implementation uses 'structure-copying' or
'structure-sharing' is then an implementation decision.

> I once read that part of the success of the relational model is that
> it did away with system dependent references...but that is surely
> another story.

It is related, because references mean 'state information'. State
information is difficult to capture in functional definitions (anyone
wants to learn about monads? :-). If it is not defined in terms of
functions, a lot of optimizations are not possible.

IOW, references (in the machine sense) open pandoras box, but...

> [1] These values form the edges of the graph, BTW. Since a value can  
> contain more than one proxy, a TMRM topic map is a directed  
> hypergraph. (In case anyone cares :o)

...but Tau+ is using identifiers. Which is sloppy from a pure
theoretic point of view as you might end up with infinitely large
structures. Not sure, whether this is a real issue.

\rho