What I think is more interesting, is a large general graph store, with a pattern matcher.

The tuple space is limited to only looking for matching tuples. But I think you should be able to enter in any query over a giant graph.

For instance, in your Instant Message example, there should be a node for “single conversation.” The individual message would be another node. The word “Joe” should be replaced by a link to the node for the whole person, Joe. (If you wanted to know that person’s name, it the node for the person Joe would have an attribute “first name” that linked to “Joe.”) The destination, in turn, links to the node for Frank. The time links to a temporal node, which may be a virtual node on a continuum. Finally, the message is an attribute that links to the particular string.

This is a MUCH more powerful model, because you can look at the graph, and say, “Show me everything that happened from time t1 to time t2,” and it will mesh with the entire system.

In fact, you can say, “Let me know whenever a message is posted between time t3-t4 where the target is Frank.”

Good question. The description of tuples as a vector of attributes looks just like del.icio.us. Bookmarks (i.e., resources) are the marbles. You can ask for resources by some tag vector:
[ 'Ruby', 'XML', 'Rinda' ]
and see what comes back.

One interesting difference is that the tags do not (currently, that I know of) have types. The Python example shows the use of type selectors (e.g., include as a criteria any Integer). Tags on del.icio.us (and probably all folksonomy sites) are unclassified. You cannot, for example, ask for all resource with tags related to database programming; something has to know that this request set includes more than just those resources explicitly tagged ‘database-programming’. You would likely also want resource tagged ‘mysql-programming’, and so on. (And this is the harder part: the metadata itself needs metadata to define such higher-level groupings. To get a richer navigation scheme, it would be good if tags/topics were assigned to facets. )

But even without such sophistication the use of tagged resources is very much like tuple spaces.

“this is the harder part: the metadata itself needs metadata to define such higher-level groupings”

This is what OWL does. For example, ((url),(topic),(mysql-programming))
is a “specialisation” of the tuple ((url),(topic),(database-programming)).

http://www.w3c.org for more.