SASSY has hit a wall, part 3

So, how to proceed?

RDF is a foundational part of this version of SASSY. It is possible that other graph databases might get used in the final version, but for now its RDF. It needs to be used correctly or it will be an unending source of problems.

This implies that probably all the blank nodes need to be replaced by resource nodes (URIs). These I will call anonymous nodes.

The plan:
+ Decide on the representation for these anonymous nodes. Sequence numbers or UUIDs seem to be the obvious alternatives.

+ Update my RDF library to create these nodes.

+ Write a program to turn the existing blank nodes into anonymous nodes.

+ Update all the SASSY projects to use the new code.

3/3

#SASSY #RDF

SASSY has hit a wall, part 2.

The rule engine uses SPARQL to find the data the rules are to be applied to. Blank nodes break this design.

When I first started SASSY I avoided blank nodes. As time went on I realised that generated names for nodes was going to save a lot of time, so I used them in lists and other places where names did not add any real information to the model.

My initial design for generated names was OK for toy examples, but was not going to work well for a multi-user distributed system. Hence I moved to using blank nodes that have identifiers which are mini-UUIDs.

So the early version of the rule engine worked OK since it used generated names. It is only the more recent data that uses the blank nodes, and the world fell apart.

2/n

#SASSY #RDF #SPARQL

SASSY has hit a wall.

I was debugging the rule engine, finding easy bugs at first, and getting steadily harder to resolve as work proceeded. This is normal.

In the end the results were really strange with some odd messages from the underlying RDF library.

It seems I have misunderstood a fundamental component of RDF, the blank node.

The blank node is often described as being useful in things like linked lists or other structural uses where naming the node with a URI would seem redundant. This leads to problems.

The correct use is as a placeholder for when the identity of the node is not known. For example, the perpetrator of a crime before a suspect has been identified. There must be somebody, but they are unknown.

Mostly both uses work. The gotcha is SPARQL. You cannot give it a blank node to search on. If you do it treats it as a variable to be bound to some value.

1/n

#SASSY #RDF