Clarification of DatasetGraph contract

[cygri]

I have some questions on the contract for DatasetGraph. What exactly is expected of an implementation of this interface? Let's say I want the implementation to work at least with ARQ's implementations of SPARQL Query, SPARQL Update, and the various dataset readers and writers.

• Can getDefaultGraph() return null? Can it throw UnsupportedOperationException?
• Can getUnionGraph() return null? Can it throw UnsupportedOperationException?
• Many (all?) of the provided implementations of DatasetGraph support the special graph names urn:x-arq:DefaultGraph, urn:x-arq:DefaultGraphNode, and urn:x-arq:UnionGraph. Is this required for all implementations of DatasetGraph?
• Is add(null, s, p, o) and remove(null, s, p, o) a valid way of adding and removing a triple from the default graph? (DatasetGraphCollection seems to support this usage, but I'm not sure it works with other implementations.)
• Is the following interpretation correct? The only difference between find(g, s, p, o) and findNG(g, s, p, o) is that a wildcard graph (null or Node.ANY) will also search the default graph in the former, but not in the latter. There is no difference between the methods with regards to the interpretation of the special graph names urn:x-arq:DefaultGraph, urn:x-arq:DefaultGraphNode and urn:x-arq:UnionGraph.
• In implementations that distinguish between empty and absent named graphs, would dsg.isEmpty() mean “all graphs are empty”, or “there are no named graphs”? In other words, after calling addGraph(g, anEmptyGraph) on an empty DatasetGraph, would an implementation be allowed to return false? Symmetry with size() and clear() would suggest that the answer is “yes”.

Thank you!

[afs]

• Can getDefaultGraph() return null? Can it throw UnsupportedOperationException?

There is always a default graph. This is defined in SPARQL - a dataset is a default graph and zero or more named graphs.

• Can getUnionGraph() return null? Can it throw UnsupportedOperationException?

There is a general purpose (and possibly inefficient) implement provided in DatasetGraphBase.

It is better to provide something (tests pass) but sometimes it might be meaning less and UnsupportedOperationException - with javadoc to explain why - would be appropriate.

• Many (all?) of the provided implementations of DatasetGraph support the special graph names urn:x-arq:DefaultGraph, urn:x-arq:DefaultGraphNode, and urn:x-arq:UnionGraph. Is this required for all implementations of DatasetGraph?

The DatasetGraph hierarchy (a tree of various styles of implementation) uses these - directly implementing DatasetGraph shouldn't necessary.

• Is add(null, s, p, o) and remove(null, s, p, o) a valid way of adding and removing a triple from the default graph? (DatasetGraphCollection seems to support this usage, but I'm not sure it works with other implementations.)

Yes.

• Is the following interpretation correct? The only difference between find(g, s, p, o) and findNG(g, s, p, o) is that a wildcard graph (null or Node.ANY) will also search the default graph in the former, but not in the latter. There is no difference between the methods with regards to the interpretation of the special graph names urn:x-arq:DefaultGraph, urn:x-arq:DefaultGraphNode and urn:x-arq:UnionGraph.

find includes the default graph , findNG does not.

• In implementations that distinguish between empty and absent named graphs, would dsg.isEmpty() mean “all graphs are empty”, or “there are no named graphs”? In other words, after calling addGraph(g, anEmptyGraph) on an empty DatasetGraph, would an implementation be allowed to return false? Symmetry with size() and clear() would suggest that the answer is “yes”.

All graphs are empty including the default graph.

Few implementations (just one now for legacy?) distinguish between empty and absent named graphs and this should not be relied upon. Best to plan around empty graphs not being preserved.

[afs]

Assuming the response 7/7/2022 answers the question.