SHACL rules infer new triples. The input is a data graph and a shapes graph which has rules; the output is a graph of inferred triples.

Whether, and how, the output graph is added back into the data graph is not part of the SHACL rules process. If being run to enrich the data graph, the process can be done inside a transaction updating the data graph.

The rule system is also capable of triple pattern matching - what datalog calls a "query" - where only necessary rules are run. (This is "backward chaining")

Once repeated application of rules is considered, there does not seem to be useful subsets of features. Fortunately, simple evaluation of (positive i.e. without negation) datalog is not a major piece of implementation. Run rules until no change; the implementation does not need to inspect the rule set structure. The literature has several better algorithms where implementation isn't partic