5.2.1 Call Variance vs. Call Subsumption

The above description gives a general characterization of tabled evaluation for definite programs but glosses over certain details. In particular, we have not specified the criteria for

• Call Similarity - whereby a newly issued call $C$ is determined to be ``sufficiently similar'' to a tabled subgoal $S$ so that $C$ can use the answers from the table of $S$ rather than re-deriving its own answers. In the first case where $C$ uses answers of a tabled subgoal it is termed a consumer; in the second case when $C$ produces its own answers it is called a generator.
• Answer Similarity - whereby a derived answer to a tabled subgoal is determined to contain information similar to that already in the set of answers for that subgoal.

Different measures of similarity are possible. XSB's engine supports two measures for call similarity: variance and subsumption. XSB's engine supports a variance-based measure for answer similarity, but allows users to program other measures in certain cases. We discuss call similarity here, but defer the discussion of answer similarity until Section 5.4).