Classical and incremental evaluators for attribute grammars

Abstract

The most interesting aspects of the present work are, in our opinion, the following two.

First, it gives a unifying view of ANC- and DNC-AG's and their evaluators allowing, in this way, to better understand the relation among different results concerning these classes.

Secondly, it stresses the role that the knowledge of the dependencies of an AG plays for the amount of recomputation of the corresponding evaluator. In this sense several interesting questions should be studied.

For instance, it would be interesting to device methods for transforming a given good (or fine) pop-assignment into a better one, i.e., producting a more efficient evaluator. Such process could consist in:

• (i) individuating where the recomputation may take place,

• (ii) merging 2 blocks of a po-partition of a nonterminal,

• (iii) adding new dependencies to a po-partition of a nonterminal, (as was done for TT(Y) at the end of section 2).