Several new attribute grammar dialects have recently been developed, all with the common goal of allowing large, complex language translators to be specified through a modular composition of smaller attribute grammars. We refer to the class of dialects as hierarchical attribute grammars. In this short article, we present a characterization of optimal incremental evaluation that indicates the unsuitability of change propagation as the basis of an optimal incremental evaluator for hierarchical attribute grammars. This result lends strong support to the use of incremental evaluators based on more applicative approaches to attribute evaluation, such as Carle and Pollock’s evaluator based on more applicative approaches to attribute evaluation, such as Carle and Pollock’s evaluator based on caching of partially attributed subtree, Pugh’s evaluator based on function caching of semantic functions, and Swierstra and Vogt’s evaluator based on functions, and Swierstra and Vogt’s evaluator based on function caching of visit sequences.