Partial Evaluation for Constraint-Based Program Analyses

We report on a case study in the application of partial evaluation, initiatedby the desire to speed up a constraint-based algorithm for control-flow analysis. We designed and implemented a dedicated partial evaluator,able to specialize the analysis wrt. a given constraint graph and thus remove the interpretive overhead, and measured it with Feeley's Schemebenchmarks. Even though the gain turned out to be rather limited, ourinvestigation yielded valuable feed back in that it provided a better understandingof the analysis, leading us to (re)invent an incremental version.We believe this phenomenon to be a quite frequent spinoff from using partial evaluation, since the removal of interpretive overhead makes the flowof control more explicit and hence pinpoints sources of inefficiency. Finally, we observed that partial evaluation in our case yields such regular,low-level specialized programs that it begs for run-time code generation