A framework for preprocessor-aware C source code analyses

Analyses of C source code usually ignore the C preprocessor because of its complexity. Instead, these analyses either define their own approximate parser (or scanner) or else they require that their input already be preprocessed. Neither approach is entirely satisfactory: the first gives up accuracy (or incurs large implementation costs), while the second loses the preprocessor-based abstractions. We describe a framework that permits analyses to be expressed in terms of both preprocessing and parsing actions, allowing the implementer to focus on the analysis. We discuss an implementation of such a framework that embeds a C preprocessor, a parser, and a Perl interpreter for the action “hooks.” Many common software engineering analyses can be written surprisingly easily using our implementation, replacing numerous ad-hoc tools. The framework’s integration of the preprocessor and the parser further enables some analyses that otherwise would be especially difficult.