BLADE: Towards Scalable Source Code Debloating

Existing source code debloaters fall short due to low scalability and high runtime overhead when applied in dynamic cloud settings, where instances are spun up on the fly. To address this challenge, we propose BLADE that leverages the common coding idioms and language restrictions to build simple yet effective heuristics for faster source-code debloating. For example, usually, coding constructs are defined before used. Thus, the probability of breaking code after the removal of a node reduces with the depth of its position in the syntax tree. Also, while debloating certain functionalities, statements from a basic block have a higher possibility of getting removed together. To utilize these insights, BLADE employs a hierarchical source code reduction, where reduction candidates are chosen with reverse pre-order traversal, so that it removes uses before the definitions. Low runtime overhead makes BLADE practical to apply code debloating to large workloads. Our evaluation shows that BLADE runs faster than existing source code debloating tools. Compared to Chisel, BLADE is, on average, 2.3 faster and provides comparable reductions in the code size and attack surfaces. In comparison to Debop, another source code debloater, BLADE, on average, is 2.75 faster.