Automated duplicated code detection and procedure extraction
Raghavan Komondoor, Susan B. Horwitz · 2003
Making changes to software is a difficult task. Up to 70% of the effort in the software process goes towards maintenance. This is mainly because programs have poor structure (due to poor initial design, or due to repeated ad hoc modifications) which makes them difficult to understand and modify. The focus of this thesis is duplication in source code, which is a major cause of poor structure in real programs. We make two contributions: (a) a novel program-slicing-based approach for detecting duplicated fragments in source code, and (b) a pair of algorithms, one that works on a single selected fragment of code, and the other that works on a group of matching fragments, for making the fragment(s) easily extractable into a separate procedure. The key, novel aspect of our duplication-detection approach is its ability to detect “difficult” groups of matching fragments, i.e., groups in which matching statements are not in the same order in all fragments, and groups in which non-matching statements intervene between matching statements. Our procedure-extraction algorithms are an advance over previous work in this area in two ways: they employ a range of transformations, including code motion and duplication of predicates, to handle a wide variety of difficult clone groups that arise in practice; and they are the first, to our knowledge, to address the extraction of fragments that contain exiting jumps (jumps from within the region containing the fragment to locations outside that are not the “fall through” exit of the region). We present experimental results, using implementations of our duplication-detection algorithm and one of our extractability algorithms, that indicate that our approaches are effective and useful in practice.