detecting merging and splitting using origin analysis.pptVIP

Detecting merging and splitting using origin analysis Lijie Zou and Michael Godfrey Software Architecture Group (SWAG) University of Waterloo Problem Developers use merging/splitting to reduce complexity, improve cohesion, … Easy to see the effects of the changes, but the intent/rationale is often lost “rcsdiff –r1.2 –r2.0 foo.c” vs. “we moved the error handling from foo.c to bar.c” Goal: Want to recover the changes and capture the intent behind these changes to better understand system evolutionary history. Two phases of our approach Use improved “origin analysis” to detect software entities involved in merging/splitting Derive patterns using detailed analysis of change of call relations and other attributes - understand intent Origin Analysis Definition: F was an apparently new entity in Vnew. “Origin analysis” is the process to decide whether F was newly introduced in Vnew, or it should be viewed as a renamed, moved, or otherwise changed version of an entity from Vold, say G. Origin Analysis – “How to” Basic techniques: match software entities from multiple attributes. Name Declaration Metrics Relation (e.g., call relation) Detecting merges/splits at the function level Name-Matcher Declaration-Matcher Metrics-Matcher Relation-Matcher … Detecting chained merges/splits at the function level Detecting merges/splits at the file level Manually File merge: A new file G is found to be composed of most functions from two old files F1 and F2 File split Scatter Plot Two phases of our approach Use improved “origin analysis” to detect software entities involved in merging/splitting Derive patterns using detailed analysis of change of call relations and other attributes - understand intent Patterns Clone elimination Pipeline extraction Service consolidation Parameterization Partial clone elimination Pattern 1: clone elimination Pattern 2: pipeline contraction Pattern 3~5 Service consolidation Two functions that perform different se