Does Pointer Analysis Improve Code Completion?
Past research of Bruch et al. uses a static points-to analysis to track objects within methods. The usage of such identified objects are used to train code-recommendation models. Though intuitive, the benefit of such a strategy has never been formally evaluated. Assuming that multiple instances of the same type are rarely used within one method, the additional complexity of this analysis may not pay off.
The goal of this thesis is to implement a transformation of Simplified Syntax Trees (SSTs), to enrich them with points-to analysis results. This allows us to evaluate whether this additional precision does, in fact, improve recommendation quality or not.
- Bruch, Monperrus, Mezini: Learning from Examples to Improve Code Completion Systems