Towards Better Graph Neural Network-based Fault Localization Through Enhanced Code Representation

TL;DR

This work addresses the scalability and accuracy challenges of GNN-based fault localization by introducing DepGraph, a Dependency-Enhanced Coverage Graph that fuses AST structure, interprocedural call graphs, and dynamic code coverage with code-change attributes. A five-iteration GGNN ranks faulty methods using a listwise loss, yielding significant improvements over Grace in Top-1, MFR, and MAR, while dramatically reducing graph size, GPU memory, and training/inference time. Across Defects4J v2.0.0, DepGraph demonstrates strong cross-project generalization and identifies fault sets that GNN often misses, especially those tied to method interactions and loops. The approach offers practical benefits for scalable fault localization and provides a foundation for future graph-representation enhancements and history-aware debugging tools.

Abstract

Automatic software fault localization plays an important role in software quality assurance by pinpointing faulty locations for easier debugging. Coverage-based fault localization, a widely used technique, employs statistics on coverage spectra to rank code based on suspiciousness scores. However, the rigidity of statistical approaches calls for learning-based techniques. Amongst all, Grace, a graph-neural network (GNN) based technique has achieved state-of-the-art due to its capacity to preserve coverage spectra, i.e., test-to-source coverage relationships, as precise abstract syntax-enhanced graph representation, mitigating the limitation of other learning-based technique which compresses the feature representation. However, such representation struggles with scalability due to the increasing complexity of software and associated coverage spectra and AST graphs. In this work, we proposed a new graph representation, DepGraph, that reduces the complexity of the graph representation by 70% in nodes and edges by integrating interprocedural call graph in the graph representation of the code. Moreover, we integrate additional features such as code change information in the graph as attributes so the model can leverage rich historical project data. We evaluate DepGraph using Defects4j 2.0.0, and it outperforms Grace by locating 20% more faults in Top-1 and improving the Mean First Rank (MFR) and the Mean Average Rank (MAR) by over 50% while decreasing GPU memory usage by 44% and training/inference time by 85%. Additionally, in cross-project settings, DepGraph surpasses the state-of-the-art baseline with a 42% higher Top-1 accuracy, and 68% and 65% improvement in MFR and MAR, respectively. Our study demonstrates DepGraph's robustness, achieving state-of-the-art accuracy and scalability for future extension and adoption.

Figures (5)

• Figure 1: Comparing graph representation for coverage and call graph for Lang-62. The green nodes correspond to the code statements (1--4). (a) shows a coverage graph linking tests to source code statements without considering the method-level call graph. (b) considers the method-level call graph, eliminating node 4 since it was not reachable in the call graph.
• Figure 2: Overview of DepGraph. The term $\text{type}$ denotes the AST node's type (e.g., M1 is a MethodDeclaration, and S2 is an IfStatement). Tests, such as T1, include test outcomes (e.g., pass or fail).
• Figure 3: An example of the Dependency-Enhanced Coverage Graph representation in DepGraph.
• Figure 4: Code Representation for the method unescape() based on the abstract syntax tree.
• Figure 5: Overlaps between the faults that DepGraph and GNN (i.e., Grace) locate in Top-1, 3, 5 and 10. The overlapping regions contain a number of faults that have the same ranking in both of the techniques. The non-overlapped regions contain faults that were uniquely located by each technique. We consider both DepGraph, and, DepGraph w/o CC (Code Change).