Boundary-Driven Path Diversity Heuristics for Improving Coverage of Difficult-to-Reach Program Paths in Search-Based Software Testing

Difficult-to-reach program paths remain a fundamental challenge in search-based software testing (SBST), often resulting in limited structural coverage and substantial computational overhead. This work presents a methodological integration that combines abstract path modeling with stagnation-aware refinement to improve exploration of difficult-to-reach paths in search-based software testing (SBST). Rather than proposing a new SBST paradigm, the contribution lies in unifying (i) a reduced, count-based abstraction of control-flow behavior and (ii) an adaptive refinement mechanism triggered by stagnation, enabling targeted redirection toward underexplored execution regions. To address search stagnation, boundary constraints are adaptively introduced when progress plateaus, enabling targeted redirection toward underexplored execution regions while avoiding unnecessary computational cost. The proposed technique was empirically evaluated against baseline control strategies, dependency-driven methods, and conventional diversity-based approaches using multiple benchmark programs. Experimental results indicate that the proposed approach achieves substantially higher branch coverage, while reducing the average number of test executions by 73.4% (2,083 vs. 7,833) and decreasing execution time by 57.2%. While the empirical results indicate consistent efficiency gains across the evaluated benchmarks, the study is limited to established research programs rather than large-scale industrial systems. Consequently, claims regarding scalability and industrial applicability should be interpreted as indicative rather than conclusive. Nevertheless, the observed reductions in execution count and runtime suggest that the proposed heuristic has characteristics that are favorable for practical deployment, motivating further evaluation on larger and more heterogeneous software systems.