Introduction
Android applications are modular systems composed of various components like Activities and Services that communicate via Inter-Component Communication (ICC), typically through "Intents." Accurately resolving these ICC links is vital for security auditing and automated testing. However, static analysis tools often struggle with the complexity of real-world Android development, such as dynamic fragment loading and atypical API usage. This paper presents a large-scale evaluation of six state-of-the-art ICC resolution tools, revealing that current techniques miss between 38% and 85% of links in complex, commercial applications.
Methodology
To conduct a fair comparison, the authors developed a multi-pronged evaluation framework. Because different tools (Epicc, IC3, IC3Dial, Gator, StoryDroid, and ICCBot) produce disparate output formats, the researchers built a system to normalize these results into a Unified Transition Graph (CTG) format.
The study utilized three distinct datasets:
- BenchHand (73 apps): Synthetic benchmarks designed to test specific code features.
- BenchSmall (31 apps): Real-world apps where a ground-truth "oracle" of 1,680 ICC links was established through dynamic analysis (using the APE tool) and manual auditing.
- BenchLarge (4,000 apps): A massive collection from F-Droid and Google Play to test scalability and efficiency.
Key Results and Insights
The Completeness Gap
A major finding is the "Synthetic Success" trap: tools perform excellently on hand-made benchmarks but fail significantly on real-world software. On the BenchSmall dataset, the False Negative (FN) rate ranged from 38% to 85%. Interestingly, 73% of these missed links were due to pre-existing Java/Android features like complex callbacks rather than newer features like Fragments.
Quantity vs. Quality
As shown in the comparative data, some tools like Gator report an exceptionally high number of ICC links. However, manual auditing reveals that many of these are false positives that do not contribute to a coherent map of the application's actual transitions.
Efficiency and Failure Patterns
Scalability remains a hurdle. For large commercial apps, tools like IC3 timed out or crashed on up to 36% of the Google Play dataset. The researchers identified eight recurring failure patterns, such as layout-related callbacks and polymorphic invocations, which consistently thwart static analysis.
Conclusion
While ICCBot and StoryDroid emerged as relatively stronger performers, no single tool provides a complete solution for real-world Android ICC resolution. Future work must focus on hybrid analysis—combining static and dynamic methods—and more accurate modeling of the Android lifecycle to bridge the current gap in analysis completeness and efficiency.
Summary updated