TL;DR
Fusion is a novel bug-reporting system for Android that uses static and dynamic analysis to help users "auto-complete" reproduction steps. By bridging the communication gap between non-technical reporters and developers, Fusion significantly reduces the number of non-reproducible bug reports.
What Problem Does This Paper Address?
Software maintenance is expensive, and resolving bug reports is one of its most critical tasks. However, there is a persistent "lexical knowledge gap" between the people who find bugs (users/testers) and the people who fix them (developers).
Most bug trackers rely on unstructured natural language. Users often provide vague descriptions, omitting the specific "steps to reproduce" that developers desperately need. This leads to a high volume of "non-reproducible" reports, which waste developer time. While some "in-field" tools exist to record crashes, they often require heavy instrumentation, only trigger on actual crashes (not logic bugs), and haven't been optimized for the resource-constrained environment of mobile apps.
Methods
The authors introduce Fusion, which operates in two distinct phases:
- Analysis Phase:
- Static Analysis (The Primer): Fusion uses tools like
apktool,dex2jar, andsrcMLto extract GUI components (buttons, menus, etc.) from the app's source code or APK. It builds a map of which actions are possible on which components and links them to specific code-level artifacts.
- Static Analysis (The Primer): Fusion uses tools like
- Dynamic Analysis (The Engine): The system "rips" the app using a systematic Depth-First Search (DFS) algorithm. Using
UIAutomator, it explores the app to capture screenshots, component locations, and event flows (how one screen leads to another). - Report Generation Phase:
- Auto-completion Engine: When a reporter starts documenting a bug, Fusion tracks their input and suggests the next likely steps based on the pre-built GUI model.
- Contextualized Reporting: Instead of just text, the reporter selects from suggested
{action, gui-component}tuples. Fusion then automatically attaches a screenshot and a traceability link to the source code for every step.
Datasets
The researchers evaluated Fusion using:
- 15 real-world bugs identified in 14 open-source Android applications sourced from the F-Droid marketplace.
- User Studies:
- Study 1: 8 graduate students created 120 bug reports (60 using Fusion, 60 using the standard Google Code Issue Tracker).
- Study 2: 20 participants attempted to reproduce the bugs using the reports generated in Study 1.
Key Insights and Ideas
- GUI as the Common Language: Mobile apps are inherently GUI-centric. By modeling the interface, Fusion creates a shared vocabulary that both a user (who sees a button) and a developer (who sees a code object) can understand.
- Proactive vs. Reactive: Unlike traditional tools that record a failure after it happens, Fusion facilitates the creation of the report, ensuring high-quality data is captured from the start.
- Reducing Cognitive Load: The auto-completion mechanism doesn't just make reporting faster; it guides the reporter to be precise, effectively "forcing" the inclusion of actionable reproduction steps.
Key Results
The study compared Fusion against the Google Code Issue Tracker (GCIT) and found a marked improvement in report quality:
- Higher Reproducibility: Only 10.8% (13/120) of Fusion-generated reports were non-reproducible, compared to 19.1% (23/120) of reports created with the standard tracker.
- Actionable Data: Fusion reports provided developers with immediate "actionable knowledge," including reliable reproduction steps and direct links to the source code responsible for the UI components involved in the bug.
Limitations and Future Work
- Analysis Overhead: While Fusion is "automatic," the initial static and dynamic analysis must be performed once for every version of the app. For extremely large or complex apps, the DFS "ripping" process might struggle to cover every possible state.
- Resource Constraints: The authors acknowledge that mobile environments are resource-constrained. While Fusion moves much of the heavy lifting to a pre-processing phase, further optimization is needed for the analysis techniques to scale.
- Future Directions: The authors suggest that mobile issue trackers can be further augmented by leveraging the unique nature of mobile hardware (sensors, touch gestures) to make reporting even more seamless.
Summary updated