android

Android app builds frequently fail due to complex dependency graphs, heterogeneous configuration schemes, and rapid ecosystem evolution. To address this, we conduct an empirical study analyzing build logs from 200 open-source Android projects and propose a five-stage diagnosis-and-repair framework. We systematically categorize build failures into four canonical patterns—the first such taxonomy in the literature. Innovatively, we integrate a large language model (GPT-5) to generate context-aware repair suggestions, which are then validated through automated build execution and manual verification. Our evaluation shows successful repair of 102 out of 135 initially failing projects (75.6% success rate). Furthermore, we identify programming language choice, project age, and application scale as statistically significant factors influencing build reliability. This work contributes both a reusable methodology and empirically grounded insights to enhance Android build robustness and maintainability.

This study addresses semantic discrepancies between parallel Java and Kotlin implementations in the Android framework, which can introduce security vulnerabilities. It presents ParaDroid, the first systematic analysis framework for detecting cross-language semantic differences at scale. ParaDroid integrates a bytecode-level intermediate representation, class-to-source mapping reconstruction, large language model–driven semantic reasoning, and comparative static and dynamic behavior analysis. Applied to 329 pairs of parallel methods in AOSP Android versions 14 through 16, the approach identified 37 instances of fragile semantic divergence, leading to the confirmation of three security vulnerabilities—two of which have been assigned CVE identifiers—and two functional defects.

This study investigates dataset usage patterns in empirical mobile application requirements engineering (RE) research to identify data source bias and its implications for external validity. Following Kitchenham et al.’s systematic mapping methodology, we analyze 43 empirical studies published between 2012 and 2023. Results reveal that over 90% rely exclusively on Google Play and Apple App Store—neglecting critical RE activities such as requirement validation and evolution—while exhibiting pronounced dataset homogeneity despite growing adoption. This work provides the first quantitative evidence of data source narrowing in mobile RE research. To mitigate this risk, we propose a “multi-source data fusion” framework advocating integration across platforms (e.g., F-Droid, GitHub), modalities (e.g., user reviews, source code, changelogs), and RE activities. The framework advances methodological rigor and practical relevance, supporting more generalizable and empirically grounded mobile RE research.

This study addresses the high development costs and significant code redundancy associated with traditional institutional mobile applications that rely heavily on native Android development. To overcome these limitations, the authors propose a full-stack solution leveraging a Django backend and an HTMX frontend, integrated via a WebView bridge to deliver a campus management system without writing any Android SDK code. The system supports core functionalities including task scheduling, inventory management, and attendance tracking, and is deployed using a self-hosted Docker Compose setup, eliminating dependence on external cloud services. Evaluated in a real-world institutional setting, this approach demonstrates for the first time that HTMX combined with Django can effectively replace conventional APK-based development, achieving a 54% reduction in development time, a 91% decrease in HTTP payload size, and a user satisfaction score of 4.2 out of 5.0 among 42 participants.

This work investigates the capability of large language models (LLMs) to generate complete, functional Android applications from scratch, with particular emphasis on system-level challenges—including state coordination, activity lifecycle management, and asynchronous operations. To enable rigorous evaluation, we introduce AppForge, the first automated, end-to-end benchmark for Android application generation, comprising 101 development tasks derived from real-world applications. We propose a novel multi-agent framework that autonomously generates functional specifications and executable test cases, all validated by domain experts to ensure fidelity and reproducibility. Extensive evaluation across 12 state-of-the-art LLMs reveals severe limitations: even the strongest model, GPT-5, achieves only 18.8% task success rate. These results expose fundamental shortcomings of current LLMs in handling complex, structured software engineering tasks requiring deep platform-specific knowledge and precise compositional reasoning.

This study addresses the scarcity of large-scale, reproducible, fine-grained data on third-party SDK dependencies in mobile applications, which hinders research into technical ecosystems and privacy infrastructures. The authors construct a public dataset comprising 334,719 app-version observations by combining static APK analysis, code-signing matching, and an automated processing pipeline, leveraging AndroZoo and Exodus Privacy rules to achieve code-level SDK identification. Covering nearly 100,000 distinct applications and 246 SDKs, the dataset enables the construction of an app–SDK bipartite network and maps SDKs to their operating companies, thereby revealing upstream technological control structures. This resource provides a reusable infrastructure for empirical studies on third-party dependencies and privacy practices in the Android ecosystem.

This work addresses the challenge of repairing visual defects in Android applications, which is often hindered by incomplete human-submitted bug reports lacking observed behavior (OB), expected behavior (EB), or steps to reproduce (S2Rs). To overcome this limitation, the authors propose the first approach that leverages GUI context to guide large language models (LLMs) in generating structured bug reports. By integrating interaction logs and screenshots, the method automatically produces comprehensive and accurate OB, EB, and S2Rs. A unified evaluation framework assessing both correctness and completeness is also introduced. Experimental results on 48 bug reports across 26 applications demonstrate that the generated reports significantly outperform both original human-written reports and existing LLM-based baselines in quality.

This work addresses key challenges in deploying small language models (SLMs) on mobile devices—including erroneous outputs, constraint violations, contextual degradation, latency incompatibility, and unstable model selection—by integrating Gemma-2.6B and Qwen3-0.6B into the Android word game Palabrita. The authors simplify the LLM’s role from generating full puzzles to producing only three hints and introduce a deterministic fallback mechanism. Embracing a “less is more” principle, they derive eight design heuristics for SLM integration on mobile platforms and systematically categorize five failure modes along with corresponding mitigation strategies. Through techniques such as multi-layer defensive parsing, context-aware retries, conversational turn rotation, and responsibility minimization, the approach enables stable, production-grade SLM deployment, substantially improving system reliability and user experience.

Dynamic analysis of Android applications at the application layer has long been constrained by reliance on physical devices, suffering from poor scalability and limited reproducibility. This work proposes a systematic rehosting approach that migrates Android framework components and preinstalled vendor binaries from real-world firmware into a fully emulated environment. By employing tailored extraction and injection strategies, these components are seamlessly integrated into the AOSP build system to produce bootable emulator images that preserve system integrity and runtime compatibility. The method enables, for the first time, large-scale rehosting of vendor-customized Android firmware in QEMU across multiple SDK versions (31–33). Evaluation on 184 firmware samples demonstrates high success rates in both image construction and booting, with only a few failures attributable to missing dependencies or emulator limitations, thereby validating the feasibility and effectiveness of this approach for scalable and reproducible dynamic analysis.

Existing large language model (LLM) evaluation benchmarks lack coverage of mobile application development scenarios, failing to capture platform-specific constraints, framework lifecycles, and complex API interactions inherent to such environments. To address this gap, this work introduces the first multilingual repair benchmark targeting three major mobile platforms—Android Native, React Native, and Flutter—comprising 384 real-world production issues. Each task is accompanied by an executable test patch enabling automated validation of cross-file and cross-artifact modifications. Experimental results reveal that state-of-the-art code LLMs achieve end-to-end repair success rates of only 3.39%–5.21% on this benchmark, substantially lower than their performance on existing datasets, thereby exposing critical limitations in multi-file fault localization and coordinated repair capabilities.