git

This study investigates the contextual applicability of Trunk-Based Development (TBD) versus Branch-Based Development (BBD) workflows to optimize developer productivity and software quality across diverse team settings. Method: We conducted a mixed-methods empirical study with 127 practitioners in Brazil, comprising semi-structured interviews (n=15) and an online survey (n=112), analyzed via qualitative coding and quantitative statistical methods. Contribution/Results: To our knowledge, this is the first systematic, industry-based comparison of TBD and BBD practices. We identify team size and developer experience as critical moderating factors: TBD significantly improves delivery efficiency in small, experienced teams, whereas BBD better accommodates large-scale or junior-heavy teams—at the cost of higher process management overhead. The findings yield a data-driven decision framework to guide workflow selection based on organizational context.

This study presents the first large-scale empirical analysis of Git history rewriting and its threats to software supply chain integrity and reproducibility. Addressing risks—including push conflicts, broken provenance, and malicious code injection—arising from history-altering operations (e.g., rebase, filter-branch) in public repositories, the authors analyze 111 million open-source projects archived by Software Heritage. Leveraging static analysis and two in-depth case studies, they propose the first evidence-driven taxonomy of Git history rewriting and develop GitHistorian, an automated detection tool. Applied at scale, the methodology identifies 1.22 million projects exhibiting history rewriting (8.7 million operations total), revealing prevalent legitimate use cases such as license updates and sensitive information removal. The work establishes a novel, scalable methodology for supply chain security assessment and delivers an open, extensible infrastructure for detecting and characterizing historical tampering.

This study addresses the challenge novice learners face in understanding Git’s distributed collaboration model, as existing tools predominantly emphasize individual operations and lack intuitive representations of collaborative states. To bridge this gap, we propose GitAcademy, a novel platform featuring an innovative “split-screen awareness” design that serves as a temporary scaffold within a fully functional, browser-embedded Git environment. This design enables a dual-view collaboration mode, allowing learners to perform local repository operations while simultaneously observing their partner’s actions in real time. By integrating a synchronized remote repository with a mirrored representation of collaborative states, the system significantly enhances social presence and peer learning effects. In an experiment with 13 learner pairs, participants consistently preferred this mode, confirming its effectiveness in improving collaborative understanding, although gains in task performance were modest.

Existing tools struggle to support fine-grained analysis of software code evolution effectively. To address this limitation, this work proposes GitEvo—a multilingual, extensible analysis framework that uniquely integrates Git version metadata with syntactic code structures, such as abstract syntax trees (ASTs), enabling deep co-modeling of version history and code structure for the first time. GitEvo facilitates cross-language tracking of code evolution, computation of evolutionary metrics, and interactive visualization. Its effectiveness has been validated on real-world repositories, demonstrating its utility both as a foundation for empirical software engineering research and as an educational platform for understanding the patterns and principles of software evolution.

This work addresses the challenge that developers often make errors when performing complex Git operations, and existing large language models (LLMs) struggle to ensure correctness and safety due to their limited formal reasoning capabilities. To overcome this limitation, the paper proposes a novel approach that integrates automated planning with LLMs, enabling the system to interpret natural language instructions and formally model the state of a Git repository to generate safe, verifiable command sequences. By combining symbolic reasoning with language understanding, the method significantly improves the success rate of Git operations compared to pure LLM-based solutions. Experimental results demonstrate consistent performance gains across multiple evaluation metrics, thereby enhancing both the reliability and interpretability of AI-powered developer assistance tools.

This work addresses the absence of Git-like data versioning in existing lakehouse architectures, which hinders collaborative development among multiple agents and human-in-the-loop review workflows. To bridge this gap, the paper introduces Git semantics into lakehouse systems by extending Apache Iceberg to support lakehouse-wide commits, branches, and merges—elevating single-table snapshots to a unified versioning model. This enables agents to operate on isolated branches while ensuring atomic, cross-table changes during release. The core abstractions are formally modeled and verified using Alloy to guarantee semantic correctness. The proposed system has been deployed in production, demonstrating the feasibility and effectiveness of the collaborative workflow it enables.

This work addresses the frequent loss of reasoning behind code changes in agent development—such as trade-offs, constraints, and discarded alternatives—due to inadequate memory mechanisms. To remedy this, the paper proposes a novel paradigm that embeds reasoning traces directly into Git as a memory substrate, leveraging native version control operations like commits, merges, and code reviews to ensure authenticity, timeliness, and verifiability. Through a Git-anchored architecture featuring multi-path routing retrieval, structured mapping, confidence-gated snippet injection, and decision synthesis, the approach enables reproduction of results without additional annotations and achieves, for the first time, cross-session reconstruction of “why” reasoning chains. Evaluated on approximately 50,000 lines of production code, it attains an answer adequacy of 0.83 with outputs compressed by three orders of magnitude (382–980 tokens) and a seed retrieval MRR of 0.31, significantly outperforming baseline methods.