We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Hacker News·Jun 14·5 min read
Weave: Merging based on language structure and not lines
TL;DR
Traditional line-based merging in version control systems often leads to conflicts when multiple developers edit the same file. Weave is a new entity-level semantic merge driver for Git that uses tree-sitter to merge code based on functions and classes rather than lines.
✦ Why It Matters
Engineers can adopt Weave to improve collaboration and reduce merge conflicts in their projects.
Key Takeaways
- Implement Weave to avoid merge conflicts.
- Utilize tree-sitter for code parsing.
- Encourage developers to claim code entities before editing.
- Explore multi-agent workflows for enhanced collaboration.
Full Summary
Git's default line-based merge strategy creates conflicts when multiple agents edit the same file, even if they modify different functions or classes—a critical limitation for multi-agent AI workflows and collaborative development. Weave is a semantic merge driver for Git that replaces line-level merging with entity-level (function and class) merging by parsing code using tree-sitter, a language-agnostic parser.
The system implements conflict prevention through pre-merge entity claiming, where agents declare ownership of specific functions before editing, enabling detection of conflicts before they occur. Weave achieved zero conflicts across 31 merge scenarios spanning 7 programming languages, demonstrating 100% success rate on tested cases.
Engineers can deploy Weave as a standalone merge driver or integrate it with multi-agent coordination systems, enabling scalable collaborative development without manual conflict resolution.