Weave – A language aware merge algorithm based on entities

Overview & Goals

• Weave is a Git merge driver that merges code at the level of “entities” (functions, classes, methods) instead of lines.
• It uses Tree-sitter to parse files, falls back to line-based merge for unsupported languages, and writes normal files back so Git workflows remain unchanged.
• A key motivation is cleaner merges when multiple changes touch the same file but different entities.

Validation & Adoption Concerns

• Several well-known Git ecosystem contributors have privately expressed support; some users see this as strong validation.
• Others question the evidence for demand and worry about promotional tactics (e.g., opening issues on many repos to suggest adoption).
• There is interest in integrating with other tools (e.g., alternative VCSs, Git frontends), but this is mostly early-stage discussion.

Multi-Agent and AI Workflows

• Proponents argue entity-level merging is critical when many AI agents edit in parallel, reducing unnecessary conflicts and saving time/tokens.
• Skeptics say AI can already resolve most textual conflicts and that humans rarely find these conflicts hard anyway.
• A related idea is moving from “fix conflicts later” to “prevent conflicts” via an MCP server where agents claim entities before editing.

Technical Approach & Alternatives

• Compared to other Tree-sitter-based tools that match raw AST nodes, Weave groups entire entities, aiming for simpler, faster, more readable conflicts.
• There’s extensive discussion of storing ASTs (or CSTs) directly in a VCS vs. treating source as blobs; some argue code is already effectively a serialized AST and the real issue is parsers/consumers, not storage.
• Concerns are raised about languages with preprocessors (C/C++) and significant whitespace (Python); Weave claims best-effort handling with Tree-sitter plus language-specific logic.
• Structural hashes are used for rename detection; a commenter doubts this will reliably handle more complex refactors.

Language Support & Tooling

• Current support includes several languages via a shared parser library; Ruby and other languages (Swift, Bash) are requested and described as relatively easy to add through Tree-sitter grammars.
• C# support exists but is not prominently documented.
• A separate tool provides entity-level diffs on arbitrary files, and another experimental tool uses entity graphs for PR triage.

Limitations, Risks & Open Questions

• Weave can still miss purely semantic conflicts (e.g., one change obsoleting another); it reportedly runs post-merge dependency checks to flag some of these.
• Some argue stricter, more conservative conflict detection might be safer in cases where merging both sides leads to dead code.
• Performance trade-offs of AST-based approaches, robustness on heavily macro’d code, and behavior during rebases (vs merges) are raised but not fully resolved.