feat: AST-aware chunking for Java (.java)

[Ciel2142]

Summary

Adds Java (.java) to QMD's AST-aware chunking. With --chunk-strategy auto,
Java sources are split at class / interface / record / enum / method / field /
import boundaries via tree-sitter instead of arbitrary regex cuts, improving
chunk coherence and retrieval quality for Java code. Implements the Java
portion of #565.

No behavior change unless --chunk-strategy auto is requested — existing
languages and the default regex strategy are untouched, and src/store.ts
needs no change.

How it works

src/ast.ts is data-driven: a language is registered purely through lookup
tables, no new code paths.

• SupportedLanguage += "java"; EXTENSION_MAP += .java
• GRAMMAR_MAP += tree-sitter-java — official package, ships a prebuilt
  ~414 KB wasm, ABI-compatible with this repo's web-tree-sitter@0.26.8
• LANGUAGE_QUERIES.java — tree-sitter query mapping Java nodes onto the
  existing cross-language capture vocabulary
• SCORE_MAP += one new key, field: 60

getASTStatus() / qmd status pick up Java automatically.

Query & scoring

Capture	Java nodes	Score
@class	class, record	100
@iface	interface	100
@enum	enum	80
@type	annotation type	80
@method	method, constructor, compact constructor	90
@field	field, interface constant, enum constant	60 (new)
@import	package, import	60

Reuses the established score vocabulary so Java stays consistent with the other
six languages; field is the only new key.

Annotations

tree-sitter-java includes a declaration's leading annotations in its
startIndex (the modifiers node is a child), so an @Service-annotated class
or @Override method breaks at the annotation — a chunk boundary never
splits an annotation from the declaration it decorates. Pinned by a test.

Scope — Kotlin

Java only, to keep this PR focused. Kotlin (.kt / .kts) AST chunking is
coming in a separate follow-up PR.

Testing

• test/ast.test.ts: detection (incl. uppercase extension + qmd:// paths),
  break points across package/import/class/interface/enum/record/method/field/
  annotation-type, score hierarchy, position ordering, annotation-boundary
  positioning, and grammar availability via getASTStatus().
• test/ast-chunking.test.ts: integration check that the new field score
  flows through the chunking pipeline.
• Green on both runners (vitest and bun test) and tsc.
• Also exercised end-to-end on a real ~360-file Spring Boot codebase: indexed +
  embedded, Java chunks land on method/class/Javadoc boundaries and are
  retrievable via keyword and semantic search.

Packaging

The Java wasm (~414 KB) is comparable to the already-shipped TypeScript grammar
— no meaningful change to install size.

Commits

• build: add tree-sitter-java grammar dependency
• feat: register Java in the AST layer
• docs: CLAUDE.md / README.md / CHANGELOG.md
• test: integration coverage for the Java field score

Part of #565 (Java; Kotlin tracked separately — not auto-closing).

🤖 Generated with Claude Code

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	pypi/​torch@​2.11.0
	pypi/​transformers@​4.57.6
	npm/​tree-sitter-java@​0.23.5
	pypi/​sentencepiece@​0.2.1
	pypi/​huggingface-hub@​0.36.2
	pypi/​gguf@​0.18.0
	pypi/​trl@​1.0.0
	pypi/​peft@​0.18.1
	pypi/​trackio@​0.4.1
	pypi/​pydantic@​2.12.3
	pypi/​pyyaml@​6.0.3
	pypi/​nvidia-ml-py@​13.595.45

View full report