No actionable comments were generated in the recent review. 🎉

sequenceDiagram
  participant Client as extractSymbolsAndCalls
  participant LuaExtractor as extractFromLua
  participant AstGrep as ast-grep
  participant ScopeTracker as ScopeFrameTracker
  Client->>LuaExtractor: route Lua file for extraction
  LuaExtractor->>AstGrep: parse source -> AST
  AstGrep-->>LuaExtractor: return AST nodes
  LuaExtractor->>ScopeTracker: emit symbols and record scope frames
  LuaExtractor->>LuaExtractor: find function_call nodes and derive callees
  LuaExtractor->>ScopeTracker: attribute calls to innermost scope or <module>
  LuaExtractor-->>Client: return symbols, calls, and scopes

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Thanks for this, it is a clean and well-scoped contribution, and CI is green across Node 18/20/22.

I verified both halves against the engine:

Discovery fix: correct and, importantly, complete. I tested our ignore@7.0.5 with the exact /* then !/src/ pattern: ignores("src") is true (the old walk bailed here, hence the empty graph), ignores("src/") is false (your fix descends), and ignores("src/foo.lua") is false, so the files under the re-included directory are actually picked up, not just the directory entered. I also confirmed the search/embedding discovery path (getIndexableFiles, which uses glob(nodir:true) plus a per-file check) is unaffected, so scoping this to the graph walk is right.

Lua extractor: faithful to the existing extractors. Good calls on taking the direct child before parameters as the name, requiring the assignment RHS to be directly a function_definition, and filtering keywords on call sites.

A few things before merge:

Add real Lua tests. tests/unit/graph-symbols.test.ts has a per-language block with concrete assertions for every other extractor. Please add a describe("Lua") that locks in the new behavior: qualified Table.method / T:m() names, local function, the T.f = function() ... end form, and call attribution to the enclosing function.

Fix the now-misleading existing test. it("handles Lua via regex fallback", ...) (currently in the "Regex fallback" block) no longer exercises the fallback, since Lua now routes to extractFromLua. It still passes because it only asserts exists, but the name and placement are now wrong. Please move it into the new Lua block and give it real assertions.
Add a regression test for the discovery fix. A whitelist .gitignore fixture (/* then !/src/) that asserts files under src are discovered. Note getGraphableFiles is not currently exported, so this likely means either exporting it for the test or asserting through the graph build.

Minor style: every other dynamically-registered grammar calls parse("x" as unknown as Lang, source). Please match that for parse("lua", ...) for consistency. Not blocking since it compiles.

The discovery fix is valuable also on its own. Thanks again.

@coderabbitai review

Thanks for the thorough review — all addressed in 8be929a:

1. Lua tests — new describe("Lua") block locking in qualified Table.method / T:m() names, local function, both the T.f = function() … end and local f = function() … end forms, plus call attribution to the enclosing function (and top-level calls falling back to <module>).
2. Misleading test — moved the former "handles Lua via regex fallback" into the Lua block with real assertions, and dropped "Lua" from the regex-fallback block title.
3. Discovery regression — added a whitelist .gitignore fixture (/* then !/src/) asserting files under the re-included dir are discovered; exported getGraphableFiles for the test.
4. Style — parse("lua" as unknown as Lang, source) to match the other dynamic grammars.

Local suite green for the touched areas; lint clean.