cocoindex-io · georgeh0 · Sep 5, 2026 · Aug 29, 2026
diff --git a/Cargo.lock b/Cargo.lock
diff --git a/docs/src/content/docs/ops/text.mdx b/docs/src/content/docs/ops/text.mdx
@@ -155,6 +155,7 @@ Languages with syntax-aware (tree-sitter) splitting — splits at logical bounda
 | JSON | `"json"` | `.json`, `.jsonl`, `.geojson` |
 | Julia | `"julia"` | `.jl` |
 | Kotlin | `"kotlin"` | `.kt`, `.kts`, `.ktm` |
+| Lua | `"lua"` | `.lua`, `.nse`, `.p8`, `.pd_lua`, `.rbxs`, `.rockspec`, `.wlua` |
 | Markdown | `"markdown"` | `.md`, `.markdown`, `.mdx` |
 | Pascal / Delphi | `"pascal"` | `.pas`, `.dpr`, `.lpr`, `.dfm` |
 | PHP | `"php"` | `.php` |
@@ -176,7 +177,7 @@ Languages with syntax-aware (tree-sitter) splitting — splits at logical bounda
 
 Note that `.h` maps to C++, not C — the C++ grammar also parses C headers. Pass `language="c"` explicitly to force the C grammar.
 
-Over 110 additional languages use separator-based splitting (e.g. `"elixir"`, `"erlang"`, `"haskell"`, `"lua"`, `"nix"`, `"perl"`, `"powershell"`, `"proto"`, `"zig"`, and more). A `language=` value that matches no known language also falls back to separator-based splitting rather than raising.
+110 additional languages use separator-based splitting (e.g. `"elixir"`, `"erlang"`, `"haskell"`, `"nix"`, `"perl"`, `"powershell"`, `"proto"`, `"zig"`, and more). A `language=` value that matches no known language also falls back to separator-based splitting rather than raising.
 
 ### `CustomLanguageConfig`
 

diff --git a/examples/rust/amazon_s3_embedding/Cargo.lock b/examples/rust/amazon_s3_embedding/Cargo.lock
diff --git a/examples/rust/code_embedding/Cargo.lock b/examples/rust/code_embedding/Cargo.lock
diff --git a/examples/rust/code_embedding_lancedb/Cargo.lock b/examples/rust/code_embedding_lancedb/Cargo.lock
diff --git a/examples/rust/gdrive_text_embedding/Cargo.lock b/examples/rust/gdrive_text_embedding/Cargo.lock
diff --git a/examples/rust/oci_object_storage_embedding/Cargo.lock b/examples/rust/oci_object_storage_embedding/Cargo.lock
diff --git a/examples/rust/paper_metadata/Cargo.lock b/examples/rust/paper_metadata/Cargo.lock
diff --git a/examples/rust/pdf_embedding/Cargo.lock b/examples/rust/pdf_embedding/Cargo.lock
diff --git a/examples/rust/text_embedding/Cargo.lock b/examples/rust/text_embedding/Cargo.lock
diff --git a/examples/rust/text_embedding_lancedb/Cargo.lock b/examples/rust/text_embedding_lancedb/Cargo.lock
diff --git a/examples/rust/text_embedding_qdrant/Cargo.lock b/examples/rust/text_embedding_qdrant/Cargo.lock
diff --git a/examples/rust/text_embedding_turbopuffer/Cargo.lock b/examples/rust/text_embedding_turbopuffer/Cargo.lock
diff --git a/python/tests/ops/test_text.py b/python/tests/ops/test_text.py
@@ -19,6 +19,7 @@ def test_detect_code_language_known_extensions() -> None:
     assert detect_code_language(filename="App.vue") == "vue"
     assert detect_code_language(filename="script.jl") == "julia"
     assert detect_code_language(filename="main.dart") == "dart"
+    assert detect_code_language(filename="init.lua") == "lua"
     assert detect_code_language(filename="Main.elm") == "elm"
     assert detect_code_language(filename="index.astro") == "astro"
     assert detect_code_language(filename="deploy.sh") == "bash"
@@ -229,6 +230,54 @@ def test_recursive_splitter_with_dart() -> None:
     assert all(isinstance(c, Chunk) for c in chunks)
 
 
+def test_recursive_splitter_with_lua() -> None:
+    """Lua splitting is syntax-aware, not separator-based.
+
+    The sample is one line, so separators give the fallback path no function
+    boundaries at all. Only the grammar can find them.
+    """
+    splitter = RecursiveSplitter()
+    code = (
+        "local function f(a) return a+1 end "
+        "local function g(b) return b*2 end "
+        "local function h(c) return c-3 end"
+    )
+    chunks = splitter.split(code, chunk_size=45, min_chunk_size=10, language="lua")
+
+    assert len(chunks) > 1, "sample must split for the check below to mean anything"
+    assert all(isinstance(c, Chunk) for c in chunks)
+    for chunk in chunks:
+        text = chunk.text.strip()
+        assert text.startswith("local function") and text.endswith("end"), (
+            f"chunk straddles a function boundary: {text!r}"
+        )
+
+
+def test_recursive_splitter_lua_keeps_table_constructor_whole() -> None:
+    """A Lua table constructor is not severed at a comma between its fields.
+
+    Separator fallback cuts `{ x = x, y = y }` at the comma, because a comma is
+    a separator to it and a table field to the grammar. A constructor longer
+    than `chunk_size` is still split, here as in every other grammar-backed
+    language.
+    """
+    splitter = RecursiveSplitter()
+    code = (
+        "local function clamp(v, lo, hi) return math.min(math.max(v, lo), hi) end\n"
+        "local Point = {}\n"
+        "Point.__index = Point\n"
+        "function Point.new(x, y) return setmetatable({ x = x, y = y }, Point) end\n"
+        "function Point:norm() return math.sqrt(self.x * self.x + self.y * self.y) end\n"
+        "return Point\n"
+    )
+    chunks = splitter.split(code, chunk_size=60, min_chunk_size=10, language="lua")
+
+    assert len(chunks) > 1, "sample must split for the check below to mean anything"
+    assert any("setmetatable({ x = x, y = y }, Point)" in c.text for c in chunks), (
+        "table constructor was severed across chunks"
+    )
+
+
 def test_recursive_splitter_with_vue() -> None:
     """Test RecursiveSplitter with Vue syntax-aware splitting."""
     splitter = RecursiveSplitter()

diff --git a/rust/code_ast/Cargo.toml b/rust/code_ast/Cargo.toml
@@ -30,6 +30,7 @@ tree-sitter-json = "0.24.8"
 tree-sitter-julia = "0.23.1"
 # The other more popular crate tree-sitter-kotlin requires tree-sitter < 0.23 for now
 tree-sitter-kotlin-ng = "1.1.0"
+tree-sitter-lua = "0.5.0"
 tree-sitter-md = "0.5.3"
 tree-sitter-pascal = "0.10.2"
 tree-sitter-php = "0.23.11"