fix: support unconstrained array item schemas

CHANGELOG.md

@@ -7,6 +7,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+- fix: support unconstrained array item schemas by @tuanzirwar (#2378)
+
 ## [0.3.36]
 
 - feat: update llama.cpp to ggml-org/llama.cpp@0c1e57098

llama_cpp/llama_grammar.py

@@ -771,7 +771,9 @@ def add_component(comp_schema, is_required):
         elif schema_type in (None, "array") and (
             "items" in schema or "prefixItems" in schema
         ):
-            items = schema.get("items") or schema["prefixItems"]
+            items = schema.get("items") or schema.get("prefixItems")
+            if items is None:
+                items = schema["items"]
             if isinstance(items, list):
                 return self._add_rule(
                     rule_name,
@@ -830,7 +832,10 @@ def add_component(comp_schema, is_required):
                 + r' "\"" space',
             )
 
-        elif (schema_type == "object") or (len(schema) == 0):
+        elif len(schema) == 0:
+            return self._add_primitive(rule_name, PRIMITIVE_RULES["value"])
+
+        elif schema_type == "object":
             return self._add_rule(
                 rule_name, self._add_primitive("object", PRIMITIVE_RULES["object"])
             )

tests/test_llama_grammar.py

@@ -76,3 +76,56 @@ def test_grammar_anyof():
     grammar = llama_cpp.LlamaGrammar.from_json_schema(json.dumps(sch))
 
     # assert grammar.grammar is not None
+
+
+def test_grammar_unconstrained_array_items():
+    schema = {"type": "array", "items": {}}
+    grammar = llama_cpp.LlamaGrammar.from_json_schema(json.dumps(schema))
+    rules = dict(line.split(" ::= ", 1) for line in grammar._grammar.splitlines())
+    assert rules["item"] == "object | array | string | number | boolean | null"
+    assert "item" in rules["root"]
+
+
+def test_grammar_unconstrained_tuple_item():
+    schema = {"type": "array", "prefixItems": [{}, {"type": "integer"}]}
+    grammar = llama_cpp.LlamaGrammar.from_json_schema(json.dumps(schema))
+    rules = dict(line.split(" ::= ", 1) for line in grammar._grammar.splitlines())
+    assert rules["tuple-0"] == "object | array | string | number | boolean | null"
+    assert "integer" in rules["root"]
+
+
+def test_grammar_empty_schema_allows_any_json_value():
+    grammar = llama_cpp.LlamaGrammar.from_json_schema("{}")
+    rules = dict(line.split(" ::= ", 1) for line in grammar._grammar.splitlines())
+    assert rules["root"] == "object | array | string | number | boolean | null"
+
+
+def test_grammar_typed_array_preserves_item_constraints():
+    grammar = llama_cpp.LlamaGrammar.from_json_schema(
+        json.dumps(
+            {
+                "type": "array",
+                "items": {"type": "integer"},
+                "minItems": 1,
+                "maxItems": 2,
+            }
+        )
+    )
+    assert (
+        "integer"
+        in dict(line.split(" ::= ", 1) for line in grammar._grammar.splitlines())[
+            "root"
+        ]
+    )
+
+
+def test_grammar_closed_tuple_preserves_prefix_items():
+    schema = {
+        "type": "array",
+        "prefixItems": [{"type": "integer"}, {"type": "string"}],
+        "items": False,
+    }
+    grammar = llama_cpp.LlamaGrammar.from_json_schema(json.dumps(schema))
+    rules = dict(line.split(" ::= ", 1) for line in grammar._grammar.splitlines())
+    assert "integer" in rules["root"]
+    assert "string" in rules["root"]