feat: support Blob v2 UDF signatures (#4091)

Make Function authoring and declaration planning treat Blob v2 as a
scalar semantic type while preserving exact Blob metadata in binding
schemas.

Covers scalar Blob outputs, expanded named-struct outputs, and
whole-result structs with Blob children.

python/python/lancedb/functions.py

@@ -49,6 +49,8 @@ from pydantic import (
     model_validator,
 )
 
+from .schema import is_blob_v2_field as _is_blob_v2_field
+
 _Int32 = conint(strict=True, ge=-(2**31), le=2**31 - 1)
 _UInt32 = conint(strict=True, ge=0, le=2**32 - 1)
 _UInt64 = conint(strict=True, ge=0, le=2**64 - 1)
@@ -518,6 +520,7 @@ class RefreshColumnResult(_RemoteValue):
 
 
 _FUNCTION_NAME = re.compile(r"^[A-Za-z_][A-Za-z0-9_.-]*$")
+_FUNCTION_BLOB_V2_TYPE = "blob_v2"
 
 
 _GRAMMAR_PRIMITIVES = (
@@ -581,20 +584,90 @@ def _validate_exact_arrow_field(field: pa.Field) -> None:
             "unsupported Arrow type for Function signature: field names "
             "must not be empty"
         )
-    if field.metadata:
+    if _is_blob_v2_field(field):
+        if not _has_supported_blob_v2_layout(field):
+            raise TypeError(
+                "unsupported Arrow type for Function signature: lance.blob.v2 "
+                f"requires a supported Blob storage layout, got {field}"
+            )
+    elif field.metadata:
         raise TypeError(
             "unsupported Arrow type for Function signature: field metadata "
             f"is not supported, got {field}"
         )
 
 
+def _has_supported_blob_v2_layout(field: pa.Field) -> bool:
+    data_type = field.type
+    if isinstance(data_type, pa.ExtensionType):
+        data_type = data_type.storage_type
+    if not pa.types.is_struct(data_type):
+        return False
+
+    fields = tuple(data_type)
+
+    def matches(spec, compare_nullable) -> bool:
+        return len(fields) == len(spec) and all(
+            actual.name == name
+            and actual.type == expected_type
+            and (not check_nullable or actual.nullable == nullable)
+            for actual, (name, expected_type, nullable), check_nullable in zip(
+                fields, spec, compare_nullable
+            )
+        )
+
+    logical_minimal = (
+        ("data", pa.large_binary(), True),
+        ("uri", pa.utf8(), True),
+    )
+    logical_full = logical_minimal + (
+        ("position", pa.uint64(), True),
+        ("size", pa.uint64(), True),
+    )
+    prepared = (
+        ("kind", pa.uint8(), True),
+        ("data", pa.large_binary(), True),
+        ("uri", pa.utf8(), True),
+        ("blob_id", pa.uint32(), True),
+        ("blob_size", pa.uint64(), True),
+        ("position", pa.uint64(), True),
+    )
+    descriptor = (
+        ("kind", pa.uint8(), False),
+        ("position", pa.uint64(), False),
+        ("size", pa.uint64(), False),
+        ("blob_id", pa.uint32(), False),
+        ("blob_uri", pa.utf8(), False),
+    )
+    return (
+        matches(logical_minimal, (True, True))
+        or matches(logical_full, (True, True, False, False))
+        or matches(prepared, (True,) * len(prepared))
+        or matches(descriptor, (False,) * len(descriptor))
+    )
+
+
+def _canonical_arrow_field(field: pa.Field) -> str:
+    _validate_exact_arrow_field(field)
+    if _is_blob_v2_field(field):
+        return _FUNCTION_BLOB_V2_TYPE
+    return _canonical_arrow_type(field.type)
+
+
 def _exact_arrow_field(field: pa.Field) -> dict[str, Any]:
     _validate_exact_arrow_field(field)
-    return {
+    if _is_blob_v2_field(field):
+        raise TypeError(
+            "unsupported Arrow type for Function signature: nested Blob v2 "
+            "fields are not supported; declare Blob parameters or named result "
+            "fields directly"
+        )
+    value = {
         "name": field.name,
         "nullable": field.nullable,
         "type": _exact_arrow_type(field.type),
     }
+    return value
 
 
 def _exact_arrow_type(data_type: pa.DataType) -> dict[str, Any]:
@@ -718,7 +791,11 @@ def _function_output(output: pa.DataType | pa.Field | pa.Schema) -> FunctionOutp
         if output.metadata:
             raise TypeError("Function output schema metadata is not supported")
         fields = tuple(output)
-    elif isinstance(output, pa.Field) and pa.types.is_struct(output.type):
+    elif (
+        isinstance(output, pa.Field)
+        and not _is_blob_v2_field(output)
+        and pa.types.is_struct(output.type)
+    ):
         _validate_exact_arrow_field(output)
         if output.nullable:
             raise ValueError("Function output must be non-nullable")
@@ -740,7 +817,7 @@ def _function_output(output: pa.DataType | pa.Field | pa.Schema) -> FunctionOutp
             raise ValueError("Function output must be non-nullable")
         return FunctionOutput(
             kind="scalar",
-            arrow_type=_canonical_arrow_type(field.type),
+            arrow_type=_canonical_arrow_field(field),
             nullable=False,
         )
 
@@ -758,7 +835,7 @@ def _function_output(output: pa.DataType | pa.Field | pa.Schema) -> FunctionOutp
         fields=tuple(
             FunctionResultField(
                 name=field.name,
-                arrow_type=_canonical_arrow_type(field.type),
+                arrow_type=_canonical_arrow_field(field),
                 nullable=False,
             )
             for field in fields
@@ -792,7 +869,7 @@ def _infer_signature(
         inputs = tuple(
             FunctionParameter(
                 name=field.name,
-                arrow_type=_canonical_arrow_type(field.type),
+                arrow_type=_canonical_arrow_field(field),
                 nullable=field.nullable,
             )
             for field in input_schema
@@ -815,7 +892,9 @@ def _infer_signature(
         inputs.append(
             FunctionParameter(
                 name=parameter.name,
-                arrow_type=_canonical_arrow_type(data_type),
+                arrow_type=_canonical_arrow_field(
+                    pa.field(parameter.name, data_type, nullable=nullable)
+                ),
                 nullable=nullable,
             )
         )

python/python/tests/test_first_class_function_slice2.py

@@ -578,6 +578,124 @@ def test_explicit_arrow_schema_is_deterministic():
     assert signature.output.nullable is False
 
 
+def test_blob_fields_use_the_scalar_function_semantic_type():
+    @udf(
+        input_schema=pa.schema([lancedb.blob("image", nullable=False)]),
+        output_schema=lancedb.blob("result", nullable=False),
+    )
+    def copy_blob(image):
+        return image
+
+    signature = copy_blob.registration_request.signature
+    assert signature.inputs[0].arrow_type == "blob_v2"
+    assert signature.output.kind == "scalar"
+    assert signature.output.arrow_type == "blob_v2"
+
+
+def test_named_struct_function_can_include_a_blob_result_field():
+    @udf(
+        input_schema=pa.schema([lancedb.blob("image", nullable=False)]),
+        output_schema=pa.schema(
+            [
+                lancedb.blob("thumbnail", nullable=False),
+                pa.field("width", pa.int32(), nullable=False),
+            ]
+        ),
+    )
+    def inspect_blob(image):
+        return {"thumbnail": image, "width": 1}
+
+    output = inspect_blob.registration_request.signature.output
+    assert output.kind == "named_struct"
+    assert [(field.name, field.arrow_type) for field in output.fields] == [
+        ("thumbnail", "blob_v2"),
+        ("width", "int32"),
+    ]
+
+
+def test_metadata_marked_blob_field_uses_the_semantic_type():
+    extension = lancedb.blob("image", nullable=False).type
+    storage = (
+        extension.storage_type if isinstance(extension, pa.ExtensionType) else extension
+    )
+    metadata_blob = pa.field(
+        "image",
+        storage,
+        nullable=False,
+        metadata={"ARROW:extension:name": "lance.blob.v2"},
+    )
+
+    @udf(
+        input_schema=pa.schema([metadata_blob]),
+        output_schema=pa.field("size", pa.int64(), nullable=False),
+    )
+    def blob_size(image):
+        return len(image)
+
+    assert blob_size.registration_request.signature.inputs[0].arrow_type == "blob_v2"
+
+
+def test_blob_marker_rejects_invalid_storage_layout():
+    malformed = pa.field(
+        "image",
+        pa.int64(),
+        nullable=False,
+        metadata={"ARROW:extension:name": "lance.blob.v2"},
+    )
+
+    with pytest.raises(TypeError, match="requires a supported Blob storage layout"):
+
+        @udf(
+            input_schema=pa.schema([malformed]),
+            output_schema=pa.field("size", pa.int64(), nullable=False),
+        )
+        def blob_size(image):
+            return len(image)
+
+
+def test_nested_blob_signature_field_has_a_clear_error():
+    nested = pa.field(
+        "value",
+        pa.struct([lancedb.blob("image", nullable=False)]),
+        nullable=False,
+    )
+    with pytest.raises(TypeError, match="nested Blob v2 fields are not supported"):
+
+        @udf(
+            input_schema=pa.schema([nested]),
+            output_schema=pa.field("size", pa.int64(), nullable=False),
+        )
+        def blob_size(value):
+            return len(value["image"])
+
+
+def test_nested_non_blob_extension_is_not_silently_unwrapped():
+    class TestExtension(pa.ExtensionType):
+        def __init__(self):
+            super().__init__(pa.int64(), "test.function.extension")
+
+        def __arrow_ext_serialize__(self):
+            return b""
+
+        @classmethod
+        def __arrow_ext_deserialize__(cls, storage_type, serialized):
+            return cls()
+
+    nested = pa.field(
+        "value",
+        pa.struct([pa.field("extended", TestExtension(), nullable=False)]),
+        nullable=False,
+    )
+    with pytest.raises(TypeError, match="unsupported Arrow type"):
+
+        @udf(
+            input_schema=pa.schema([nested]),
+            output_schema=pa.field("result", pa.int64(), nullable=False),
+        )
+        def extension_value(value):
+            return value["extended"]
+
+
 def test_nested_struct_output_uses_canonical_exact_json():
     token = pa.struct(
         [