feat(singlestoredb): add vector type support

Author: kesmit13

ibis/backends/singlestoredb/converter.py

@@ -2,6 +2,7 @@
 
 import datetime
 import json
+from functools import partial
 
 from ibis.formats.pandas import PandasData
 
@@ -359,27 +360,30 @@ def convert_SingleStoreDB_type(self, type_name):
 
         # SingleStoreDB-specific mappings
         singlestore_specific = {
-            "VECTOR": dt.binary,
+            "VECTOR": partial(dt.Array, dt.float32),  # Default to float32 array
             "BSON": dt.JSON,
             "GEOGRAPHY": dt.geometry,
             # Vector binary types
-            "FLOAT32_VECTOR": dt.binary,
-            "FLOAT64_VECTOR": dt.binary,
-            "INT8_VECTOR": dt.binary,
-            "INT16_VECTOR": dt.binary,
-            "INT32_VECTOR": dt.binary,
-            "INT64_VECTOR": dt.binary,
+            "FLOAT32_VECTOR": partial(dt.Array, dt.float32),
+            "FLOAT64_VECTOR": partial(dt.Array, dt.float64),
+            "INT8_VECTOR": partial(dt.Array, dt.int8),
+            "INT16_VECTOR": partial(dt.Array, dt.int16),
+            "INT32_VECTOR": partial(dt.Array, dt.int32),
+            "INT64_VECTOR": partial(dt.Array, dt.int64),
             # Vector JSON types
-            "FLOAT32_VECTOR_JSON": dt.JSON,
-            "FLOAT64_VECTOR_JSON": dt.JSON,
-            "INT8_VECTOR_JSON": dt.JSON,
-            "INT16_VECTOR_JSON": dt.JSON,
-            "INT32_VECTOR_JSON": dt.JSON,
-            "INT64_VECTOR_JSON": dt.JSON,
+            "FLOAT32_VECTOR_JSON": partial(dt.Array, dt.float32),
+            "FLOAT64_VECTOR_JSON": partial(dt.Array, dt.float64),
+            "INT8_VECTOR_JSON": partial(dt.Array, dt.int8),
+            "INT16_VECTOR_JSON": partial(dt.Array, dt.int16),
+            "INT32_VECTOR_JSON": partial(dt.Array, dt.int32),
+            "INT64_VECTOR_JSON": partial(dt.Array, dt.int64),
         }
 
         ibis_type = singlestore_specific.get(normalized_name)
         if ibis_type is not None:
+            # Handle partials (like VECTOR types)
+            if hasattr(ibis_type, "func"):
+                return ibis_type()  # Call the partial function
             return ibis_type
 
         # Default to string for unknown types

ibis/backends/singlestoredb/datatypes.py

@@ -162,15 +162,6 @@ def _type_from_cursor_info(
     flags = _FieldFlags(flags)
     typename = _type_codes.get(type_code)
 
-    # Handle SingleStoreDB vector types that may not be in _type_codes
-    if type_code in (3001, 3002, 3003, 3004, 3005, 3006):  # Vector types
-        # SingleStoreDB VECTOR types - map to Binary for now
-        # Could be enhanced to Array[Float32] or other appropriate types in future
-        return dt.Binary(nullable=True)
-    elif type_code in (2001, 2002, 2003, 2004, 2005, 2006):  # Vector JSON types
-        # SingleStoreDB VECTOR_JSON types - map to JSON
-        return dt.JSON(nullable=True)
-
     if typename is None:
         raise NotImplementedError(
             f"SingleStoreDB type code {type_code:d} is not supported"
@@ -212,10 +203,6 @@ def _type_from_cursor_info(
         # making them indistinguishable from TINYINT. The DESCRIBE-based schema
         # detection (via to_ibis method) can properly distinguish these types.
         typ = dt.Boolean
-    elif typename == "VECTOR":
-        # SingleStoreDB VECTOR type - typically used for AI/ML workloads
-        # For now, map to Binary; could be enhanced to Array[Float32] in future
-        typ = dt.Binary
     elif flags.is_set:
         # Sets are limited to strings in SingleStoreDB
         typ = dt.Array(dt.string)
@@ -254,7 +241,12 @@ def _type_from_cursor_info(
         typ = dt.Geometry
     else:
         typ = _type_mapping[typename]
-        if issubclass(typ, dt.SignedInteger) and flags.is_unsigned:
+        # Only apply unsigned logic to actual type classes, not partials
+        if (
+            hasattr(typ, "__mro__")
+            and issubclass(typ, dt.SignedInteger)
+            and flags.is_unsigned
+        ):
             typ = getattr(dt, f"U{typ.__name__}")
 
     # Projection columns are always nullable
@@ -304,20 +296,20 @@ def _decimal_length_to_precision(*, length: int, scale: int, is_unsigned: bool)
     # SingleStoreDB-specific types
     "BSON": dt.JSON,
     # Vector types for machine learning and AI workloads
-    "VECTOR": dt.Binary,  # General vector type
-    "FLOAT32_VECTOR": dt.Binary,
-    "FLOAT64_VECTOR": dt.Binary,
-    "INT8_VECTOR": dt.Binary,
-    "INT16_VECTOR": dt.Binary,
-    "INT32_VECTOR": dt.Binary,
-    "INT64_VECTOR": dt.Binary,
+    "VECTOR": partial(dt.Array, dt.Float32),  # General vector type
+    "FLOAT32_VECTOR": partial(dt.Array, dt.Float32),
+    "FLOAT64_VECTOR": partial(dt.Array, dt.Float64),
+    "INT8_VECTOR": partial(dt.Array, dt.Int8),
+    "INT16_VECTOR": partial(dt.Array, dt.Int16),
+    "INT32_VECTOR": partial(dt.Array, dt.Int32),
+    "INT64_VECTOR": partial(dt.Array, dt.Int64),
     # Vector JSON types (stored as JSON with vector semantics)
-    "FLOAT32_VECTOR_JSON": dt.JSON,
-    "FLOAT64_VECTOR_JSON": dt.JSON,
-    "INT8_VECTOR_JSON": dt.JSON,
-    "INT16_VECTOR_JSON": dt.JSON,
-    "INT32_VECTOR_JSON": dt.JSON,
-    "INT64_VECTOR_JSON": dt.JSON,
+    "FLOAT32_VECTOR_JSON": partial(dt.Array, dt.Float32),
+    "FLOAT64_VECTOR_JSON": partial(dt.Array, dt.Float64),
+    "INT8_VECTOR_JSON": partial(dt.Array, dt.Int8),
+    "INT16_VECTOR_JSON": partial(dt.Array, dt.Int16),
+    "INT32_VECTOR_JSON": partial(dt.Array, dt.Int32),
+    "INT64_VECTOR_JSON": partial(dt.Array, dt.Int64),
     # Extended types (SingleStoreDB-specific extensions)
     "GEOGRAPHY": dt.Geometry,  # Enhanced geospatial support
 }

ibis/backends/singlestoredb/tests/test_datatypes.py

@@ -54,7 +54,7 @@ def test_basic_type_mappings(self):
             # Collection types
             "SET": partial(dt.Array, dt.String),
             # SingleStoreDB-specific types
-            "VECTOR": dt.Binary,
+            "VECTOR": partial(dt.Array, dt.Float32),
             "GEOGRAPHY": dt.Geometry,
         }
 
@@ -74,7 +74,10 @@ def test_singlestoredb_specific_types(self):
         """Test SingleStoreDB-specific type extensions."""
         # Test VECTOR type
         assert "VECTOR" in _type_mapping
-        assert _type_mapping["VECTOR"] == dt.Binary
+        expected_vector_type = partial(dt.Array, dt.Float32)
+        actual_vector_type = _type_mapping["VECTOR"]
+        assert actual_vector_type.func == expected_vector_type.func
+        assert actual_vector_type.args == expected_vector_type.args
 
         # Test GEOGRAPHY type
         assert "GEOGRAPHY" in _type_mapping
@@ -147,8 +150,9 @@ def test_vector_type_handling(self):
             scale=0,
             multi_byte_maximum_length=1,
         )
-        # Vector types are currently mapped to Binary
-        assert isinstance(result, dt.Binary)
+        # Vector types are mapped to Array[Float32]
+        assert isinstance(result, dt.Array)
+        assert isinstance(result.value_type, dt.Float32)
 
         # Test FLOAT64_VECTOR type too
         result2 = _type_from_cursor_info(
@@ -158,7 +162,8 @@ def test_vector_type_handling(self):
             scale=0,
             multi_byte_maximum_length=1,
         )
-        assert isinstance(result2, dt.Binary)
+        assert isinstance(result2, dt.Array)
+        assert isinstance(result2.value_type, dt.Float64)
 
     def test_timestamp_with_timezone(self):
         """Test TIMESTAMP type includes UTC timezone by default."""
@@ -472,12 +477,16 @@ def test_convert_singlestoredb_type_method(self):
         assert converter.convert_SingleStoreDB_type("GEOMETRY") == dt.geometry
 
         # Test SingleStoreDB-specific types
-        assert converter.convert_SingleStoreDB_type("VECTOR") == dt.binary
+        vector_result = converter.convert_SingleStoreDB_type("VECTOR")
+        assert isinstance(vector_result, dt.Array)
+        assert isinstance(vector_result.value_type, dt.Float32)
         assert converter.convert_SingleStoreDB_type("GEOGRAPHY") == dt.geometry
 
         # Test case insensitivity
         assert converter.convert_SingleStoreDB_type("varchar") == dt.string
-        assert converter.convert_SingleStoreDB_type("Vector") == dt.binary
+        vector_result_case = converter.convert_SingleStoreDB_type("Vector")
+        assert isinstance(vector_result_case, dt.Array)
+        assert isinstance(vector_result_case.value_type, dt.Float32)
 
         # Test unknown type defaults to string
         assert converter.convert_SingleStoreDB_type("UNKNOWN_TYPE") == dt.string

ibis/backends/tests/snapshots/test_sql/test_mixed_qualified_and_unqualified_predicates/singlestoredb/out.sql

@@ -6,14 +6,14 @@ FROM (
     `t1`.`x`,
     `t1`.`y`,
     AVG(`t1`.`x`) OVER (
-      ORDER BY CASE WHEN NULL IS NULL THEN 1 ELSE 0 END, NULL ASC
+      ORDER BY NULL ASC NULLS LAST
       ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
     ) AS _w
   FROM (
     SELECT
       `t0`.`x`,
       SUM(`t0`.`x`) OVER (
-        ORDER BY CASE WHEN NULL IS NULL THEN 1 ELSE 0 END, NULL ASC
+        ORDER BY NULL ASC NULLS LAST
         ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
       ) AS `y`
     FROM `t` AS `t0`

ibis/backends/tests/snapshots/test_sql/test_order_by_no_deference_literals/singlestoredb/out.sql

@@ -4,4 +4,4 @@ SELECT
   'foo' AS `s`
 FROM `test` AS `t0`
 ORDER BY
-  CASE WHEN `t0`.`a` IS NULL THEN 1 ELSE 0 END, `t0`.`a` ASC
+  `t0`.`a` ASC NULLS LAST

ibis/backends/tests/snapshots/test_sql/test_rewrite_context/singlestoredb/out.sql

@@ -1,4 +1,4 @@
 SELECT
-  NTILE(2) OVER (ORDER BY RAND() ASC) - 1 AS `new_col`
+  NTILE(2) OVER (ORDER BY RAND() ASC NULLS LAST) - 1 AS `new_col`
 FROM `test` AS `t0`
 LIMIT 10

ibis/backends/tests/test_numeric.py

@@ -1374,11 +1374,6 @@ def test_clip(backend, alltypes, df, ibis_func, pandas_func):
     raises=PyDruidProgrammingError,
     reason="SQL query requires 'MIN' operator that is not supported.",
 )
-@pytest.mark.notyet(
-    ["singlestoredb"],
-    raises=SingleStoreDBOperationalError,
-    reason="Complex nested SQL exceeds SingleStoreDB stack size causing stack overflow",
-)
 def test_histogram(con, alltypes):
     n = 10
     hist = con.execute(alltypes.int_col.histogram(nbins=n).name("hist"))