Types: Add support for BINARY columns and improve support for FLOATs

Author: amotl

src/sqlalchemy_cratedb/compiler.py

@@ -197,6 +197,9 @@ def visit_unique_constraint(self, constraint, **kw):
                       "they will be omitted when generating DDL statements.")
         return None
 
+    def visit_create_index(self, create, **kw):
+        return "SELECT 1;"
+
 
 class CrateTypeCompiler(compiler.GenericTypeCompiler):
 
@@ -245,6 +248,30 @@ def visit_FLOAT_VECTOR(self, type_, **kw):
             raise ValueError("FloatVector must be initialized with dimension size")
         return f"FLOAT_VECTOR({dimensions})"
 
+    def visit_BLOB(self, type_, **kw):
+        return "STRING"
+
+    def visit_FLOAT(self, type_, **kw):
+        """
+        From `sqlalchemy.sql.sqltypes.Float`.
+
+        When a :paramref:`.Float.precision` is not provided in a
+        :class:`_types.Float` type some backend may compile this type as
+        an 8 bytes / 64 bit float datatype. To use a 4 bytes / 32 bit float
+        datatype a precision <= 24 can usually be provided or the
+        :class:`_types.REAL` type can be used.
+        This is known to be the case in the PostgreSQL and MSSQL dialects
+        that render the type as ``FLOAT`` that's in both an alias of
+        ``DOUBLE PRECISION``. Other third party dialects may have similar
+        behavior.
+        """
+        if not type_.precision:
+            return "FLOAT"
+        elif type_.precision <= 24:
+            return "FLOAT"
+        else:
+            return "DOUBLE"
+
 
 class CrateCompiler(compiler.SQLCompiler):
 

src/sqlalchemy_cratedb/dialect.py

@@ -35,6 +35,7 @@
 from crate.client.exceptions import TimezoneUnawareException
 from .sa_version import SA_VERSION, SA_1_4, SA_2_0
 from .type import FloatVector, ObjectArray, ObjectType
+from .type.binary import LargeBinary
 
 TYPES_MAP = {
     "boolean": sqltypes.Boolean,
@@ -153,7 +154,8 @@ def process(value):
 
 colspecs = {
     sqltypes.DateTime: DateTime,
-    sqltypes.Date: Date
+    sqltypes.Date: Date,
+    sqltypes.LargeBinary: LargeBinary,
 }
 
 

src/sqlalchemy_cratedb/type/__init__.py

@@ -1,4 +1,5 @@
 from .array import ObjectArray
+from .binary import LargeBinary
 from .geo import Geopoint, Geoshape
 from .object import ObjectType
 from .vector import FloatVector, knn_match

src/sqlalchemy_cratedb/type/binary.py

@@ -0,0 +1,44 @@
+import base64
+import sqlalchemy as sa
+
+
+class LargeBinary(sa.String):
+
+    """A type for large binary byte data.
+
+    The :class:`.LargeBinary` type corresponds to a large and/or unlengthed
+    binary type for the target platform, such as BLOB on MySQL and BYTEA for
+    PostgreSQL.  It also handles the necessary conversions for the DBAPI.
+
+    """
+
+    __visit_name__ = "large_binary"
+
+    def bind_processor(self, dialect):
+        if dialect.dbapi is None:
+            return None
+
+        # TODO: DBAPIBinary = dialect.dbapi.Binary
+
+        def process(value):
+            if value is not None:
+                # TODO: return DBAPIBinary(value)
+                return base64.b64encode(value).decode()
+            else:
+                return None
+
+        return process
+
+    # Python 3 has native bytes() type
+    # both sqlite3 and pg8000 seem to return it,
+    # psycopg2 as of 2.5 returns 'memoryview'
+    def result_processor(self, dialect, coltype):
+        if dialect.returns_native_bytes:
+            return None
+
+        def process(value):
+            if value is not None:
+                return base64.b64decode(value)
+            return value
+
+        return process