Reduced copying

Author: ankane

pgvector/bit.py

@@ -74,12 +74,9 @@ def from_text(cls, value: str) -> Bit:
         return cls(str(value))
 
     @classmethod
-    def from_binary(cls, value: bytes) -> Bit:
-        if not isinstance(value, bytes):
-            raise ValueError('expected bytes')
-
+    def from_binary(cls, value: bytes | bytearray | memoryview) -> Bit:
         length, = unpack_from('>i', value)
-        data = value[4:]
+        data = memoryview(value)[4:].tobytes()
 
         if len(data) != (length + 7) // 8:
             raise ValueError('invalid length')

pgvector/halfvec.py

@@ -64,17 +64,18 @@ def from_text(cls, value: str) -> HalfVector:
         return cls(cls._from_text(value))
 
     @classmethod
-    def from_binary(cls, value: bytes) -> HalfVector:
+    def from_binary(cls, value: bytes | bytearray | memoryview) -> HalfVector:
         dim, unused = struct.unpack_from('>HH', value)
-        data = value[4:]
+        data = memoryview(value)[4:]
 
         if len(data) != 2 * dim:
             raise ValueError('invalid length')
 
         if unused != 0:
             raise ValueError('expected unused to be 0')
 
-        arr = array.array('H', data)
+        arr = array.array('H')
+        arr.frombytes(data)
         if sys.byteorder != 'big':
             arr.byteswap()
 

pgvector/psycopg/halfvec.py

@@ -35,8 +35,6 @@ class HalfVectorBinaryLoader(HalfVectorLoader):
     format = Format.BINARY
 
     def load(self, data: Buffer) -> HalfVector | None:
-        if isinstance(data, (bytearray, memoryview)):
-            data = bytes(data)
         return HalfVector.from_binary(data)
 
 

pgvector/psycopg/sparsevec.py

@@ -35,8 +35,6 @@ class SparseVectorBinaryLoader(SparseVectorLoader):
     format = Format.BINARY
 
     def load(self, data: Buffer) -> SparseVector | None:
-        if isinstance(data, (bytearray, memoryview)):
-            data = bytes(data)
         return SparseVector.from_binary(data)
 
 

pgvector/psycopg/vector.py

@@ -46,8 +46,6 @@ class VectorBinaryLoader(VectorLoader):
     format = Format.BINARY
 
     def load(self, data: Buffer) -> Vector | None:
-        if isinstance(data, (bytearray, memoryview)):
-            data = bytes(data)
         return Vector.from_binary(data)
 
 

pgvector/sparsevec.py

@@ -127,7 +127,7 @@ def from_text(cls, value: str) -> SparseVector:
         return cls._from_parts(int(dim), indices, values)
 
     @classmethod
-    def from_binary(cls, value: bytes) -> SparseVector:
+    def from_binary(cls, value: bytes | bytearray | memoryview) -> SparseVector:
         dim, nnz, unused = unpack_from('>iii', value)
 
         if len(value) != 12 + 8 * nnz:

pgvector/vector.py

@@ -62,17 +62,18 @@ def from_text(cls, value: str) -> Vector:
         return cls(cls._from_text(value))
 
     @classmethod
-    def from_binary(cls, value: bytes) -> Vector:
+    def from_binary(cls, value: bytes | bytearray | memoryview) -> Vector:
         dim, unused = struct.unpack_from('>HH', value)
-        data = value[4:]
+        data = memoryview(value)[4:]
 
         if len(data) != 4 * dim:
             raise ValueError('invalid length')
 
         if unused != 0:
             raise ValueError('expected unused to be 0')
 
-        arr = array.array('f', data)
+        arr = array.array('f')
+        arr.frombytes(data)
         if sys.byteorder != 'big':
             arr.byteswap()