Improved internal storage format for Bit [skip ci]

Author: ankane

pgvector/bit.py

@@ -7,8 +7,8 @@
 class Bit:
     def __init__(self, value: bytes | str | list[bool] | np.ndarray[tuple[int], np.dtype[np.bool | np.uint8]]) -> None:
         if isinstance(value, bytes):
-            self._len = 8 * len(value)
-            self._data = value
+            _len = 8 * len(value)
+            _data = value
         else:
             if isinstance(value, str):
                 value = [v != '0' for v in value]
@@ -30,28 +30,34 @@ def __init__(self, value: bytes | str | list[bool] | np.ndarray[tuple[int], np.d
                 if value.ndim != 1:
                     raise ValueError('expected ndim to be 1')
 
-            self._len = len(value)
-            self._data = np.packbits(value).tobytes()
+            _len = len(value)
+            _data = np.packbits(value).tobytes()
+
+        self._value = pack('>i', _len) + _data
 
     def __repr__(self) -> str:
         return f'Bit({self.to_text()})'
 
     def __eq__(self, other: object) -> bool:
         if isinstance(other, self.__class__):
-            return self._len == other._len and self._data == other._data
+            return self.to_binary() == other.to_binary()
         return False
 
+    def _len(self):
+        _len, = unpack_from('>i', self._value)
+        return _len
+
     def to_list(self) -> list[bool]:
         return self.to_numpy().tolist()
 
     def to_numpy(self) -> np.ndarray[tuple[int], np.dtype[np.bool]]:
-        return np.unpackbits(np.frombuffer(self._data, dtype=np.uint8), count=self._len).astype(bool)
+        return np.unpackbits(np.frombuffer(self._value[4:], dtype=np.uint8), count=self._len()).astype(bool)
 
     def to_text(self) -> str:
-        return ''.join(format(v, '08b') for v in self._data)[:self._len]
+        return ''.join(format(v, '08b') for v in self._value[4:])[:self._len()]
 
     def to_binary(self) -> bytes:
-        return pack('>i', self._len) + self._data
+        return self._value
 
     @classmethod
     def from_text(cls, value: str) -> Bit:
@@ -63,8 +69,7 @@ def from_binary(cls, value: bytes) -> Bit:
             raise ValueError('expected bytes')
 
         bit = cls.__new__(cls)
-        bit._len = unpack_from('>i', value)[0]
-        bit._data = value[4:]
+        bit._value = value
         return bit
 
     @classmethod