Fix hashed linting and backward compatibility issues.

formulaic/transforms/hashed.py

@@ -1,16 +1,26 @@
+from __future__ import annotations
+
+import sys
 from hashlib import md5
 from numbers import Number
-from typing import Any, Callable, Dict, Iterable, List, Optional, Union
+from typing import Any, Callable, Dict, Iterable, List, Optional, Union, TYPE_CHECKING
 
 import numpy as np
 
 from formulaic.materializers.types import FactorValues
 
 from .contrasts import Contrasts, encode_contrasts
 
+if TYPE_CHECKING:
+    from formulaic.model_spec import ModelSpec
+
 
 def md5_to_int(s: str) -> int:
-    return int(md5(s.encode(), usedforsecurity=False).hexdigest(), 16)
+    if sys.version_info >= (3, 9):
+        hashed = md5(s.encode(), usedforsecurity=False)
+    else:
+        hashed = md5(s.encode())
+    return int(hashed.hexdigest(), 16)
 
 
 def hashed(
@@ -22,7 +32,7 @@ def hashed(
     *,
     hash_func: Callable[[str], int] = md5_to_int,
     spans_intercept: bool = False,
-):
+) -> FactorValues:
     """
     Deterministically hashes the values of a factor into a fixed number of levels.
     If `levels` is large, you will likely want to use this transform in conjunction
@@ -48,16 +58,14 @@ def encoder(
         reduced_rank: bool,
         drop_rows: List[int],
         encoder_state: Dict[str, Any],
-        model_spec,
-    ):
+        model_spec: ModelSpec,
+    ) -> FactorValues:
         values = np.array(values)
-        empty_state = {}
         return encode_contrasts(
             values,
             contrasts=contrasts,
             levels=np.arange(levels),
             reduced_rank=reduced_rank,
-            _state=empty_state,
             _spec=model_spec,
         )