Review

Author: jonashaag

pandas/io/common.py

@@ -5,6 +5,7 @@
 import codecs
 from collections import abc
 import dataclasses
+import functools
 import gzip
 from io import (
     BufferedIOBase,
@@ -1130,20 +1131,19 @@ def _is_binary_mode(handle: FilePath | BaseBuffer, mode: str) -> bool:
     )
 
 
-def _get_binary_io_classes(_cache: list[type] = []) -> tuple[type, ...]:
+@functools.lru_cache
+def _get_binary_io_classes() -> tuple[type, ...]:
     """IO classes that that expect bytes"""
-    if not _cache:
-        binary_classes: tuple[type, ...] = (BufferedIOBase, RawIOBase)
-
-        # python-zstandard doesn't use any of the builtin base classes; instead we
-        # have to use the `zstd.ZstdDecompressionReader` class for isinstance checks.
-        # Unfortunately `zstd.ZstdDecompressionReader` isn't exposed by python-zstandard
-        # so we have to get it from a `zstd.ZstdDecompressor` instance.
-        # See also https://github.com/indygreg/python-zstandard/pull/165.
-        zstd = import_optional_dependency("zstandard", errors="ignore")
-        if zstd is not None:
-            with zstd.ZstdDecompressor().stream_reader(b"") as reader:
-                binary_classes += (type(reader),)
-
-        _cache[:] = binary_classes
-    return tuple(_cache)
+    binary_classes: tuple[type, ...] = (BufferedIOBase, RawIOBase)
+
+    # python-zstandard doesn't use any of the builtin base classes; instead we
+    # have to use the `zstd.ZstdDecompressionReader` class for isinstance checks.
+    # Unfortunately `zstd.ZstdDecompressionReader` isn't exposed by python-zstandard
+    # so we have to get it from a `zstd.ZstdDecompressor` instance.
+    # See also https://github.com/indygreg/python-zstandard/pull/165.
+    zstd = import_optional_dependency("zstandard", errors="ignore")
+    if zstd is not None:
+        with zstd.ZstdDecompressor().stream_reader(b"") as reader:
+            binary_classes += (type(reader),)
+
+    return binary_classes