BUG: Help python csv engine read binary buffers (#27925)

* BUG: Help python csv engine read binary buffers

The file buffer given to read_csv could have been opened in
binary mode, but the python csv reader errors on binary buffers.

closes #23779

Author: fiendish

doc/source/whatsnew/v1.0.0.rst

@@ -158,7 +158,7 @@ MultiIndex
 I/O
 ^^^
 
--
+- :meth:`read_csv` now accepts binary mode file buffers when using the Python csv engine (:issue:`23779`)
 -
 
 Plotting

pandas/io/common.py

@@ -5,7 +5,7 @@
 import csv
 import gzip
 from http.client import HTTPException  # noqa
-from io import BytesIO
+from io import BufferedIOBase, BytesIO
 import mmap
 import os
 import pathlib
@@ -344,9 +344,9 @@ def _get_handle(
     try:
         from s3fs import S3File
 
-        need_text_wrapping = (BytesIO, S3File)
+        need_text_wrapping = (BufferedIOBase, S3File)
     except ImportError:
-        need_text_wrapping = (BytesIO,)
+        need_text_wrapping = BufferedIOBase
 
     handles = list()
     f = path_or_buf
@@ -422,8 +422,10 @@ def _get_handle(
     if is_text and (compression or isinstance(f, need_text_wrapping)):
         from io import TextIOWrapper
 
-        f = TextIOWrapper(f, encoding=encoding, newline="")
-        handles.append(f)
+        g = TextIOWrapper(f, encoding=encoding, newline="")
+        if not isinstance(f, BufferedIOBase):
+            handles.append(g)
+        f = g
 
     if memory_map and hasattr(f, "fileno"):
         try:

pandas/tests/io/parser/test_common.py

@@ -2020,9 +2020,34 @@ def test_file_handles_with_open(all_parsers, csv1):
     # Don't close user provided file handles.
     parser = all_parsers
 
-    with open(csv1, "r") as f:
-        parser.read_csv(f)
-        assert not f.closed
+    for mode in ["r", "rb"]:
+        with open(csv1, mode) as f:
+            parser.read_csv(f)
+            assert not f.closed
+
+
+@pytest.mark.parametrize(
+    "fname,encoding",
+    [
+        ("test1.csv", "utf-8"),
+        ("unicode_series.csv", "latin-1"),
+        ("sauron.SHIFT_JIS.csv", "shiftjis"),
+    ],
+)
+def test_binary_mode_file_buffers(all_parsers, csv_dir_path, fname, encoding):
+    # gh-23779: Python csv engine shouldn't error on files opened in binary.
+    parser = all_parsers
+
+    fpath = os.path.join(csv_dir_path, fname)
+    expected = parser.read_csv(fpath, encoding=encoding)
+
+    with open(fpath, mode="r", encoding=encoding) as fa:
+        result = parser.read_csv(fa)
+    tm.assert_frame_equal(expected, result)
+
+    with open(fpath, mode="rb") as fb:
+        result = parser.read_csv(fb, encoding=encoding)
+    tm.assert_frame_equal(expected, result)
 
 
 def test_invalid_file_buffer_class(all_parsers):