Fix BUG: read_sql tries to convert blob/varbinary to string with pyarrow backend

doc/source/whatsnew/v2.3.0.rst

@@ -134,6 +134,7 @@ MultiIndex
 I/O
 ^^^
 - :meth:`DataFrame.to_excel` was storing decimals as strings instead of numbers (:issue:`49598`)
+- Bug in :func:`read_sql` causing an unintended exception when byte data was being converted to string when using the pyarrow dtype_backend (:issue:`59242`)
 -
 
 Period

pandas/core/internals/construction.py

@@ -968,10 +968,6 @@ def convert(arr):
                     # i.e. maybe_convert_objects didn't convert
                     convert_to_nullable_dtype = dtype_backend != "numpy"
                     arr = maybe_infer_to_datetimelike(arr, convert_to_nullable_dtype)
-                    if convert_to_nullable_dtype and arr.dtype == np.dtype("O"):
-                        new_dtype = StringDtype()
-                        arr_cls = new_dtype.construct_array_type()
-                        arr = arr_cls._from_sequence(arr, dtype=new_dtype)
                 elif dtype_backend != "numpy" and isinstance(arr, np.ndarray):
                     if arr.dtype.kind in "iufb":
                         arr = pd_array(arr, copy=False)

pandas/tests/io/test_sql.py

@@ -4358,3 +4358,27 @@ def test_xsqlite_if_exists(sqlite_buildin):
         (5, "E"),
     ]
     drop_table(table_name, sqlite_buildin)
+
+
+@pytest.mark.parametrize("dtype_backend", ["pyarrow", "numpy_nullable", lib.no_default])
+def test_bytes_column(sqlite_buildin, dtype_backend):
+    pa = pytest.importorskip("pyarrow")
+    """
+    Regression test for (#59242)
+    Bytes being returned in a column that could not be converted
+    to a string would raise a UnicodeDecodeError
+    when using dtype_backend='pyarrow' or dtype_backend='numpy_nullable'
+    """
+    query = """
+    select cast(x'0123456789abcdef0123456789abcdef' as blob) a
+    """
+    df = pd.read_sql(query, sqlite_buildin, dtype_backend=dtype_backend)
+    expected = DataFrame(
+        [
+            {
+                "a": b"\x01#Eg\x89\xab\xcd\xef\x01#Eg\x89\xab\xcd\xef",
+            }
+        ],
+        dtype=(pd.ArrowDtype(pa.binary()) if dtype_backend == "pyarrow" else "O"),
+    )
+    tm.assert_frame_equal(df, expected)