Fixed issue with reading out of order lists, added exception for pyarrow.

Author: eicchen

pandas/io/parsers/readers.py

@@ -1535,16 +1535,15 @@ def read(self, nrows: int | None = None) -> DataFrame:
             else:
                 dtype = None
 
-            if dtype is None and get_option("future.usecols_use_order"):
+            if get_option("future.usecols_use_order"):
                 if usecols is None or isfunction(usecols):
                     # Doesn't change anything if function or None gets passed
                     pass
                 elif len(usecols) == len(columns):
                     # uses size of number in usecols to determine corresponding columns
-                    usecols_sorted = sorted(
-                        range(len(usecols)), key=lambda i: usecols[i]
-                    )
-                    columns = [columns[i] for i in usecols_sorted]
+                    value_ranked = {v: i for i, v in enumerate(sorted(usecols))}
+                    usecols_pressed = [value_ranked[v] for v in usecols]
+                    columns = [columns[i] for i in usecols_pressed]
                     col_dict = {k: col_dict[k] for k in columns}
 
             if dtype is not None:

pandas/tests/io/parser/usecols/test_usecols_basic.py

@@ -554,22 +554,26 @@ def test_usecols_dtype(all_parsers):
 def test_usecols_order(all_parsers, usecols, usecols_use_order):
     # TODOE add portion in doc for 3.0 transition
     parser = all_parsers
+    pyarrow_flag = False
     data = """\
 a,b,c,d
 1,2,3,0
 4,5,6,0
 7,8,9,0
 10,11,12,13"""
 
-    msg = "The pyarrow engine does not allow 'usecols' to be integer column positions"
-    if parser.engine == "pyarrow" and isinstance(usecols[0], int):
-        with pytest.raises(ValueError, match=msg):
-            parser.read_csv(StringIO(data), usecols=usecols)
-        return
-
-    result = parser.read_csv(StringIO(data), usecols=usecols)
+    if parser.engine == "pyarrow":
+        if isinstance(usecols[0], int):
+            msg = "The pyarrow engine does not allow 'usecols' to be integer column"
+            with pytest.raises(ValueError, match=msg):
+                parser.read_csv(StringIO(data), usecols=usecols)
+            return
+        else:
+            # looks like pyarrow already considers column order by default.
+            # Modifies test to account for it in selecting expected df
+            pyarrow_flag = True
 
-    if usecols_use_order:
+    if usecols_use_order or pyarrow_flag:
         expected = DataFrame(
             {"d": [0, 0, 0, 13], "a": [1, 4, 7, 10], "c": [3, 6, 9, 12]}
         )
@@ -579,4 +583,5 @@ def test_usecols_order(all_parsers, usecols, usecols_use_order):
         )
 
     with option_context("future.usecols_use_order", usecols_use_order):
+        result = parser.read_csv(StringIO(data), usecols=usecols)
         tm.assert_frame_equal(result, expected)