Merge pull request #5770 from y-p/GH5766

BUG: regression in read_csv parser handling of usecols GH5766

Author: y-p

pandas/io/tests/test_parsers.py

@@ -2049,6 +2049,16 @@ def test_usecols(self):
                                 header=None, usecols=['b', 'c'])
         tm.assert_frame_equal(result2, result)
 
+
+        # 5766
+        result = self.read_csv(StringIO(data), names=['a', 'b'],
+                               header=None, usecols=[0, 1])
+
+        expected = self.read_csv(StringIO(data), names=['a', 'b', 'c'],
+                                 header=None)
+        expected = expected[['a', 'b']]
+        tm.assert_frame_equal(result, expected)
+
         # length conflict, passed names and usecols disagree
         self.assertRaises(ValueError, self.read_csv, StringIO(data),
                           names=['a', 'b'], usecols=[1], header=None)

pandas/parser.pyx

@@ -898,6 +898,9 @@ cdef class TextReader:
             if i < self.leading_cols:
                 # Pass through leading columns always
                 name = i
+            elif self.usecols and nused == len(self.usecols):
+                # Once we've gathered all requested columns, stop. GH5766
+                break
             else:
                 name = self._get_column_name(i, nused)
                 if self.has_usecols and not (i in self.usecols or