BUG: Avoid flaky usecols set in C engine

Closes gh-14792.

Author: gfyoung

doc/source/whatsnew/v0.20.0.txt

@@ -318,3 +318,4 @@ Bug Fixes
 - Require at least 0.23 version of cython to avoid problems with character encodings (:issue:`14699`)
 - Bug in converting object elements of array-like objects to unsigned 64-bit integers (:issue:`4471`)
 - Bug in ``pd.pivot_table()`` where no error was raised when values argument was not in the columns (:issue:`14938`)
+- Bug in ``pd.read_csv()`` for the C engine where ``usecols`` were being indexed incorrectly with ``parse_dates`` (:issue:`14792`)

pandas/io/parsers.py

@@ -997,14 +997,14 @@ def _validate_usecols_arg(usecols):
 
     if usecols is not None:
         if callable(usecols):
-            return usecols
+            return usecols, None
         usecols_dtype = lib.infer_dtype(usecols)
         if usecols_dtype not in ('empty', 'integer',
                                  'string', 'unicode'):
             raise ValueError(msg)
 
-        return set(usecols)
-    return usecols
+        return set(usecols), usecols_dtype
+    return usecols, None
 
 
 def _validate_parse_dates_arg(parse_dates):
@@ -1473,7 +1473,8 @@ def __init__(self, src, **kwds):
         self._reader = _parser.TextReader(src, **kwds)
 
         # XXX
-        self.usecols = _validate_usecols_arg(self._reader.usecols)
+        self.usecols, self.usecols_dtype = _validate_usecols_arg(
+            self._reader.usecols)
 
         passed_names = self.names is None
 
@@ -1550,11 +1551,22 @@ def close(self):
 
     def _set_noconvert_columns(self):
         names = self.orig_names
-        usecols = self.usecols
+        if self.usecols_dtype == 'integer':
+            # A set of integers will be converted to a list in
+            # the correct order every single time.
+            usecols = list(self.usecols)
+        elif (callable(self.usecols) or
+                self.usecols_dtype not in ('empty', None)):
+            # The names attribute should have the correct columns
+            # in the proper order for indexing with parse_dates.
+            usecols = self.names[:]
+        else:
+            # Usecols is empty.
+            usecols = None
 
         def _set(x):
-            if usecols and is_integer(x):
-                x = list(usecols)[x]
+            if usecols is not None and is_integer(x):
+                x = usecols[x]
 
             if not is_integer(x):
                 x = names.index(x)
@@ -1792,7 +1804,7 @@ def __init__(self, f, **kwds):
         self.skipinitialspace = kwds['skipinitialspace']
         self.lineterminator = kwds['lineterminator']
         self.quoting = kwds['quoting']
-        self.usecols = _validate_usecols_arg(kwds['usecols'])
+        self.usecols, _ = _validate_usecols_arg(kwds['usecols'])
         self.skip_blank_lines = kwds['skip_blank_lines']
 
         self.names_passed = kwds['names'] or None

pandas/io/tests/parser/usecols.py

@@ -200,6 +200,31 @@ def test_usecols_with_parse_dates(self):
                            parse_dates=parse_dates)
         tm.assert_frame_equal(df, expected)
 
+        # See gh-14792
+        s = """a,b,c,d,e,f,g,h,i,j
+        2016/09/21,1,1,2,3,4,5,6,7,8"""
+        parse_dates = [0]
+        usecols = list('abcdefghij')
+        cols = {'a': Timestamp('2016-09-21'),
+                'b': [1], 'c': [1], 'd': [2],
+                'e': [3], 'f': [4], 'g': [5],
+                'h': [6], 'i': [7], 'j': [8]}
+        expected = DataFrame(cols, columns=usecols)
+        df = self.read_csv(StringIO(s), usecols=usecols,
+                           parse_dates=parse_dates)
+        tm.assert_frame_equal(df, expected)
+
+        s = """a,b,c,d,e,f,g,h,i,j\n2016/09/21,1,1,2,3,4,5,6,7,8"""
+        parse_dates = [[0, 1]]
+        usecols = list('abcdefghij')
+        cols = {'a_b': '2016/09/21 1',
+                'c': [1], 'd': [2], 'e': [3], 'f': [4],
+                'g': [5], 'h': [6], 'i': [7], 'j': [8]}
+        expected = DataFrame(cols, columns=['a_b'] + list('cdefghij'))
+        df = self.read_csv(StringIO(s), usecols=usecols,
+                           parse_dates=parse_dates)
+        tm.assert_frame_equal(df, expected)
+
     def test_usecols_with_parse_dates_and_full_names(self):
         # See gh-9755
         s = """0,1,20140101,0900,4