COMPAT: Consider Python 2.x tarfiles file-like

Tarfile.ExFileObject has no "next" method in
Python 2.x, making it an invalid file-like
object in read_csv. However, they can be
read in just fine, meaning our check is too
strict for file-like. This commit relaxes
the check to just look for "__iter__".

Closes pandas-devgh-16530.

Author: gfyoung

doc/source/whatsnew/v0.20.2.txt

@@ -64,6 +64,7 @@ I/O
 - Bug in pd.read_csv() when comment is passed in space deliminted text files (:issue:`16472`)
 - Bug that would force importing of the clipboard routines unnecessarily, potentially causing an import error on startup (:issue:`16288`)
 - Bug that raised IndexError HTML-rendering an empty DataFrame (:issue:`15953`)
+- Bug in ``pd.read_csv()`` in which tarfile object inputs were raising an error in Python 2.x for the C engine (:issue:`16530`)
 
 
 Plotting

pandas/core/dtypes/inference.py

@@ -171,7 +171,7 @@ def is_file_like(obj):
     if not (hasattr(obj, 'read') or hasattr(obj, 'write')):
         return False
 
-    if not is_iterator(obj):
+    if not hasattr(obj, "__iter__"):
         return False
 
     return True

pandas/tests/dtypes/test_inference.py

@@ -120,9 +120,9 @@ class MockFile(object):
     m = MockFile()
     assert not is_file(m)
 
+    # gh-16530: Valid iterator just means we have the
+    # __iter__ attribute for our purposes.
     MockFile.__iter__ = lambda self: self
-    MockFile.__next__ = lambda self: 0
-    MockFile.next = MockFile.__next__
 
     # Valid write-only file
     m = MockFile()

pandas/tests/io/parser/c_parser_only.py

@@ -8,6 +8,7 @@
 """
 
 import sys
+import tarfile
 
 import pytest
 import numpy as np
@@ -16,7 +17,7 @@
 import pandas.util.testing as tm
 from pandas import DataFrame
 from pandas import compat
-from pandas.compat import StringIO, range, lrange
+from pandas.compat import StringIO, range, lrange, u
 
 
 class CParserTests(object):
@@ -446,3 +447,46 @@ def test_comment_whitespace_delimited(self):
                               [7, np.nan],
                               [8, np.nan]])
         tm.assert_frame_equal(df, expected)
+
+    def test_file_like_no_next(self):
+        # gh-16530: the file-like need not have a "next" or "__next__"
+        # attribute despite having an "__iter__" attribute.
+        #
+        # NOTE: This is only true for the C engine, not Python engine.
+        class NoNextBuffer(StringIO):
+            def __next__(self):
+                raise AttributeError("No next method")
+
+            next = __next__
+
+        data = "a\n1"
+
+        expected = pd.DataFrame({"a": [1]})
+        result = self.read_csv(NoNextBuffer(data))
+
+        tm.assert_frame_equal(result, expected)
+
+    def test_read_tarfile(self):
+        # see gh-16530
+        #
+        # Unfortunately, Python's CSV library can't handle
+        # tarfile objects (expects string, not bytes when
+        # iterating through a file-like).
+        expected = pd.DataFrame({"a": [1]})
+
+        tar_path = u("__%s__.tar" % tm.rands(10))
+        file_path = u("__%s__.csv" % tm.rands(10))
+
+        with tm.ensure_clean(tar_path) as tar_path:
+            with tm.ensure_clean(file_path) as file_path:
+                expected.to_csv(file_path, index=False)
+
+                tar = tarfile.open(tar_path, "w")
+                tar.add(file_path)
+                tar.close()
+
+                tar = tarfile.open(tar_path, "r")
+                data_file = tar.extractfile(file_path)
+                out = self.read_csv(data_file)
+
+                tm.assert_frame_equal(out, expected)