BUG: parser_source as a filename with multibyte characters in Windows(non utf-8 filesystem)

pandas/io/tests/data/日本語ファイル名テスト_read_csv_in_win_filesystem.csv

@@ -0,0 +1,3 @@
+A,B,C
+100,200,300
+aaa,bbb,ccc

pandas/io/tests/test_parsers.py

@@ -1822,6 +1822,10 @@ def test_utf16_example(self):
  result = self.read_table(buf, encoding='utf-16')
  self.assertEquals(len(result), 50)
 
+ def test_read_csv_example_in_windows_filesystem(self):
+ path = tm.get_data_path(u'日本語ファイル名テスト_read_csv_in_win_filesystem.csv')
+ self.read_csv(path)
+
  def test_converters_corner_with_nas(self):
  # skip aberration observed on Win64 Python 3.2.2
  if hash(np.int64(-1)) != -2:

pandas/parser.pyx

@@ -535,7 +535,10 @@ cdef class TextReader:
 
  if isinstance(source, basestring):
  if not isinstance(source, bytes):
- source = source.encode('utf-8')
+ if sys.getfilesystemencoding() is None:
+ source = source.encode('utf-8')
+ else:
+ source = source.encode(sys.getfilesystemencoding())
 
  if self.memory_map:
  ptr = new_mmap(source)