diff options
Diffstat (limited to 'tests/test_phystokens.py')
-rw-r--r-- | tests/test_phystokens.py | 11 |
1 files changed, 3 insertions, 8 deletions
diff --git a/tests/test_phystokens.py b/tests/test_phystokens.py index c1e51f1c..9ff053c4 100644 --- a/tests/test_phystokens.py +++ b/tests/test_phystokens.py @@ -29,7 +29,7 @@ class PhysTokensTest(CoverageTest): """Tokenize `source`, then put it back together, should be the same.""" tokenized = "" for line in source_token_lines(source): - text = "".join([t for _,t in line]) + text = "".join(t for _, t in line) tokenized += text + "\n" # source_token_lines doesn't preserve trailing spaces, so trim all that # before comparing. @@ -86,11 +86,6 @@ if sys.version_info < (3, 0): run_in_temp_dir = False - if sys.version_info >= (2,4): - default_encoding = 'ascii' - else: - default_encoding = 'iso-8859-1' - def test_detect_source_encoding(self): # Various forms from http://www.python.org/dev/peps/pep-0263/ source = "# coding=cp850\n\n" @@ -110,11 +105,11 @@ if sys.version_info < (3, 0): def test_dont_detect_source_encoding_on_third_line(self): # A coding declaration doesn't count on the third line. source = "\n\n# coding=cp850\n\n" - self.assertEqual(source_encoding(source), self.default_encoding) + self.assertEqual(source_encoding(source), 'ascii') def test_detect_source_encoding_of_empty_file(self): # An important edge case. - self.assertEqual(source_encoding(""), self.default_encoding) + self.assertEqual(source_encoding(""), 'ascii') def test_bom(self): # A BOM means utf-8. |