pytest.fail with non-ascii characters raises an internal pytest error

[nicoddemus]

Fix #1178

(oops, sent the branch to pytest's repository by accident, sorry)

[RonnyPfannschmidt]

Does this one have a test for non Unicode non ASCII?

[nicoddemus]

I think test_pytest_fail_notrace covers it? (right above the test I added)

[RonnyPfannschmidt]

At first glance it seems to test only ASCII

[nicoddemus]

Hmmm good catch!

def test_pytest_fail_notrace_non_ascii_bytes(testdir):
    testdir.makepyfile(u"""
        # coding: utf-8
        import pytest

        def test_hello():
            pytest.fail('oh oh: ☺', pytrace=False)
    """)
    result = testdir.runpytest()
    if sys.version_info[0] >= 3:
        result.stdout.fnmatch_lines(['*test_hello*', "oh oh: ☺"])
    else:
        result.stdout.fnmatch_lines(['*test_hello*', "oh oh: *"])
    assert 'def test_hello' not in result.stdout.str()

Fails:

    and: u'INTERNALERROR>   File "X:\\pytest\\_pytest\\python.py", line 743, in _repr_failure_py'
    and: u'INTERNALERROR>     return py._builtin._totext(excinfo.value)'
    and: u'INTERNALERROR>   File "X:\\pytest\\_pytest\\runner.py", line 438, in __repr__'
    and: u'INTERNALERROR>     return py._builtin._totext(self.msg)'
    and: u"INTERNALERROR> UnicodeDecodeError: 'ascii' codec can't decode byte 0xe2 in position 7: ordinal not in range(128)"

Not sure what the appropriate fix for this issue would be then (totext might not be it). Do you have a suggestion? I will think a little more about it meanwhile.

[nicoddemus]

This seems to solve it:

diff --git a/_pytest/runner.py b/_pytest/runner.py
index 4f66aa9..cde94c8 100644
--- a/_pytest/runner.py
+++ b/_pytest/runner.py
@@ -435,7 +435,10 @@ class OutcomeException(Exception):

     def __repr__(self):
         if self.msg:
-            return py._builtin._totext(self.msg)
+            val = self.msg
+            if isinstance(val, bytes):
+                val = py._builtin._totext(val, errors='replace')
+            return val
         return "<%s instance>" %(self.__class__.__name__,)
     __str__ = __repr__

What do you think @RonnyPfannschmidt?

EDIT: updated diff

[RonnyPfannschmidt]

I'm fine e with anything that works in the demonstrated error cases

I suspect at least another upcoming overhaul

[nicoddemus]

Done

[RonnyPfannschmidt]

👍