With -r 76870, on Fedora 11, amd x80_64, regrtest of test_format fails 
when the -j  option is combined with the -v option. It happens whether 
testing everything or just the one test.

  ./python -Ebb Lib/test/regrtest -j3 -vv 
or
  ./python -Ebb Lib/test/regrtest -j3 -vv test_format

The error report is like
"""
FAILED (errors=1)
test test_format failed -- Traceback (most recent call last):
  File "/home/jbase/checkouts/py/py32/Lib/test/test_format.py", line 
221, in test_format
    testformat("%r", "\u0374", "'\u0374'")   # printable
  File "/home/jbase/checkouts/py/py32/Lib/test/test_format.py", line 20, 
in testformat
    (formatstr, args, output), end=' ')
UnicodeEncodeError: 'ascii' codec can't encode character '\u0374' in 
position 8: ordinal not in range(128)
1 test failed:
    test_format
"""

~jim

I can duplicate this on Fedora Core 6, x86, r76872. I simplified the
command line down to:

./python Lib/test/regrtest.py -j1 -v test_format

(and any value for j >= 1 causes the error).

This is because in -j mode stdout is a pipe, and so python defaults to
ascii for the output encoding.  I'm not sure what the best way is to fix
this.

Maybe the "sys.stdout.write" method can be wrapped in a decorator or a
context manager?

See example of decorator attached.

James, can you set PYTHONIOENCODING to a reasonable value in the test environment and see if it fixes the bug? (More info on http://docs.python.org/dev/using/cmdline#envvar-PYTHONIOENCODING)

Funny, because it works here (Mandriva Linux, py3k branch).
Can you still reproduce?

(in any case, it doesn't make much sense to run in verbose mode with -jN, since it will interleave output from several tests)

Unable to reproduce, here. It seems fixed.

Seems to work, even when forcing LANG=C.