This allows unicode arguments to be passed as keywords to functions to
help "from __future__ import unicode_literals". See
http://mail.python.org/pipermail/python-dev/2009-January/085113.html.

The patch works when the unicode contains only ascii, but crashes with
the following input:

>>> def f(hehe): return 42
... 
>>> f(**{u'hehe': 1})
42
>>> f(**{u'héhé': 1})
Segmentation fault

The two PyString_AsString(keyword) calls are now wrong and should be
changed.

This patch should fix that.

Sorry, but _PyUnicode_AsDefaultEncodedString returns either a borrowed reference (if 
errors==NULL) or a new reference (if errors!=NULL).
In either case it's wrong to DECREF the string when you simply take the buffer's 
address:

>>> f(**{u'someLongString':2})
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: f() got an unexpected keyword argument 'ÛÛÛÛÛÛÛgString'

Interesting, I didn't know that about _PyUnicode_AsDefaultEncodedString.
Here's another patch.

The patch looks good now.

Thanks for the reviews! Applied in r68805.

Really trivial back port of r68805 from trunk.

release blocker for 2.6.5

Why backport? Isn't this a new feature?

Maybe not.  It's arguably a bug that adding unicode_literals breaks keyword arguments.  Guido seems to (mostly) agree that it's a bug we just never got around to fixing.

http://www.mail-archive.com/python-dev@python.org/msg45346.html

Done in r78027.