The pprint function in the python 3.1 pprint module fails when
printing a dictionary containing more than one item and with one item
being a user-defined type. It seems pprint tries to sort the keys but
fails, (maybe because calling __lt__ on a user-defined type doesn't
bind its first argument to 'self'? Looking into pprint.py would
probably yield the answer).

This seems related to bpo-3976 but it was fixed in r76389 and
r76390. My example below fails in r79147. I'm not in a position to check more recent revisions right now.

In Python 2.6, the following works fine:

Python 2.6.5 (r265:79063, Apr 16 2010, 13:09:56) 
[GCC 4.4.3] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> from pprint import pprint
>>> class A(object): pass
... 
>>> pprint({A:1, 1:2})
{1: 2, <class __main__.A at 0xb77dc47c>: 1}

But in Python 3.1, it fails with the error below:

Python 3.1.2 (r312:79147, Apr 15 2010, 12:35:07) 
[GCC 4.4.3] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> from pprint import pprint
>>> class A: pass
... 
>>> pprint({A:1, 1:2})
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python3.1/pprint.py", line 55, in pprint
    printer.pprint(object)
  File "/usr/lib/python3.1/pprint.py", line 132, in pprint
    self._format(object, self._stream, 0, 0, {}, 0)
  File "/usr/lib/python3.1/pprint.py", line 155, in _format
    rep = self._repr(object, context, level - 1)
  File "/usr/lib/python3.1/pprint.py", line 242, in _repr
    self._depth, level)
  File "/usr/lib/python3.1/pprint.py", line 254, in format
    return _safe_repr(object, context, maxlevels, level)
  File "/usr/lib/python3.1/pprint.py", line 296, in _safe_repr
    items = sorted(object.items(), key=_safe_tuple)
  File "/usr/lib/python3.1/pprint.py", line 89, in __lt__
    rv = self.obj.__lt__(other.obj)
TypeError: expected 1 arguments, got 0

FWIW, the problem still occurs on the most recent release31-maint checkout (as of r85323), and does not happen on py3k (3.2a2).

If I'm understanding this correctly, this fails on 3.1 and not (although, actually, it does) on py3k/3.2 because:

• pprint._safe_key.__lt__ checks "rv = self.obj.__lt__(other.obj)" and falls back to id comparison if rv is NotImplemented

• If the object passed to _safe_key is a class, self.obj.__lt__ will expect *self* as well as the other object. Therefore the verification above fails with "TypeError: expected 1 arguments, got 0". You can see that pprint works with an instance:

>>> pprint.pprint({A(): 1, 1: 2})
{<__main__.A object at 0x8594d4c>: 1, 1: 2}

• Finally, this works on py3k *for your example* because, for some reason, on py3k the comparison is being based on the 1 key. That is, the comparison on _safe_key.__lt__ happens to be 1.__lt__(A), instead of A.__lt__(1). Perhaps hashing changed after the 3.1 release?

Anyway, py3k still fails when you force the comparison to happen on the class:

>>> class B(object): pass
... 
>>> pprint.pprint({A: 1, B: 2})
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/rbp/python/dev/py3k/Lib/pprint.py", line 55, in pprint
    printer.pprint(object)
  File "/home/rbp/python/dev/py3k/Lib/pprint.py", line 132, in pprint
    self._format(object, self._stream, 0, 0, {}, 0)
  File "/home/rbp/python/dev/py3k/Lib/pprint.py", line 155, in _format
    rep = self._repr(object, context, level - 1)
  File "/home/rbp/python/dev/py3k/Lib/pprint.py", line 245, in _repr
    self._depth, level)
  File "/home/rbp/python/dev/py3k/Lib/pprint.py", line 257, in format
    return _safe_repr(object, context, maxlevels, level)
  File "/home/rbp/python/dev/py3k/Lib/pprint.py", line 299, in _safe_repr
    items = sorted(object.items(), key=_safe_tuple)
  File "/home/rbp/python/dev/py3k/Lib/pprint.py", line 89, in __lt__
    rv = self.obj.__lt__(other.obj)
TypeError: expected 1 arguments, got 0
>>> 

So, basically, the fix on bpo-3976 does't (always) work when there are classes as dict keys. I think switching from

rv = self.obj.__lt__(other.obj)
if rv is NotImplemented:

to something like

try:
rv = self.obj < other.obj
except TypeError:
rv = (str(type(self.obj)), id(self.obj)) < \
(str(type(other.obj)), id(other.obj))

solves this. Or we can check first whether self.obj is a 'type', but I think it gets convoluted.

If anyone can confirm that that's the way to go, I can produce one (though it's a trivial one). Raymond?

A simpler change would to replace:
rv = self.obj.__lt__(other.obj)
with
rv = type(self.obj).__lt__(self.obj, other.obj)

Confirmed on python 3.2 and 3.3.

Here's a patch with fix and tests.

Note that class objects are not comparable and have the same type so they
fall back on sorting by id. Should we sort them by name as a special case instead?

New changeset 03cda5360dc6 by Florent Xicluna in branch '3.2':
Issues bpo-10017 and bpo-14998: Fix TypeError using pprint on dictionaries with unorderable key.
http://hg.python.org/cpython/rev/03cda5360dc6

New changeset 4d0dcfbdf45b by Florent Xicluna in branch 'default':
Issues bpo-10017 and bpo-14998: Fix TypeError using pprint on dictionaries with unorderable key.
http://hg.python.org/cpython/rev/4d0dcfbdf45b

Thanks for this patch.
I've reviewed the issue and merged the patch of issue bpo-14998 which fixes both cases.