In Py>=2.6, kwargs can be general mappings

[robertwb]

See  http://www.mail-archive.com/cython-dev@codespeak.net/msg08051.html and  http://bugs.python.org/issue1686487

While testing type inference in mpi4py, I noticed that code like the
one below is failing in the last line. It seems that in  Py>=2.6,
Python makes a coercion on "kw" and builds an actual dict instance
before calling.

class Foo:
   def view(self, **kw):
       print (kw)
       for k in kw.keys():
           print (k)

Foo().view(a=0, b=1, c=2)

class MyDict(object):
   def __getitem__(self, k):
       assert k == 'a'
       return 7
   def keys(self):
       return ['a']

Foo().view(**MyDict()) # <-- Here is the issue

[scoder]

The expected implementation is

tmp = MyDict()
if not isinstace(tmp, dict):
    tmp = dict(tmp)
Foo().view(**tmp)

[scoder]

Implemented here:

 http://hg.cython.org/cython-devel/rev/df9b7285ec6a

[scoder]

Better implementation here:

 http://hg.cython.org/cython-devel/rev/5d1ea5364347