behaviour:
>>> name = 'woo\raa'
>>> open(name, 'w')
aa', mode 'w' at 0x295a8a0>

expected:
>>> name = 'woo\raa'
>>> open(name, 'w')
<open file 'woo\raa', mode 'w' at 0x295a8a0>

note:
don't ask why i tried this chunk of code in the first place

Funny one :D  Reproduced on linux:

  >>> open('/tmp/t\nest', 'w')
  <open file '/tmp/t
  est', mode 'w' at 0x7f11268a19c0>

file_repr in Objects/fileobject.c calls http://docs.python.org/c-api/unicode#PyUnicode_AsUnicodeEscapeString, equivalent to encode('unicode-escape'), so backslashes should be escaped.

Duh, obviously that code branch is used only for unicode paths:

  >>> open('/tmp/t\nest', 'w')
  <open file '/tmp/t
  est', mode 'w' at 0x7f6f0f3dd9c0>
  >>> open(u'/tmp/t\nest', 'w')
  <open file u'/tmp/t\nest', mode 'w' at 0x7f6f0f3dda50>

There does not seem to be something similar in http://docs.python.org/c-api/string, so I guess one would have to create intermediary objects to decode str to unicode, transform with unicode-escape and convert back to str.

The attached patch seems to do the trick (not sure if it's the best way to fix the issue though):
>>> open('woo\raa')
<open file 'woo\raa', mode 'r' at 0xb77c2aa8>
>>> open('woo\ra\'a', 'w')
<open file "woo\ra'a", mode 'w' at 0xb77c2b88>
>>> open('woo\ra\'a"', 'w')
<open file 'woo\ra\'a"', mode 'w' at 0xb77c2b18>
>>> 

It's more or less equivalent to:
- return "<open file '%s', mode '%s' at %p>" % (fname, mode, addr)
+ return "<open file %s, mode '%s' at %p>" % (repr(fname), mode, addr)

1. PyObject_Repr() should IMO be preferred (it's the abstract, high-level function).
2. You must check the result for NULL before calling PyString_AsString() on it.

I think you want to decref the result of PyObject_Repr after the fact, too

New changeset 6c1964dee98b by Ezio Melotti in branch '2.7':
#14161: fix the __repr__ of file objects to escape the file name.
http://hg.python.org/cpython/rev/6c1964dee98b

New changeset 86c749151660 by Ezio Melotti in branch '2.7':
#14161: fix compile error under Windows.
http://hg.python.org/cpython/rev/86c749151660

New changeset 6b1fad34d893 by Ezio Melotti in branch '2.7':
#14161: fix test failures on Windows.
http://hg.python.org/cpython/rev/6b1fad34d893

Not sure why this is still open -- probably I was just waiting for the buildbots and forgot it open.  The issue seems fixed, so I'm going to close it now.

wtf? you made it possible to return NULL in some case

sorry for the buzz, i got myself up to date on the c api now

why is there still the unicode case?
the PyObject_Repr variant should work fine in both cases