The docs on file's constructor says that the 'U' mode should not work
with '+', and yet 'rU+' does not throw an error.

Attaching a patch that checks for '+' in the mode string, updates the 
docs, and tests bad mode strings.

I can confirm that:

- the patch applies cleanly to the release26-maint branch, with the
  exception of the Misc/NEWS portion::

  $ hg branch
  release26-maint
  $ ./python -E -tt Lib/test/regrtest.py test_file
  test_file
  1 test OK.
  $ patch -p0 < /tmp/issue2091.diff 
  patching file Objects/fileobject.c
  Hunk #1 succeeded at 209 (offset 54 lines).
  Hunk #2 succeeded at 2269 (offset 194 lines).
  patching file Misc/NEWS
  Hunk #1 FAILED at 12.
  1 out of 1 hunk FAILED -- saving rejects to file Misc/NEWS.rej
  patching file Lib/test/test_file.py
  Hunk #1 succeeded at 176 (offset 55 lines).

- the added test fails before rebuilding Objects/fileobject.c::

  $ ./python -E -tt Lib/test/regrtest.py test_file
  test_file
  test test_file failed -- Traceback (most recent call last):
    File "/home/tseaver/projects/hg-repo/py-2.6/Lib/test/test_file.py", line 181, in testModeStrings
      f = open(TESTFN, mode)
  IOError: [Errno 2] No such file or directory: '@test'
  
  1 test failed:
      test_file

- the added test passes after rebuilding Objects/fielobject.c::

  $ make
  gcc -pthread -c -fno-strict-aliasing -g -O2 -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes  -I. -IInclude -I./Include   -DPy_BUILD_CORE -o Objects/fileobject.o Objects/fileobject.c
  ...
  $ ./python -E -tt Lib/test/regrtest.py test_file
  test_file
  1 test OK.

The patch is now way out of date to the extent that I can't find the code in fileobject.c, perhaps I'm just blind  Can someone please provide a new patch, thanks.

Of course, the implementation is now in the io module:
Modules/_io/_iomodule.c *and* Lib/_pyio.py

In Python 3, the documentation no longer mentions that 'U' should not work with '+' (or 'w' or 'a', for that matter), and the code throws ValueError if 'U' is used with 'w' or 'a', but not '+'.

On the other hand, the documentation *does* mention that 'U' is for backwards compatibility and shouldn't be used with new code.

In light of this, how do you suggest to proceed with this issue?

It seems to me that adding the proper check is a good idea. It also may not be obvious to some that 'U' is now more or less an alias for 'r'. I have updated the patch so that it at least applies.

I also removed a redundant `reading = 1`