Seems at test for os.environ that uses mapping_tests.py to make 
sure the object returned by os.environ has a mapping 
environment is failing under 2.4 (fine under 2.5) at least on OS X:

test.test_support.TestFailed: Traceback (most recent call last):
  File "/Users/drifty/Code/cvs/Trees/python24/Lib/test/
mapping_tests.py", line 204, in test_update
    d.update(SimpleUserDict())
  File "/Users/drifty/Code/cvs/Trees/python24/Lib/os.py", line 467, 
in update
    for k, v in items:
  File "/Users/drifty/Code/cvs/Trees/python24/Lib/test/
mapping_tests.py", line 202, in __getitem__
    return self.d[i]
KeyError: 0

Logged In: YES 
user_id=31435

Confirming that test_os fails the same way on WinXP, on 
current release24-maint branch.  But it does not fail under 
the released 2.4.  Brett, can you clarify whether you're 
talking about 2.4 or about release24-maint here?

Logged In: YES 
user_id=21627

This is now fixed in os.py 1.86 and 1.83.2.2, test_os.py 1.30.

Logged In: YES 
user_id=89016

The new implementation of os._Environ.update() assumes that
the argument dict is a mapping, if it provides a method
items(), otherwise it treats it as an item sequence.
Unfortunately the SimpleUserDict from mapping_tests.py
provides only keys() and __getitem__().

So what is the minimal mapping interface? UserDict.DictMixin
claims that keys and __getitem__ (and __setitem__ and
__delitem__) is enough and adding __contains__, __iter__ and
iteritems increases performance.

So should update() try iteritems() or should an items()
method be added to the SimpleUserDict in the test?

CVS HEAD does not fail, because a test method test_update()
has been added there, which hides the original method from
mapping_tests.BasicTestMappingProtocol. Adding a call
mapping_tests.BasicTestMappingProtocol.test_update(self)
there results in the same exception as on the
release24-maint branch.

Logged In: YES 
user_id=89016

Oops, seems are comments have crossed, thanks for the fix.