I was playing around with python 3's io functions, and I found that when trying to write to 
an encoded utf-16 file that TextIOWrapper.write re-writes the utf-16 bom for every string:

>>> f=open('foo', 'w', encoding='utf-16')
>>> print('1234', file=f)
>>> print('5678', file=f)
>>> open('foo', 'rb').read()
b'\xff\xfe1\x002\x003\x004\x00\xff\xfe\n\x00\xff\xfe5\x006\x007\x008\x00\xff\xfe\n\x00'
>>> open('foo', 'r', encoding='utf-16').read()
'1234\ufeff\n\ufeff5678\ufeff\n'
>>> 

With the attached patch, it appears to generate the correct file:

>>> f=open('foo', 'w', encoding='utf-16')
>>> print('1234', file=f)
>>> print('5678', file=f)
>>> open('foo', 'rb').read()
b'\xff\xfe1\x002\x003\x004\x00\n\x005\x006\x007\x008\x00\n\x00'
>>> open('foo', 'r', encoding='utf-16').read()
'1234\n5678\n'
>>>

I was aware of this but tried to ignore it until it became important. :-(

Committed fix in r59832. Thanks!

P.S. Guido, what this comment, in write(), is about?
# XXX What if we were just reading?