Logged In: YES 
user_id=113328

Unicode objects do not have this behaviour. For example:

>>> u''.join(gen(None))
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
  File "<stdin>", line 3, in gen
TypeError: gen() TypeError

The offending code is at line 1610 or so of stringobject.c.
The equivalent Unicode code starts at line 3955 of
unicodeobject.c.

The string code does a 2-pass approach to calculate the size
of the result, allocate space, and then build the value. The
Unicode version resizes as it goes along. This *may* be a
significant speed optimisation (on the assumption that
strings are more commonly used than Unicode objects), but I
can't test (no MSVC7 to build with).

If the speed issue is not significant, I'd recommend
rewriting the string code to use the same approach the
Unicode code uses. Otherwise, the documentation for str.join
should clarify these points:

1. The sequence being joined is materialised as a tuple
(PySequence_Fast) - this may have an impact on generators
which use a lot of memory.
2. TypeErrors produced by materialising the sequence being
joined will be caught and re-raised with a different message.