One question which is asked with surprising frequency in #python is how
to yield multiple objects at a time from an iterable object. That is,
given [1, 2, 3, 4, 5, 6], get [(1, 2), (3, 4), (5, 6)].

The "grouper" function in the itertools recipes page provides one
pattern for this. A similar function (which behaves differently when the
length of the iterable is not evenly divisible by n) looks like this:

def group(iterable, n=2):
    return itertools.izip(*(iter(iterable),)*n)

This code is fairly opaque to the novice. It is ugly, and takes a bit of
head-scratching to realize exactly what it is doing. Because this
operation is asked for with some frequency, and because this general
implementation is so ugly, I believe it belongs in the library.

There is a related function which is asked for much less frequently, but
which is a more general case of the group() function listed above. This
other function has a third "step" argument. This argument controls how
far each group is in advance of the previous group. For example:

list(group([1, 2, 3, 4], n=2, step=1)) -> [(1, 2), (2, 3), (3, 4)]

The original function is equivalent to this function when step is equal
to n.

Please find attached a patch with implementation, documentation, and tests.

Sorry, I'm not interested in adding this to the module. Discussions to-
date on the subject seem to show more interest in playing with grouper
variants than in actual use cases. While the recipe given in the docs
is somewhat opaque, it runs at C-speed (zero trips around the eval-
loop) and it is encapsulated in a re-usable function. Writing this in C
does nothing to improve the situation. Also, when people like to play
with variants, there is no general agreement on useful requirements
(like fill-in behavior or raising an exception on uneven length
inputs). Trying to write option to meet all needs (n=2, step=1) makes
the code more difficult to learn and use -- see several variants in
Alex's Python Cookbook. Another issue is that we have to be very
selective about adding tools to the module. Each addition makes the
overall toolset harder to use -- it is better to have a good set of
basic building blocks.