Dict comprehensions need to be backported.

Here is a patch to backport dict and set comprehensions to the trunk.

The patch preserves the Python 3.x's syntax and semantics of the
feature. Although this makes dict and set comprehensions is inconsistent
with list comprehension, I believe this is reasonable since the backport
aimed to improve the compatibility of Python 2.x with Python 3.x.

The patch also adds support for dict and set comprehensions to the
'compiler' package. However, the support is quirky and use the list
comprehension semantics of Python 2.x. This allowed me to keep the patch
simple and to keep myself sane ;-).

Finally, the patch changes the name of the following syntax nodes:
'gen_expr', 'gen_iter', 'gen_if', and 'testlist_gexp'. I don't know if
we have compatibility requirements for this; so I don't know if this is
an issue.

Alexandre, I can't get this to apply cleanly to my svn trunk checkout 
(r74328) (see attached log).  Am I doing something stupid?

You need to apply the set literal patch I posted in issue2335 first.

Ah, that works.  Thanks!  (I'll pay more attention to the tracker 
Dependencies field in future.)

If there are no objections, I will commit this patch later this week
after issue #2335.

Committed in r77422.

How about fixing set.__repr__ ?

Great they are backported to 2.7.  Dictionary and set comprehensions are not documented in the tutorial of 2.7, though they are in the language reference. Is it OK to add these to the tutorial? I've uploaded a patch.

> Is it OK to add these to the tutorial? I've uploaded a patch.

Yes, but please create a new issue.

> How about fixing set.__repr__ ?

Is it broken?

I think metal means that the different ways set is repr'd in 2.7 and 3.

In 2.7:

In [9]: a = {x for x in 'abracadabra' if x not in 'abc'}

In [10]: repr(a)
Out[10]: "set(['r', 'd'])"

In 3.2：
In [6]: a = {x for x in 'abracadabra' if x not in 'abc'}

In [7]: repr(a)
Out[7]: "{'r', 'd'}"

We were saying not changing __repr__ in #2335.