Support for nonlocal declaration

[spkersten]

This PR implements #494.

There is one note I should make: An error message is generated when nonlocal x follows a definition of x, like in this case:

def outer():
    x = 1
    def inner():
        x = 2
        nonlocal x
    inner()
    print(x)

However, Python 3.4 reports just a warning. I made this choice, because it was the easiest way to report an error (as it should) when there is a nonlocal declaration of a variable with the same name as a function parameter:

def outer():
    x = 1
    def inner(x):
        nonlocal x

I don't think it is such a bad idea to report an error in the first case too. In the first code snippet, the output is 2. Which is (at least to me) a bit counter intuitive as the nonlocal x apparently affects an earlier statement. Furthermore, generating an error message in this case does not exclude any uses of nonlocal as it can be moved to the front of the function with changing the semantics.

Nevertheless, if you feel this is overly restrictive, let me know and I will try to find another way of dealing with this.

[JukkaL]

Minor nit: redundant spaces around =

[JukkaL]

Looks good! This fixes one of the biggest remaining Python syntax incompatibility issues.

Some notes above. Also, there should be a test case for type checking code using nonlocal (to test that assignment to variable defined using nonlocal uses the inferred type from the outer scope rather than inferring a completely new type).

[spkersten]

Thanks for the quick review. I'll fix the issues and add some test cases.

"there should be a test case for type checking", that was a subtle way of saying that it doesn't work yet :-). I thought that my changes in semanal.py would be enough, because that is what happens for global. Turns out it is not enough and there is a bug for global. I'll try to fix both.

[JukkaL]

Thanks for addressing the issues! I'm going to merge this now.