Python nested functions variable scoping

In Python 3, you can use the nonlocal statement to access non-local, non-global scopes.

The nonlocal statement causes a variable definition to bind to a previously created variable in the nearest scope. Here are some examples to illustrate:

def sum_list_items(_list):
    total = 0

    def do_the_sum(_list):
        for i in _list:
            total += i

    do_the_sum(_list)

    return total

sum_list_items([1, 2, 3])

The above example will fail with the error: UnboundLocalError: local variable 'total' referenced before assignment

Using nonlocal we can get the code to work:

def sum_list_items(_list):
    total = 0

    def do_the_sum(_list):

        # Define the total variable as non-local, causing it to bind
        # to the nearest non-global variable also called total.
        nonlocal total

        for i in _list:
            total += i

    do_the_sum(_list)

    return total

sum_list_items([1, 2, 3])

But what does "nearest" mean? Here is another example:

def sum_list_items(_list):

    total = 0

    def do_the_sum(_list):

        # The nonlocal total binds to this variable.
        total = 0

        def do_core_computations(_list):

            # Define the total variable as non-local, causing it to bind
            # to the nearest non-global variable also called total.
            nonlocal total

            for i in _list:
                total += i

        do_core_computations(_list)

    do_the_sum(_list)

    return total

sum_list_items([1, 2, 3])

In the above example, total will bind to the variable defined inside the do_the_sum function, and not the outer variable defined in the sum_list_items function, so the code will return 0. Note that it is still possible to do double nesting such as this: if total is declared nonlocal in do_the_sum the above example would work as expected.

def sum_list_items(_list):

    # The nonlocal total binds to this variable.
    total = 0

    def do_the_sum(_list):

        def do_core_computations(_list):

            # Define the total variable as non-local, causing it to bind
            # to the nearest non-global variable also called total.
            nonlocal total

            for i in _list:
                total += i

        do_core_computations(_list)

    do_the_sum(_list)

    return total

sum_list_items([1, 2, 3])

In the above example the nonlocal assignment traverses up two levels before it locates the total variable that is local to sum_list_items.

Here's an illustration that gets to the essence of David's answer.

def outer():
    a = 0
    b = 1

    def inner():
        print a
        print b
        #b = 4

    inner()

outer()

With the statement b = 4 commented out, this code outputs 0 1, just what you'd expect.

But if you uncomment that line, on the line print b, you get the error

UnboundLocalError: local variable 'b' referenced before assignment

It seems mysterious that the presence of b = 4 might somehow make b disappear on the lines that precede it. But the text David quotes explains why: during static analysis, the interpreter determines that b is assigned to in inner, and that it is therefore a local variable of inner. The print line attempts to print the b in that inner scope before it has been assigned.

When I run your code I get this error:

UnboundLocalError: local variable '_total' referenced before assignment

This problem is caused by this line:

_total += PRICE_RANGES[key][0]

The documentation about Scopes and Namespaces says this:

A special quirk of Python is that – if no global statement is in effect – assignments to names always go into the innermost scope. Assignments do not copy data — they just bind names to objects.

So since the line is effectively saying:

_total = _total + PRICE_RANGES[key][0]

it creates _total in the namespace of recurse(). Since _total is then new and unassigned you can't use it in the addition.

This is a variation of redman's solution, but using a proper namespace instead of an array to encapsulate the variable:

def foo():
    class local:
        counter = 0
    def bar():
        print(local.counter)
        local.counter += 1
    bar()
    bar()
    bar()

foo()
foo()

I'm not sure if using a class object this way is considered an ugly hack or a proper coding technique in the python community, but it works fine in python 2.x and 3.x (tested with 2.7.3 and 3.2.3). I'm also unsure about the run-time efficiency of this solution.