Nonlocal

We say that a variable defined in a frame is local to that frame. A variable is nonlocal to a frame if it is defined in the environment that the frame belongs to but not the frame itself, i.e. in its parent or ancestor frame.

So far, we know that we can access variables in parent frames:

def make_adder(x):
    """
    Returns a function which takes in one argument y, and returns x + y.

    >>> add_three = make_adder(3)
    >>> add_three(2)
    5
    """
    def adder(y):
        return x + y
    return adder

Here, when we call make_adder, we create a function adder that is able to look up the name x in make_adder's frame and use its value.

However, we haven't been able to modify variables defined in parent frames. Consider the following function:

def make_adder_cumulative(x):
    """
    Returns a function which takes in one argument y and returns y + x + the sum of all previous y's.
    It also updates x to this new total.

    >>> add_all = make_adder_cumulative(3)
    >>> add_all(2) # 3 + 2
    5
    >>> add_all(5) # 3 + 2 + 5
    10
    """
    def adder(y):
        total = x + y
        x = total
        return total
    return adder

Here, when we call make_adder_cumulative, we create a function adder that is able to look up the name x in make_adder_cumulative's frame and use its value. However, adder also attempts to update the variable x in its parent frame. Running this function's doctests, we find that it causes the following error:

UnboundLocalError: local variable 'x' referenced before assignment

Why does this happen? When we execute an assignment statement, we are either creating a new binding in our current frame or we are updating an old one in the current frame. For example, the line x = x + y in adder, is creating the local variable x inside adder's frame. This assignment statement tells Python to expect a variable called x inside adder's frame, so Python will not look in parent frames for this variable. However, notice that we tried to compute x + y before the local variable was created! That's why we get the UnboundLocalError.

To avoid this problem, we introduce the nonlocal keyword. It allows us to update a variable in a parent frame! The only exception is the global frame. You cannot update a variable in the global frame using nonlocal.

Consider this example:

def make_adder_cumulative(x):
    """
    Returns a function which takes in one argument y and returns y + x + the sum of all previous y's.
    It also updates x to this new total.

    >>> add_all = make_adder_cumulative(3)
    >>> add_all(2) # 3 + 2
    5
    >>> add_all(5) # 3 + 2 + 5
    10
    """
    def adder(y):
        nonlocal x
        total = x + y
        x = total
        return total
    return adder

The line nonlocal x tells Python that x will not be local to this frame, so it will look for it in parent frames. Now we can update x without running into problems.

Mutability

We say that an object is mutable if its state can change as code is executed. The process of changing an object's state is called mutation. Examples of mutable objects include lists and dictionaries. Examples of objects that are not mutable include tuples and functions.

We have seen how to use the == operator to check if two expressions evaluate to equal values. We now introduce a new comparison operator, is, that checks whether two expressions evaluate to the same values.

Wait, what's the difference? For primitive values, there is none:

>>> 2 + 2 == 3 + 1
True
>>> 2 + 2 is 3 + 1
True

This is because all primitives have the same identity under the hood. However, with non-primitive values, such as lists, each object has its own identity. That means you can construct two objects that may look exactly the same but have different identities.

>>> lst1 = [1, 2, 3, 4]
>>> lst2 = [1, 2, 3, 4]
>>> lst1 == lst2
True
>>> lst1 is lst2
False

Here, although the lists referred to by lst1 and lst2 have equal contents, they are not the same object. In other words, they are the same in terms of equality, but not in terms of identity.

This is important in our discussion of mutability because when we mutate an object, we simply change its state, not its identity.

>>> lst1 = [1, 2, 3, 4]
>>> lst2 = lst1
>>> lst1.append(5)
>>> lst2
[1, 2, 3, 4, 5]
>>> lst1 is lst2
True

Iterators

An iterable is any object that can be iterated through, or gone through one element at a time. One construct that we've used to iterate through an iterable is a for loop:

for elem in iterable:
    # do something

for loops work on any object that is iterable. We previously described it as working with any sequence -- all sequences are iterable, but there are other objects that are also iterable! We define an iterable as an object on which calling the built-in function iter function returns an iterator. An iterator is another type of object that allows us to iterate through an iterable by keeping track of which element is next in the sequence.

To illustrate this, consider the following block of code, which does the exact same thing as a the for statement above:

iterator = iter(iterable)
try:
    while True:
        elem = next(iterator)
        # do something
except StopIteration:
    pass

Here's a breakdown of what's happening:

• First, the built-in iter function is called on the iterable to create a corresponding iterator.
• To get the next element in the sequence, the built-in next function is called on this iterator.
• When next is called but there are no elements left in the iterator, a StopIteration error is raised. In the for loop construct, this exception is caught and execution can continue.

Calling iter on an iterable multiple times returns a new iterator each time with distinct states (otherwise, you'd never be able to iterate through a iterable more than once). You can also call iter on the iterator itself, which will just return the same iterator without changing its state. However, note that you cannot call next directly on an iterable.

Let's see the iter and next functions in action with an iterable we're already familiar with -- a list.

>>> lst = [1, 2, 3, 4]
>>> next(lst)             # Calling next on an iterable
TypeError: 'list' object is not an iterator
>>> list_iter = iter(lst) # Creates an iterator for the list
>>> list_iter
<list_iterator object ...>
>>> next(list_iter)       # Calling next on an iterator
1
>>> next(list_iter)       # Calling next on the same iterator
2
>>> next(iter(list_iter)) # Calling iter on an iterator returns itself
3
>>> list_iter2 = iter(lst)
>>> next(list_iter2)      # Second iterator has new state
1
>>> next(list_iter)       # First iterator is unaffected by second iterator
4
>>> next(list_iter)       # No elements left!
StopIteration
>>> lst                   # Original iterable is unaffected
[1, 2, 3, 4]

Since you can call iter on iterators, this tells us that that they are also iterables! Note that while all iterators are iterables, the converse is not true - that is, not all iterables are iterators. You can use iterators wherever you can use iterables, but note that since iterators keep their state, they're only good to iterate through an iterable once:

>>> list_iter = iter([4, 3, 2, 1])
>>> for e in list_iter:
...     print(e)
4
3
2
1
>>> for e in list_iter:
...     print(e)

Analogy: An iterable is like a book (one can flip through the pages) and an iterator for a book would be a bookmark (saves the position and can locate the next page). Calling iter on a book gives you a new bookmark independent of other bookmarks, but calling iter on a bookmark gives you the bookmark itself, without changing its position at all. Calling next on the bookmark moves it to the next page, but does not change the pages in the book. Calling next on the book wouldn't make sense semantically. We can also have multiple bookmarks, all independent of each other.

Iterable Uses

We know that lists are one type of built-in iterable objects. You may have also encountered the range(start, end) function, which creates an iterable of ascending integers from start (inclusive) to end (exclusive).

>>> for x in range(2, 6):
...     print(x)
...
2
3
4
5

Ranges are useful for many things, including performing some operations for a particular number of iterations or iterating through the indices of a list.

There are also some built-in functions that take in iterables and return useful results:

• map(f, iterable) - Creates an iterable over f(x) for x in iterable. In some cases, computing a list of the values in this iterable will give us the same result as [func(x) for x in iterable]. However, it's important to keep in mind that iterators can potentially have infinite values because they are evaluated lazily, while lists cannot have infinite elements.
• filter(f, iterable) - Creates iterator over x for each x in iterable if f(x)
• zip(iterables*) - Creates an iterable over co-indexed tuples with elements from each of the iterables
• reversed(iterable) - Creates iterator over all the elements in the input iterable in reverse order
• list(iterable) - Creates a list containing all the elements in the input iterable
• tuple(iterable) - Creates a tuple containing all the elements in the input iterable
• sorted(iterable) - Creates a sorted list containing all the elements in the input iterable

Generators

We can create our own custom iterators by writing a generator function, which returns a special type of iterator called a generator. Generator functions have yield statements within the body of the function instead of return statements. Calling a generator function will return a generator object and will not execute the body of the function.

For example, let's consider the following generator function:

def countdown(n):
    print("Beginning countdown!")
    while n >= 0:
        yield n
        n -= 1
    print("Blastoff!")

Calling countdown(k) will return a generator object that counts down from k to 0. Since generators are iterators, we can call iter on the resulting object, which will simply return the same object. Note that the body is not executed at this point; nothing is printed and no numbers are output.

>>> c = countdown(5)
>>> c
<generator object countdown ...>
>>> c is iter(c)
True

So how is the counting done? Again, since generators are iterators, we call next on them to get the next element! The first time next is called, execution begins at the first line of the function body and continues until the yield statement is reached. The result of evaluating the expression in the yield statement is returned. The following interactive session continues from the one above.

>>> next(c)
Beginning countdown!
5

Unlike functions we've seen before in this course, generator functions can remember their state. On any consecutive calls to next, execution picks up from the line after the yield statement that was previously executed. Like the first call to next, execution will continue until the next yield statement is reached. Note that because of this, Beginning countdown! doesn't get printed again.

>>> next(c)
4
>>> next(c)
3

The next 3 calls to next will continue to yield consecutive descending integers until 0. On the following call, a StopIteration error will be raised because there are no more values to yield (i.e. the end of the function body was reached before hitting a yield statement).

>>> next(c)
2
>>> next(c)
1
>>> next(c)
0
>>> next(c)
Blastoff!
StopIteration

Separate calls to countdown will create distinct generator objects with their own state. Usually, generators shouldn't restart. If you'd like to reset the sequence, create another generator object by calling the generator function again.

>>> c1, c2 = countdown(5), countdown(5)
>>> c1 is c2
False
>>> next(c1)
5
>>> next(c2)
5

Here is a summary of the above:

• A generator function has a yield statement and returns a generator object.
• Calling the iter function on a generator object returns the same object without modifying its current state.
• The body of a generator function is not evaluated until next is called on a resulting generator object. Calling the next function on a generator object computes and returns the next object in its sequence. If the sequence is exhausted, StopIteration is raised.

• A generator "remembers" its state for the next next call. Therefore,

  • the first next call works like this:

    1. Enter the function and run until the line with yield.
    2. Return the value in the yield statement, but remember the state of the function for future next calls.

  • And subsequent next calls work like this:

    1. Re-enter the function, start at the line after the yield statement that was previously executed, and run until the next yield statement.
    2. Return the value in the yield statement, but remember the state of the function for future next calls.
• Calling a generator function returns a brand new generator object (like calling iter on an iterable object).
• A generator should not restart unless it's defined that way. To start over from the first element in a generator, just call the generator function again to create a new generator.

Another useful tool for generators is the yield from statement (introduced in Python 3.3). yield from will yield all values from an iterator or iterable.

>>> def gen_list(lst):
...     yield from lst
...
>>> g = gen_list([1, 2, 3, 4])
>>> next(g)
1
>>> next(g)
2
>>> next(g)
3
>>> next(g)
4
>>> next(g)
StopIteration