Understanding Python's "with" statement

Mon, Jul 9, 2012

What is it?

Python’s with statement provides a very convenient way of dealing with the situation where you have to do a setup and teardown to make something happen. A very good example for this is the situation where you want to gain a handler to a file, read data from the file and the close the file handler.

Without the with statement, one would write something along the lines of:

file = open("/tmp/foo.txt")
data = file.read()

There are two annoying things here. First, you end up forgetting to close the file handler. The second is how to handle exceptions that may occur once the file handler has been obtained. One could write something like this to get around this:

file = open("/tmp/foo.txt")
    data = file.read()

While this works well, it is unnecessarily verbose. This is where with is useful. The good thing about with apart from the better syntax is that it is very good handling exceptions. The above code would look like this, when using with:

with open("/tmp/foo.txt") as file:
   data = file.read()

How does it work?

While this might look like magic, the way Python handles with is more clever than magic. The basic idea is that the statement after with has to evaluate an object that responds to an __enter__() as well as an __exit__() function.

After the statement that follows with is evaluated, the __enter__() function on the resulting object is called. The value returned by this function is assigned to the variable following as. After every statement in the block is evaluated, the __exit__() function is called.

This can be demonstrated with the following example:

#!/usr/bin/env python
# with_example01.py

class Sample:
    def __enter__(self):
        print "In __enter__()"
        return "Foo"

    def __exit__(self, type, value, trace):
        print "In __exit__()"

def get_sample():
    return Sample()

with get_sample() as sample:
    print "sample:", sample

When executed, this will result in:

bash-3.2$ ./with_example01.py
In __enter__()
sample: Foo
In __exit__()

As you can see,

  1. The __enter__() function is executed
  2. The value returned by it - in this case "Foo" is assigned to sample
  3. The body of the block is executed, thereby printing the value of sample ie. "Foo"
  4. The __exit__() function is called.

What makes with really powerful is the fact that it can handle exceptions. You would have noticed that the __exit__() function for Sample takes three arguments - val, type and trace. These are useful in exception handling. Let’s see how this works by modifying the above example.

#!/usr/bin/env python
# with_example02.py

class Sample:
    def __enter__(self):
        return self

    def __exit__(self, type, value, trace):
        print "type:", type
        print "value:", value
        print "trace:", trace

    def do_something(self):
        bar = 1/0
        return bar + 10

with Sample() as sample:

Notice how in this example, instead of get_sample(), with takes Sample(). It does not matter, as long as the statement that follows with evaluates to an object that has an __enter__() and __exit__() functions. In this case, Sample()’s __enter__() returns the newly created instance of Sample and that is what gets passed to sample.

When executed:

bash-3.2$ ./with_example02.py
type: <type 'exceptions.ZeroDivisionError'>
value: integer division or modulo by zero
trace: <traceback object at 0x1004a8128>
Traceback (most recent call last):
  File "./with_example02.py", line 19, in <module>
  File "./with_example02.py", line 15, in do_something
    bar = 1/0
ZeroDivisionError: integer division or modulo by zero

Essentially, if there are exceptions being thrown from anywhere inside the block, the __exit__() function for the object is called. As you can see, the type, value and the stack trace associated with the exception thrown is passed to this function. In this case, you can see that there was a ZeroDivisionError exception being thrown. People implementing libraries can write code that clean up resources, close files etc. in their __exit__() functions.

Thus, Python’s with is a nifty construct that makes code a little less verbose and makes cleaning up during exceptions a bit easier.

I have put the code examples given here on Github.