Subsection 20.4.1 Introduction
Now that you know how to write unit tests using the assert statement, it’s important for you to understand how to write testable functions. Not all functions can be tested.
Consider the following function:
def add(x, y):
"""Adds two numbers and displays the sum"""
print(x + y)
How would you write an assert statement to check that this function works? Think about it a moment. Would this work?
assert add(2, 3) == 5
Answer: no. An assert statement cannot verify that what a function displays on the screen is correct. It can only check that the contents of variables are correct. This function is not testable.
A testable function is a function that produces a result that can be checked by an assert statement. Generally, it does so in one of three ways:
It returns its result
It stores its result in a global variable
It modifies the state of an object passed as a parameter
Functions that display their output using print are not testable functions.
Check your understanding
Checkpoint 20.4.1.
sum = 0
def add(x, y):
global sum
sum = x + y
Yes.
Correct. In a unit test, an assert statement could check the value of sum after a call to add.
No.
Incorrect. The function stores its result in a global variable, which can be checked by an assert statement.
Subsection 20.4.2 Design by Contract
In addition to producing a result that can be checked by an assert statement, a testable function must have a clear specification. In order to write unit tests for a function, you must have a precise understanding of what the function should do.
A function specification describes what value the function produces, given its parameter values, and is generally expressed in the form of a docstring. For example, consider the sumnums function given earlier in this chapter:
def sumnums(lo, hi):
"""returns the sum of the numbers in the range [lo..hi]
Precondition: lo <= hi
"""
...
The docstring is this function’s specification. Given this specification, you might write a unit test that contains the following assert:
assert sumnums(1, 3) == 6
An alternate way to write the docstring is as follows:
def sumnums(lo, hi):
"""computes the sum of a range of numbers
Precondition: lo <= hi
Postcondition: returns the sum of the numbers in the range [lo..hi]
"""
...
This docstring contains three elements: a brief description; a precondition; and a postcondition. We’ve discussed the concept of a precondition earlier in this chapter. The postcondition is new.
Functions that include a precondition and a postcondition in their docstring embody a software engineering idea called design by contract. The idea is that a function specification forms a contract between the function and the code calling the function. If the code calling the function passes parameters that satisfy the function’s precondition, then the function should be expected to produce what it says it will produce. If the parameters do not satisfy the function’s precondition, then the function does not have to produce a valid result. In the design by contract approach, a testable function is one where the function’s postcondition can be verified by an assert statement.
In this example, you can think of the function’s docstring as promising to calling code: “If you give me two parameters, lo and hi, such that lo is less than or equal to hi, I promise to return the sum of the numbers in the range lo..hi, inclusive.”
To write a precondition, think about the parameter values that the function is designed to handle, and write a boolean expression that expresses what parameter values are valid. For example, consider a function that computes the average weight, given a total weight and a number of items:
def compute_average(total_weight: float, num_items: float) -> float:
return total / num_items
This function will work if num_items is greater than zero, but will fail if num_items is zero. So, an appropriate precondition would be num_items > 0. A complete docstring would look like this:
def compute_average(total_weight: float, num_items: float) -> float:
"""computes the average weight, given `total_weight` of items and `num_items`
Precondition: num_items > 0
Postcondition: returns average item weight
"""
Sometimes, your precondition will be expressed more loosely, using English. Consider this function which extracts the first word from a string containing text:
def get_first_word(text: str) -> str:
"""extracts the first word from `text`"""
space_loc = text.find(' ')
return text[0:space_loc]
This function will produce nonsense if the string doesn’t contain a space. So, an appropriate precondition might be “text contains 2 or more words separated by spaces”. The docstring might be:
def get_first_word(text: str) -> str:
"""extracts the first word from `text`
Precondition: `text` contains 2 or more words separated by spaces
Postcondition: returns the first word in `text`
"""
Following the design by contract idea and writing function specifications that include preconditions and postconditions is an excellent way to design testable functions, because, as we’ll see in the next section, it makes it possible to reason precisely about what the function should do when given various parameter values. Even if you don’t use precondition and postcondition terminology in your docstrings, it helps to think in those terms.
Check your understanding
Checkpoint 20.4.3.
Consider the following function. What would an appropriate precondition be?
def getfirst(msg):
"""returns first character of msg"""
return msg[0]
len(msg) <= 0
Incorrect. The function would crash if this condition were true.
len(msg) > 0
Correct. The function cannot return the first character of an empty msg.
msg == ""
Incorrect. The function would crash if this condition were true.
none of these
Incorrect. Try again.
