Topics: Mocking in Python

Today, I want to talk about mocking. I became interested in this topic a few months back when I started to work in a data engineering project at work. In this project we have a lot of tests that relies on mocking to test code with external dependencies. If you don’t like reading long blog posts, consider listening to one of the podcast episodes I did on this topic: there is one at Talk Python to Me and the other at Test and Code.

Introduction

Before getting started you should note that mocking is a controversial topic. Some people think it is great while others regard it only as a last resort that should be be avoided. Personally, I believe that mocking can be a great tool to improve testing but should be used with care. Most importantly, it should not be used as a tool to fix badly written code. However, the Zen of Python tells us that practicality beats purity. So rather use mocking than no testing at all.

What is mocking?

A mock simulates the existence and behavior of a real object. This allows developers to

  • Improve the quality of their tests
  • Test code in a controlled environment
  • Test code which has external dependencies

The python library for mocking is unittest.mock. If you prefer using pytest, don’t worry - pytest supports the unittest.mock library. You can also take a look at pytest libraries for mocking like pytest-mock.

Example class

In the spirit of our magical universe The Tales of Castle Kilmere we will work with an example class Spell in this blog post. Our Spell class is simple: each spell has a name, an incantation and a defining feature:

class Spell:
   """Creates a spell"""
   def __init__(self, name: str, incantation: str, defining_feature: str):
      self.name = name
      self.incantation = incantation
      self.defining_feature = defining_feature

Naming conventions and Test Doubles

Similar to the question of whether mocking is good or bad there is disagreement about the terminology of mocking. Some people prefer a more fine-grained terminology in which different types of ‘mocking behavior’ are distinguished. Others believe that this makes things unnecessarily complicated. For completeness, let’s take a look at the more fine-grained view. In this view, objects which are not real can be either dummies, fakes, stubs, mocks or spies. All of these are so called test-doubles. Note that the definitions of the different kinds of test doubles are not set in stone but are controversial and vary between sources.

Dummies

A dummy is an object which is passed around but not intended to be used in your tests. It will have no effect on the behaviour of your tests. An example for a dummy could be an attribute that is needed in order to instantiate a class but not required for the test. Looking at our Spell class we might want to test casting a spell but won’t require the defining_feature attribute for this:

class Spell:
   """Creates a spell"""
   def __init__(self, name: str, incantation: str, defining_feature: str):
      self.name = name
      self.incantation = incantation
      self.defining_feature = defining_feature

   def cast(self):
      print(f"{self.incantation}!")

def test_cast_spell():
   name = "The Stuporus Ratiato spell"
   incantation = "Stuporus Ratiato"
   defining_feature = "" # this is a dummy

   assert spell.cast == "Stuporus Ratiatio!"

Fakes

A fake implements a fake version of a class or method. It has a working implementation but takes some kind of shortcut such that it is not suitable for production. For example, we could implement an in memory database which is used for testing but not in production. In this setting, the in memory database would function as a fake.

Stubs

A stub is a non-real object which has pre-programmed behavior. Most of the time, stubs simply return fixed values (also refered to as canned data). For example, let’s say the Spell class has a method get_similar_spells which searches a database for similar spells. The logic behind this function is very complex and the function takes several minutes to complete. During testing, we don’t want to wait several minutes for a result. Therefore, we could replace the real implementation with a stub that returns hard-coded values; taking only a small fraction of the time.

Mocks

Mocks are closely related to stubs. A mock does not have predetermined behavior. Instead, it has to be configured with our expectations. The important difference between mocks and stubs is that a mock records which calls have been executed. Therefore, it can be used to verify not only the result (which can be done with a stub, too) but HOW the result was achieved / that the correct methods have been invoked. Mocks are the type of test double we will be talking about in this blog post.

If you are confused about the difference between stubs and mocks, that a look at this stackoverflow post. It might be helpful.

Spies

Spies are used to wrap real objects and, by default, route all method calls to the original object. In this sense, they intercept and record all calls to a real object. This allows us to verify calls to the original object (e.g. how often a certain method has been called) without replacing the original object (as, for example, a mock does). I haven’t use spies myself yet and find them the hardest to understand.

When should we mock?

Mocking can be used whenever we don’t want to actually call an object. Let’s say our Spell class has a function save which writes file to disk and function remove which deletes local file:

class Spell:
   """Creates a spell"""
   def __init__(self, name: str, incantation: str, defining_feature: str):
      self.name = name
      self.incantation = incantation
      self.defining_feature = defining_feature

   def save(self, save_path: str) -> dict:
      spell_as_dict = {"name": self.name,
	               "incantation": self.incantation,
	               "description": self.description}
      with open(save_path, "w") as file:
         json.dump(spell_as_dict, file, indent=4)

      return spell_as_dict

   def remove(self, file_path: str):
      os.remove(file_path)

When testing these methods we don’t want to write to disk every time the test runs Same holds for function ‘remove’

How to mock

The unittest.mock library has three core functionalities:

  1. The Mock class
  2. The MagicMock class
  3. The patch method

The unittest.mock.Mock class

The Mock class can be used for mocking any object. A mock simulates the object it replaces. To achieve this, it creates attributes on the fly. In other words: you can access whatever methods and attributes you like, the mock object will simply create them. This is quite magical, don’t you think?

from unittest.mock import Mock
my_mock = Mock()

# Let's try to access some attribute
my_mock.fancy_attribute
>>> <Mock name='mock.fancy_attribute' id='140586483377056'>

# What about a method with inputs?
my_mock.fancy_method(3, 4)
>>> <Mock name='mock.fancy_method()' id='140586482172880'>

# We can assert that certain methods can be called
my_mock.fancy_method.assert_called_with(3, 4)
>>> True

When taking a close look at the example above we can see that each call returns a mock object with a different ID. This is an important feature of mocks: they will automatically create child mocks when you access a property or method. You should keep in mind that the methods and attributes you create have nothing to do with the ’true’ implementation of a method. For example, we could mock the json library and access the method dumps on the mock object. We could call dumps with all kinds of attributes (or non at all) without the mock complaining about it:

json = Mock()
json.dumps()
>>> <Mock name='mock.dumps()' id='140586482201840'> 
json.dumps.assert_called_once()
>>> True

We can configue a mock object to do exactly what we want it to do. For instance, we can set the return value of a mock, or side effects it should exhibit:

my_mock = Mock()
my_mock.return_value = 3
my_mock()
>>> 3

my_mock = Mock(side_effect=Exception("Python rocks!"))
my_mock()
---------------------------------------------------------------------------
Exception                                 Traceback (most recent call last)
<ipython-input-14-faa07e9a9283> in <module>
----> 1 my_mock()

~/anaconda3/lib/python3.8/unittest/mock.py in __call__(self, *args, **kwargs)
   1079         self._mock_check_sig(*args, **kwargs)
   1080         self._increment_mock_call(*args, **kwargs)
-> 1081         return self._mock_call(*args, **kwargs)
   1082 
   1083 

~/anaconda3/lib/python3.8/unittest/mock.py in _mock_call(self, *args, **kwargs)
   1083 
   1084     def _mock_call(self, /, *args, **kwargs):
-> 1085         return self._execute_mock_call(*args, **kwargs)
   1086 
   1087     def _increment_mock_call(self, /, *args, **kwargs):

~/anaconda3/lib/python3.8/unittest/mock.py in _execute_mock_call(self, *args, **kwargs)
   1138         if effect is not None:
   1139             if _is_exception(effect):
-> 1140                 raise effect
   1141             elif not _callable(effect):
   1142                 result = next(effect)

Exception: Python rocks!

Mock vs. MagicMock

I mentioned that the unittest.mock library contains two classes: Mock and MagicMock. So what is the difference between them? MagicMock is a subclass of Mock. It contains all magic methods pre-created and ready to use (e.g. __str__, __len__, etc.). Therefore, you should use MagicMock when you need magic methods, and Mock if you don’t need them.

The unittest.mock.patch method

The third main functionality contained in the unittest.mock library is the patch function. So what is patching? The patch() function looks up an object in a given module and replaces it with another object. By default, the object is replaced with a new MagicMock (but this can be changed). When you want to use patch you have to use the syntax patch(‘package.module.target’). For example, consider we have a file example.py which contains the following code:

from db import db_write
def foo():
   x = db_write()
   return x

We want to mock the call to the db_write function in our tests. With patching, this can be dones as follows:

from unittest.mock import patch
from example import foo

@patch('example.db_write')
def test_foo(mock_write):
   mock_write.return_value = 10
   x = foo()
   assert x == 10

What is happening when using patching like this? In simple terms, the call to db_write is replaced with a call to MagicMock:

def foo():
   x = MagicMock()
   return x

By replacing the call to db_write with a call to MagicMock, everything happening in db_write is not being executed. The only thing that is called is the MagicMock object!

In the example above we used patch() as a decorator. We could also use it as a context manager or with manual starting/stopping. When to use which version depends on the scope you want/need. Consider the decorator example:

from unittest.mock import patch
from example import foo

@patch('example.db_write')
def test_foo_decorator(mock_write):
   mock_write.return_value = 10
   x = foo()
   assert x == 10

In this example, the db_write function will be replaced by a MagicMock for the entire test function. In other words: in the test function test_foo_decorator you don’t have access to the true implementation of db_write anymore, only to the mocked version. A more fine-grained scheme can be achieved using a context manager:

def test_foo_context_manager(mock_write):
   with patch('example.db_write`, return_value = 10):
      x1 = foo()
      assert x1 == 10

   x2 = foo() # When calling foo here, db_write won't be mocked

Manual starting and stopping can be useful when you need to mock certain parts of your code for all tests in your test file. I haven’t yet used this functionality myself but you can find a good description here.

Note: there is also patch.object and patch.dict. We won’t discuss them here to avoid confusion, please check out the documentation to find out how they are used.

Where do we patch?

When using patching it is important to know where patch is applied. Specifically, we should patch where the object is looked up. This might not be the place where the object is defined! In the example above (the one where our file is named example.py) we import the method db_write from the module db:

from db import db_write
def foo():
   x = db_write()
   return x

Consequently, the example module knows about db_write. However, it does not know about db. This is nicely illustrated when importing the example module (e.g. in an IPython shell) and calling dir(example) to see a list of attributes for the example object.

import example
dir(example)
>>>
['__builtins__',
 '__cached__',
 '__doc__',
 '__file__',
 '__loader__',
 '__name__',
 '__package__',
 '__spec__',
 'db_write',
 'foo']

As we can see here, db_write is in the namespace of example, but db is not in the namespace. Therefore, to mock the call to db_write in the foo function, we have to patch example.db_write:

from unittest.mock import patch
from example import foo

@patch('example.db_write')
def test_foo_decorator(mock_write):
   mock_write.return_value = 10
   x = foo()
   assert x == 10

Think about this for a moment - db_write is defined in the module db but used in the module example.

Let’s drive this point home with another example. Let’s say that we import the entire db module in example.py:

import db
def foo():
   x = db.db_write()
   return x

In this case, the example module knows about db but not db_write:

import example
dir(example)
>>>
['__builtins__',
 '__cached__',
 '__doc__',
 '__file__',
 '__loader__',
 '__name__',
 '__package__',
 '__spec__',
 'db',
 'foo']

To mock the call to db_write in the foo function, we need to mock db.db_write or, if you find it easier to understand example.db.db_write:

from unittest.mock import patch
from example import foo

@patch('example.db.db_write')
def test_foo_decorator(mock_write):
   mock_write.return_value = 10
   x = foo()
   assert x == 10

Common mocking problems and how to solve them

In the beginning of this post we saw that we can configure Mock objects. For example, we can set their return value or side effects like exceptions. The fact that Mock objects create attributes and methods on the fly makes them sensitive to mistakes. For example, if you call .asert_called_once() instead of .assert_called_once() on a Mock, the test will not raise an AssertionError because you have created a new method on the Mock object called asert_called_once() instead of calling the actual function.

Another easy mistake occurs when forgetting that a Mock object does not know the interface of your class / method. In other words: the Mock object does not know which attributes it can be called with. We talked about that earlier with the json.dumps() example. We can illustrate this problem when taking a closer look at the attributes of the thing we are mocking. Let’s say we we mock our Spell class. When calling dir() on the mocked object we see all kinds of attributes:

import spell_class
from unittest.mock import patch

with patch('spell_class.Spell') as mocked_spell:
   print(dir(mocked_spell))
>>>
['assert_any_call', 'assert_called', 'assert_called_once', 'assert_called_once_with', 'assert_called_with', 'assert_has_calls', 'assert_not_called', 'attach_mock', 'call_args', 'call_args_list', 'call_count', 'called', 'configure_mock', 'method_calls', 'mock_add_spec', 'mock_calls', 'reset_mock', 'return_value', 'side_effect']

For example, we can see the different assertions that can be made on the Mock object. Hover, we don’t see the actual methods of the Spell class, e.g. the save or remove method! This is because the call to patch returns a MagicMock which is completely unrelated to the Spell class.

Problems like these can be solved using the spec=True attribute in the patch call. This will cause the MagicMock to look like the object we are patching. We could no longer call methods without argument or with the wrong arguments. Also, the misspelled .asert_called_once() would raise an Exception:

import spell_class
from unittest.mock import patch

with patch('spell_class.Spell', spec=True) as mocked_spell:
   print(dir(mocked_spell))
>>> ['__class__', '__delattr__', '__dict__', '__dir__', '__doc__', '__eq__', '__format__', '__ge__', '__getattribute__', '__gt__', '__hash__', '__init__', '__init_subclass__', '__le__', '__lt__', '__module__', '__ne__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', '__weakref__', 'assert_any_call', 'assert_called', 'assert_called_once', 'assert_called_once_with', 'assert_called_with', 'assert_has_calls', 'assert_not_called', 'attach_mock', 'call_args', 'call_args_list', 'call_count', 'called', 'configure_mock', 'method_calls', 'mock_add_spec', 'mock_calls', 'remove', 'reset_mock', 'return_value', 'save', 'side_effect']

with patch('spell_class.Spell', spec=True) as mocked_spell:
   mocked_spell.asert_called_once()
>>>
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-3-d3058c55a4ad> in <module>
      3 
      4 with patch('spell_class.Spell', spec=True) as mocked_spell:
----> 5    mocked_spell.asert_called_once()
      6 

~/anaconda3/lib/python3.8/unittest/mock.py in __getattr__(self, name)
    635         elif self._mock_methods is not None:
    636             if name not in self._mock_methods or name in _all_magics:
--> 637                 raise AttributeError("Mock object has no attribute %r" % name)
    638         elif _is_magic(name):
    639             raise AttributeError(name)

AttributeError: Mock object has no attribute 'asert_called_once'

Similar to spec are autospec and specset. Take a look at the documentation or at the PyCon talk “Demystifying the patch function” for more details.

One caveat when using spec (or autospec or specset) is that you can not access attributes created in the __init__ method, that is, attributes that exist on the instances and not the class. Due to the internal workings of spec (and also autospec and specset) they cannot know about any dynamically created attributes. They only know about visible attributes. For our Spell example above, you won’t see the attributes name, incantation and description in the output of dir(mocked_spell). They are not valid attributes of the mock object anymore. Also, speccing comes at a cost and will slow down your tests

Code design that prevents mocking

When you stumble into mock hell (a term introduced and explained in the talk “Mocking and Patching Pitfalls”) during development, it might be the right time to consider one of the design principles that prevent the need for mocking. For example, you could use dependency injection or an adaptor pattern. These principles are explained in the PyCon talk “Stop using Mocks - For a While”.

Wrap-Up

Mocking is a controversial topic. It should be used with care and never to fix badly written code. The Python library for mocking - unittest.mock - comes with three core functionalities: Mock, MagicMock and patch. Mocking might seem confusing in the beginning but once you understand the basics (for example where to patch) it can be very helpful, especially with production code. There are lots of functionalities in the unittest.mock library that help prevent common problems, so it is useful to take a look at the documentation when you want to start using mocking. Lastly, don’t get confused by the different naming conventions! If you feel more comfortable with the fine-grained view on test doubles great! But sticking to a simpler scheme is also fine.