Before we can look closely at creating classes, we need to talk a little bit about what a class is and how we're sure we're using it correctly. The central idea here is that everything in Python is an object.

When we write literal values like "Hello, world!" or 42, we're actually creating instances of built-in classes. We can fire up interactive Python and use the built-in type() function on the class that defines the properties of these objects:

>>> type("Hello, world!")
<class 'str'>
>>> type(42)
<class 'int'>

The point of object-oriented programming is to solve a problem via the interactions of objects. When we write 6*7, the multiplication of the two objects is handled by a method of the built-in int class. For more complex behaviors, we'll often need to write unique, new classes.

Here are the first two core rules of how Python objects work:

• Everything in Python is an object
• Every object is defined by being an instance of at least one class

These rules have many interesting consequences. A class definition we write, using the class statement, creates a new object of class type. When we create an instance of a class, the class object will be used to create and initialize the instance object.

What's the distinction between class and type? The class statement lets us define new types. Because the class statement is what we use, we'll call them classes throughout the text. See Python objects, types, classes, and instances - a glossary by Eli Bendersky: https://eli.thegreenplace.net/2012/03/30/python-objects-types-classes-and-instances-a-glossary for this useful quote:

"The terms "class" and "type" are an example of two names referring to the same concept."

We'll follow common usage and call the annotations type hints.

There's another important rule:

• A variable is a reference to an object. Think of a yellow sticky note with a name scrawled on it, slapped on a thing.

This doesn't seem too earth-shattering but it's actually pretty cool. It means the type information – what an object is – is defined by the class(es) associated with the object. This type information is not attached to the variable in any way. This leads to code like the following being valid but very confusing Python:

>>> a_string_variable = "Hello, world!"
>>> type(a_string_variable)
<class 'str'>
>>> a_string_variable = 42
>>> type(a_string_variable)
<class 'int'>

We created an object using a built-in class, str. We assigned a long name, a_string_variable, to the object. Then, we created an object using a different built-in class, int. We assigned this object the same name. (The previous string object has no more references and ceases to exist.)

Here are the two steps, shown side by side, showing how the variable is moved from object to object:

Figure 2.1: Variable names and objects

The various properties are part of the object, not the variable. When we check the type of a variable with type(), we see the type of the object the variable currently references. The variable doesn't have a type of its own; it's nothing more than a name. Similarly, asking for the id() of a variable shows the ID of the object the variable refers to. So obviously, the name a_string_variable is a bit misleading if we assign the name to an integer object.

Type checking

Let's push the relationship between object and type a step further, and look at some more consequences of these rules. Here's a function definition:

>>> def odd(n):
...     return n % 2 != 0
>>> odd(3)
True
>>> odd(4)
False

This function does a little computation on a parameter variable, n. It computes the remainder after division, the modulo. If we divide an odd number by two, we'll have one left over. If we divide an even number by two, we'll have zero left over. This function returns a true value for all odd numbers.

What happens when we fail to provide a number? Well, let's just try it and see (a common way to learn Python!). Entering code at the interactive prompt, we'll get something like this:

>>> odd("Hello, world!")
Traceback (most recent call last):
  File "<doctestexamples.md[9]>", line 1, in <module>
odd("Hello, world!")
  File "<doctestexamples.md[6]>", line 2, in odd
    return n % 2 != 0
TypeError: not all arguments converted during string formatting

This is an important consequence of Python's super-flexible rules: nothing prevents us from doing something silly that may raise an exception. This is an important tip:

In our example, the % operator provided by the str class doesn't work the same way as the % operator provided by the int class, raising an exception. For strings, the % operator isn't used very often, but it does interpolation: "a=%d" % 113 computes a string 'a=113'; if there's no format specification like %d on the left side, the exception is a TypeError. For integers, it's the remainder in division: 355 % 113 returns an integer, 16.

This flexibility reflects an explicit trade-off favoring ease of use over sophisticated prevention of potential problems. This allows a person to use a variable name with little mental overhead.

Python's internal operators check that operands meet the requirements of the operator. The function definition we wrote, however, does not include any runtime type checking. Nor do we want to add code for runtime type checking. Instead, we use tools to examine code as part of testing. We can provide annotations, called type hints, and use tools to examine our code for consistency among the type hints.

First, we'll look at the annotations. In a few contexts, we can follow a variable name with a colon, :, and a type name. We can do this in the parameters to functions (and methods). We can also do this in assignment statements. Further, we can also add -> syntax to a function (or a class method) definition to explain the expected return type.

Here's how type hints look:

>>> def odd(n: int) -> bool:
...     return n % 2 != 0

We've added two type hints to our odd() little function definition. We've specified that argument values for the n parameter should be integers. We've also specified that the result will be one of the two values of the Boolean type.

While the hints consume some storage, they have no runtime impact. Python politely ignores these hints; this means they're optional. People reading your code, however, will be more than delighted to see them. They are a great way to inform the reader of your intent. You can omit them while you're learning, but you'll love them when you go back to expand something you wrote earlier.

The mypy tool is commonly used to check the hints for consistency. It's not built into Python, and requires a separate download and install. We'll talk about virtual environments and installation of tools later in this chapter, in the Third-party libraries section. For now, you can use python -m pip install mypy or conda install mypy if you're using the conda tool.

Let's say we had a file, bad_hints.py, in a src directory, with these two functions and a few lines to call the main() function:

def odd(n: int) -> bool:
    return n % 2 != 0
def main():
    print(odd("Hello, world!"))
if __name__ == "__main__":
    main()

When we run the mypy command at the OS's terminal prompt:

% mypy –strict src/bad_hints.py

The mypy tool is going to spot a bunch of potential problems, including at least these:

src/bad_hints.py:12: error: Function is missing a return type annotation
src/bad_hints.py:12: note: Use "-> None" if function does not return a value
src/bad_hints.py:13: error: Argument 1 to "odd" has incompatible type "str"; expected "int"

The def main(): statement is on line 12 of our example because our file has a pile of comments not shown above. For your version, the error might be on line 1. Here are the two problems:

• The main() function doesn't have a return type; mypy suggests including -> None to make the absence of a return value perfectly explicit.
• More important is line 13: the code will try to evaluate the odd() function using a str value. This doesn't match the type hint for odd() and indicates another possible error.

Most of the examples in this book will have type hints. We think they're always helpful, especially in a pedagogical context, even though they're optional. Because most of Python is generic with respect to type, there are a few cases where Python behavior is difficult to describe via a succinct, expressive hint. We'll steer clear of these edge cases in this book.

Python Enhancement Proposal (PEP) 585 covers some new language features to make type hints a bit simpler. We've used mypy version 0.812 to test all of the examples in this book. Any older version will encounter problems with some of the newer syntax and annotation techniques.

Now that we've talked about how parameters and attributes are described with type hints, let's actually build some classes.

We don't have to write much Python code to realize that Python is a very clean language. When we want to do something, we can just do it, without having to set up a bunch of prerequisite code. The ubiquitous hello world in Python, as you've likely seen, is only one line.

Similarly, the simplest class in Python 3 looks like this:

class MyFirstClass: 
    pass 

There's our first object-oriented program! The class definition starts with the class keyword. This is followed by a name (of our choice) identifying the class and is terminated with a colon.

The class definition line is followed by the class contents, indented. As with other Python constructs, indentation is used to delimit the classes, rather than braces, keywords, or brackets, as many other languages use. Also, in line with the style guide, use four spaces for indentation unless you have a compelling reason not to (such as fitting in with somebody else's code that uses tabs for indents).

Since our first class doesn't actually add any data or behaviors, we simply use the pass keyword on the second line as a placeholder to indicate that no further action needs to be taken.

We might think there isn't much we can do with this most basic class, but it does allow us to instantiate objects of that class. We can load the class into the Python 3 interpreter, so we can interactively play with it. To do this, save the class definition mentioned earlier in a file named first_class.py and then run the python -i first_class.py command. The -i argument tells Python to run the code and then drop to the interactive interpreter. The following interpreter session demonstrates a basic interaction with this class:

>>> a = MyFirstClass()
>>> b = MyFirstClass()
>>> print(a)
<__main__.MyFirstClass object at 0xb7b7faec>
>>> print(b)
<__main__.MyFirstClass object at 0xb7b7fbac>

This code instantiates two objects from the new class, assigning the object variable names a and b. Creating an instance of a class is a matter of typing the class name, followed by a pair of parentheses. It looks much like a function call; calling a class will create a new object. When printed, the two objects tell us which class they are and what memory address they live at. Memory addresses aren't used much in Python code, but here, they demonstrate that there are two distinct objects involved.

We can see they're distinct objects by using the is operator:

>>> a is b
False

This can help reduce confusion when we've created a bunch of objects and assigned different variable names to the objects.

Adding attributes

Now, we have a basic class, but it's fairly useless. It doesn't contain any data, and it doesn't do anything. What do we have to do to assign an attribute to a given object?

In fact, we don't have to do anything special in the class definition to be able to add attributes. We can set arbitrary attributes on an instantiated object using dot notation. Here's an example:

class Point: 
    pass 
p1 = Point() 
p2 = Point() 
p1.x = 5 
p1.y = 4 
p2.x = 3 
p2.y = 6 
print(p1.x, p1.y) 
print(p2.x, p2.y) 

If we run this code, the two print statements at the end tell us the new attribute values on the two objects:

5 4
3 6

This code creates an empty Point class with no data or behaviors. Then, it creates two instances of that class and assigns each of those instances x and y coordinates to identify a point in two dimensions. All we need to do to assign a value to an attribute on an object is use the <object>.<attribute> = <value> syntax. This is sometimes referred to as dot notation. The value can be anything: a Python primitive, a built-in data type, or another object. It can even be a function or another class!

Creating attributes like this is confusing to the mypy tool. There's no easy way to include the hints in the Point class definition. We can include hints on the assignment statements, like this: p1.x: float = 5. In general, there's a much, much better approach to type hints and attributes that we'll examine in the Initializing the object section, later in this chapter. First, though, we'll add behaviors to our class definition.

Making it do something

Now, having objects with attributes is great, but object-oriented programming is really about the interaction between objects. We're interested in invoking actions that cause things to happen to those attributes. We have data; now it's time to add behaviors to our classes.

Let's model a couple of actions on our Point class. We can start with a method called reset, which moves the point to the origin (the origin is the place where x and y are both zero). This is a good introductory action because it doesn't require any parameters:

class Point: 
    def reset(self): 
        self.x = 0 
        self.y = 0 
p = Point() 
p.reset() 
print(p.x, p.y) 

This print statement shows us the two zeros on the attributes:

0 0

In Python, a method is formatted identically to a function. It starts with the def keyword, followed by a space, and the name of the method. This is followed by a set of parentheses containing the parameter list (we'll discuss that self parameter, sometimes called the instance variable, in just a moment), and terminated with a colon. The next line is indented to contain the statements inside the method. These statements can be arbitrary Python code operating on the object itself and any parameters passed in, as the method sees fit.

We've omitted type hints in the reset() method because it's not the most widely used place for hints. We'll look at the best place for hints in the Initializing the object section. We'll look a little more at these instance variables, first, and how the self variable works.

Talking to yourself

The one difference, syntactically, between methods of classes and functions outside classes is that methods have one required argument. This argument is conventionally named self; I've never seen a Python programmer use any other name for this variable (convention is a very powerful thing). There's nothing technically stopping you, however, from calling it this or even Martha, but it's best to acknowledge the social pressure of the Python community codified in PEP 8 and stick with self.

The self argument to a method is a reference to the object that the method is being invoked on. The object is an instance of a class, and this is sometimes called the instance variable.

We can access attributes and methods of that object via this variable. This is exactly what we do inside the reset method when we set the x and y attributes of the self object.

Notice that when we call the p.reset() method, we do not explicitly pass the self argument into it. Python automatically takes care of this part for us. It knows we're calling a method on the p object, so it automatically passes that object, p, to the method of the class, Point.

For some, it can help to think of a method as a function that happens to be part of a class. Instead of calling the method on the object, we could invoke the function as defined in the class, explicitly passing our object as the self argument:

>>> p = Point() 
>>> Point.reset(p) 
>>> print(p.x, p.y)

The output is the same as in the previous example because, internally, the exact same process has occurred. This is not really a good programming practice, but it can help to cement your understanding of the self argument.

What happens if we forget to include the self argument in our class definition? Python will bail with an error message, as follows:

>>> class Point:
...     def reset():
...         pass
...
>>> p = Point()
>>> p.reset()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: reset() takes 0 positional arguments but 1 was given

The error message is not as clear as it could be ("Hey, silly, you forgot to define the method with a self parameter" could be more informative). Just remember that when you see an error message that indicates missing arguments, the first thing to check is whether you forgot the self parameter in the method definition.

More arguments

How do we pass multiple arguments to a method? Let's add a new method that allows us to move a point to an arbitrary position, not just to the origin. We can also include a method that accepts another Point object as input and returns the distance between them:

import math
class Point:
    def move(self, x: float, y: float) -> None:
        self.x = x
        self.y = y
    def reset(self) -> None:
        self.move(0, 0)
    def calculate_distance(self, other: "Point") -> float:
        return math.hypot(self.x - other.x, self.y - other.y)

We've defined a class with two attributes, x, and y, and three separate methods, move(), reset(), and calculate_distance().

The move() method accepts two arguments, x and y, and sets the values on the self object. The reset() method calls the move() method, since a reset is just a move to a specific known location.

The calculate_distance() method computes the Euclidean distance between two points. (There are a number of other ways to look at distance. In the Chapter 3, When Objects Are Alike, case study, we'll look at some alternatives.) For now, we hope you understand the math. The definition is , which is the math.hypot() function. In Python we'll use self.x, but mathematicians often prefer to write .

Here's an example of using this class definition. This shows how to call a method with arguments: include the arguments inside the parentheses and use the same dot notation to access the method name within the instance. We just picked some random positions to test the methods. The test code calls each method and prints the results on the console:

>>> point1 = Point()
>>> point2 = Point()
>>> point1.reset()
>>> point2.move(5, 0)
>>> print(point2.calculate_distance(point1))
5.0
>>> assert point2.calculate_distance(point1) == point1.calculate_distance(
...    point2
... )
>>> point1.move(3, 4)
>>> print(point1.calculate_distance(point2))
4.47213595499958
>>> print(point1.calculate_distance(point1))
0.0

The assert statement is a marvelous test tool; the program will bail if the expression after assert evaluates to False (or zero, empty, or None). In this case, we use it to ensure that the distance is the same regardless of which point called the other point's calculate_distance() method. We'll see a lot more use of assert in Chapter 13, Testing Object-Oriented Programs, where we'll write more rigorous tests.

Initializing the object

If we don't explicitly set the x and y positions on our Point object, either using move or by accessing them directly, we'll have a broken Point object with no real position. What will happen when we try to access it?

Well, let's just try it and see. Try it and see is an extremely useful tool for Python study. Open up your interactive interpreter and type away. (Using the interactive prompt is, after all, one of the tools we used to write this book.)

The following interactive session shows what happens if we try to access a missing attribute. If you saved the previous example as a file or are using the examples distributed with the book, you can load it into the Python interpreter with the python -i more_arguments.py command:

>>> point = Point()
>>> point.x = 5
>>> print(point.x)
5
>>> print(point.y)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: 'Point' object has no attribute 'y'

Well, at least it threw a useful exception. We'll cover exceptions in detail in Chapter 4, Expecting the Unexpected. You've probably seen them before (especially the ubiquitous SyntaxError, which means you typed something incorrectly!). At this point, simply be aware that it means something went wrong.

The output is useful for debugging. In the interactive interpreter, it tells us the error occurred at line 1, which is only partially true (in an interactive session, only one statement is executed at a time). If we were running a script in a file, it would tell us the exact line number, making it easy to find the offending code. In addition, it tells us that the error is an AttributeError, and gives a helpful message telling us what that error means.

We can catch and recover from this error, but in this case, it feels like we should have specified some sort of default value. Perhaps every new object should be reset() by default, or maybe it would be nice if we could force the user to tell us what those positions should be when they create the object.

Interestingly, mypy can't determine whether y is supposed to be an attribute of a Point object. Attributes are – by definition – dynamic, so there's no simple list that's part of a class definition. However, Python has some widely followed conventions that can help name the expected set of attributes.

Most object-oriented programming languages have the concept of a constructor, a special method that creates and initializes the object when it is created. Python is a little different; it has a constructor and an initializer. The constructor method, __new__(), is rarely used unless you're doing something very exotic. So, we'll start our discussion with the much more common initialization method, __init__().

The Python initialization method is the same as any other method, except it has a special name, __init__. The leading and trailing double underscores mean this is a special method that the Python interpreter will treat as a special case.

Let's add an initialization function on our Point class that requires the user to supply x and y coordinates when the Point object is instantiated:

class Point:
    def __init__(self, x: float, y: float) -> None:
        self.move(x, y)
    def move(self, x: float, y: float) -> None:
        self.x = x
        self.y = y
    def reset(self) -> None:
        self.move(0, 0)
    def calculate_distance(self, other: "Point") -> float:
        return math.hypot(self.x - other.x, self.y - other.y)

Constructing a Point instance now looks like this:

point = Point(3, 5) 
print(point.x, point.y) 

Now, our Point object can never go without both x and y coordinates! If we try to construct a Point instance without including the proper initialization parameters, it will fail with a not enough arguments error similar to the one we received earlier when we forgot the self argument in a method definition.

Most of the time, we put our initialization statements in an __init__() function. It's very important to be sure that all of the attributes are initialized in the __init__() method. Doing this helps the mypy tool by providing all of the attributes in one obvious place. It helps people reading your code, also; it saves them from having to read the whole application to find mysterious attributes set outside the class definition.

While they're optional, it's generally helpful to include type annotations on the method parameters and result values. After each parameter name, we've included the expected type of each value. At the end of the definition, we've included the two-character -> operator and the type returned by the method.

Type hints and defaults

As we've noted a few times now, hints are optional. They don't do anything at runtime. There are tools, however, that can examine the hints to check for consistency. The mypy tool is widely used to check type hints.

If we don't want to make the two arguments required, we can use the same syntax Python functions use to provide default arguments. The keyword argument syntax appends an equals sign after each variable name. If the calling object does not provide this argument, then the default argument is used instead. The variables will still be available to the function, but they will have the values specified in the argument list. Here's an example:

class Point:
    def __init__(self, x: float = 0, y: float = 0) -> None:
        self.move(x, y)

The definitions for the individual parameters can get long, leading to very long lines of code. In some examples, you'll see this single logical line of code expanded to multiple physical lines. This relies on the way Python combines physical lines to match ()'s. We might write this when the line gets long:

class Point:
    def __init__(
        self, 
        x: float = 0, 
        y: float = 0
    ) -> None:
        self.move(x, y)

This style isn't used very often, but it's valid and keeps the lines shorter and easier to read.

The type hints and defaults are handy, but there's even more we can do to provide a class that's easy to use and easy to extend when new requirements arise. We'll add documentation in the form of docstrings.

Explaining yourself with docstrings

Python can be an extremely easy-to-read programming language; some might say it is self-documenting. However, when carrying out object-oriented programming, it is important to write API documentation that clearly summarizes what each object and method does. Keeping documentation up to date is difficult; the best way to do it is to write it right into our code.

Python supports this through the use of docstrings. Each class, function, or method header can have a standard Python string as the first indented line inside the definition (the line that ends in a colon).

Docstrings are Python strings enclosed within apostrophes (') or quotation marks ("). Often, docstrings are quite long and span multiple lines (the style guide suggests that the line length should not exceed 80 characters), which can be formatted as multi-line strings, enclosed in matching triple apostrophe (''') or triple quote (""") characters.

A docstring should clearly and concisely summarize the purpose of the class or method it is describing. It should explain any parameters whose usage is not immediately obvious, and is also a good place to include short examples of how to use the API. Any caveats or problems an unsuspecting user of the API should be aware of should also be noted.

One of the best things to include in a docstring is a concrete example. Tools like doctest can locate and confirm these examples are correct. All the examples in this book are checked with the doctest tool.

To illustrate the use of docstrings, we will end this section with our completely documented Point class:

class Point:
    """
    Represents a point in two-dimensional geometric coordinates
    >>> p_0 = Point()
    >>> p_1 = Point(3, 4)
    >>> p_0.calculate_distance(p_1)
    5.0
    """
    def __init__(self, x: float = 0, y: float = 0) -> None:
        """
        Initialize the position of a new point. The x and y
        coordinates can be specified. If they are not, the
        point defaults to the origin.
        :param x: float x-coordinate
        :param y: float x-coordinate
        """
        self.move(x, y)
    def move(self, x: float, y: float) -> None:
        """
        Move the point to a new location in 2D space.
        :param x: float x-coordinate
        :param y: float x-coordinate
        """
        self.x = x
        self.y = y
    def reset(self) -> None:
        """
        Reset the point back to the geometric origin: 0, 0
        """
        self.move(0, 0)
    def calculate_distance(self, other: "Point") -> float:
        """
        Calculate the Euclidean distance from this point 
        to a second point passed as a parameter.
        :param other: Point instance
        :return: float distance
        """
        return math.hypot(self.x - other.x, self.y - other.y)

Try typing or loading (remember, it's python -i point.py) this file into the interactive interpreter. Then, enter help(Point)<enter> at the Python prompt.

You should see nicely formatted documentation for the class, as shown in the following output:

Help on class Point in module point_2:
class Point(builtins.object)
 |  Point(x: float = 0, y: float = 0) -> None
 |  
 |  Represents a point in two-dimensional geometric coordinates
 |  
 |  >>> p_0 = Point()
 |  >>> p_1 = Point(3, 4)
 |  >>> p_0.calculate_distance(p_1)
 |  5.0
 |  
 |  Methods defined here:
 |  
 |  __init__(self, x: float = 0, y: float = 0) -> None
 |      Initialize the position of a new point. The x and y
 |      coordinates can be specified. If they are not, the
 |      point defaults to the origin.
 |      
 |      :param x: float x-coordinate
 |      :param y: float x-coordinate
 |  
 |  calculate_distance(self, other: 'Point') -> float
 |      Calculate the Euclidean distance from this point
 |      to a second point passed as a parameter.
 |      
 |      :param other: Point instance
 |      :return: float distance
 |  
 |  move(self, x: float, y: float) -> None
 |      Move the point to a new location in 2D space.
 |      
 |      :param x: float x-coordinate
 |      :param y: float x-coordinate
 |  
 |  reset(self) -> None
 |      Reset the point back to the geometric origin: 0, 0
 |  
 |  ----------------------------------------------------------------
 |  Data descriptors defined here:
 |  
 |  __dict__
 |      dictionary for instance variables (if defined)
 |  
 |  __weakref__
 |      list of weak references to the object (if defined)

Not only is our documentation every bit as polished as the documentation for built-in functions, but we can run python -m doctest point_2.py to confirm the example shown in the docstring.

Further, we can run mypy to check the type hints, also. Use mypy –-strict src/*.py to check all of the files in the src folder. If there are no problems, the mypy application doesn't produce any output. (Remember, mypy is not part of the standard installation, so you'll need to add it. Check the preface for information on extra packages that need to be installed.)

Now we know how to create classes and instantiate objects. You don't need to write too many classes (or non-object-oriented code, for that matter) before you start to lose track of them. For small programs, we generally put all our classes into one file and add a little script at the end of the file to start them interacting. However, as our projects grow, it can become difficult to find the one class that needs to be edited among the many classes we've defined. This is where modules come in. Modules are Python files, nothing more. The single file in our small program is a module. Two Python files are two modules. If we have two files in the same folder, we can load a class from one module for use in the other module.

The Python module name is the file's stem; the name without the .py suffix. A file named model.py is a module named model. Module files are found by searching a path that includes the local directory and the installed packages.

The import statement is used for importing modules or specific classes or functions from modules. We've already seen an example of this in our Point class in the previous section. We used the import statement to get Python's built-in math module and use its hypot() function in the distance calculation. Let's start with a fresh example.

If we are building an e-commerce system, we will likely be storing a lot of data in a database. We can put all the classes and functions related to database access into a separate file (we'll call it something sensible: database.py). Then, our other modules (for example, customer models, product information, and inventory) can import classes from the database module in order to access the database.

Let's start with a module called database. It's a file, database.py, containing a class called Database. A second module called products is responsible for product-related queries. The classes in the products module need to instantiate the Database class from the database module so that they can execute queries on the product table in the database.

There are several variations on the import statement syntax that can be used to access the Database class. One variant is to import the module as a whole:

>>> import database
>>> db = database.Database("path/to/data")

This version imports the database module, creating a database namespace. Any class or function in the database module can be accessed using the database.<something> notation.

Alternatively, we can import just the one class we need using the from...import syntax:

>>> from database import Database
>>> db = Database("path/to/data")

This version imported only the Database class from the database module. When we have a few items from a few modules, this can be a helpful simplification to avoid using longer, fully qualified names like database.Database. When we import a number of items from a number of different modules, this can be a potential source of confusion when we omit the qualifiers.

If, for some reason, products already has a class called Database, and we don't want the two names to be confused, we can rename the class when used inside the products module:

>>> from database import Database as DB
>>> db = DB("path/to/data")

We can also import multiple items in one statement. If our database module also contains a Query class, we can import both classes using the following code:

from database import Database, Query

We can import all classes and functions from the database module using this syntax:

from database import * 

We've got several reasons for avoiding this:

• When we explicitly import the database class at the top of our file using from database import Database, we can easily see where the Database class comes from. We might use db = Database() 400 lines later in the file, and we can quickly look at the imports to see where that Database class came from. Then, if we need clarification as to how to use the Database class, we can visit the original file (or import the module in the interactive interpreter and use the help(database.Database) command). However, if we use the from database import * syntax, it takes a lot longer to find where that class is located. Code maintenance becomes a nightmare.
• If there are conflicting names, we're doomed. Let's say we have two modules, both of which provide a class named Database. Using from module_1 import * and from module_2 import * means the second import statement overwrites the Database name created by the first import. If we used import module_1 and import module_2, we'd use the module names as qualifiers to disambiguate module_1.Database from module_2.Database.
• In addition, most code editors are able to provide extra functionality, such as reliable code completion, the ability to jump to the definition of a class, or inline documentation, if normal imports are used. The import * syntax can hamper their ability to do this reliably.
• Finally, using the import * syntax can bring unexpected objects into our local namespace. Sure, it will import all the classes and functions defined in the module being imported from, but unless a special __all__ list is provided in the module, this import will also import any classes or modules that were themselves imported into that file!

Every name used in a module should come from a well-specified place, whether it is defined in that module, or explicitly imported from another module. There should be no magic variables that seem to come out of thin air. We should always be able to immediately identify where the names in our current namespace originated. We promise that if you use this evil syntax, you will one day have extremely frustrating moments of where on earth can this class be coming from?

Organizing modules

As a project grows into a collection of more and more modules, we may find that we want to add another level of abstraction, some kind of nested hierarchy on our modules' levels. However, we can't put modules inside modules; one file can hold only one file after all, and modules are just files.

Files, however, can go in folders, and so can modules. A package is a collection of modules in a folder. The name of the package is the name of the folder. We need to tell Python that a folder is a package to distinguish it from other folders in the directory. To do this, place a (normally empty) file in the folder named __init__.py. If we forget this file, we won't be able to import modules from that folder.

Let's put our modules inside an ecommerce package in our working folder, which will also contain a main.py file to start the program. Let's additionally add another package inside the ecommerce package for various payment options.

We need to exercise some caution in creating deeply nested packages. The general advice in the Python community is "flat is better than nested." In this example, we need to create a nested package because there are some common features to all of the various payment alternatives.

The folder hierarchy will look like this, rooted under a directory in the project folder, commonly named src:

src/
 +-- main.py
 +-- ecommerce/
     +-- __init__.py
     +-- database.py
     +-- products.py
     +-- payments/
     |   +-- __init__.py
     |   +-- common.py
     |   +-- square.py
     |   +-- stripe.py
     +-- contact/
         +-- __init__.py
         +-- email.py

The src directory will be part of an overall project directory. In addition to src, the project will often have directories with names like docs and tests. It's common for the project parent directory to also have configuration files for tools like mypy among others. We'll return to this in Chapter 13, Testing Object-Oriented Programs.

When importing modules or classes between packages, we have to be cautious about the structure of our packages. In Python 3, there are two ways of importing modules: absolute imports and relative imports. We'll look at each of them separately.

Absolute imports

Absolute imports specify the complete path to the module, function, or class we want to import. If we need access to the Product class inside the products module, we could use any of these syntaxes to perform an absolute import:

>>> import ecommerce.products
>>> product = ecommerce.products.Product("name1") 

Or, we could specifically import a single class definition from the module within a package:

>>> from ecommerce.products import Product 
>>> product = Product("name2") 

Or, we could import an entire module from the containing package:

>>> from ecommerce import products 
>>> product = products.Product("name3") 

The import statements use the period operator to separate packages or modules. A package is a namespace that contains module names, much in the way an object is a namespace containing attribute names.

These statements will work from any module. We could instantiate a Product class using this syntax in main.py, in the database module, or in either of the two payment modules. Indeed, assuming the packages are available to Python, it will be able to import them. For example, the packages can also be installed in the Python site-packages folder, or the PYTHONPATH environment variable could be set to tell Python which folders to search for packages and modules it is going to import.

With these choices, which syntax do we choose? It depends on your audience and the application at hand. If there are dozens of classes and functions inside the products module that we want to use, we'd generally import the module name using the from ecommerce import products syntax, and then access the individual classes using products.Product. If we only need one or two classes from the products module, we can import them directly using the from ecommerce.products import Product syntax. It's important to write whatever makes the code easiest for others to read and extend.

Relative imports

When working with related modules inside a deeply nested package, it seems kind of redundant to specify the full path; we know what our parent module is named. This is where relative imports come in. Relative imports identify a class, function, or module as it is positioned relative to the current module. They only make sense inside module files, and, further, they only make sense where there's a complex package structure.

For example, if we are working in the products module and we want to import the Database class from the database module next to it, we could use a relative import:

from .database import Database 

The period in front of database says use the database module inside the current package. In this case, the current package is the package containing the products.py file we are currently editing, that is, the ecommerce package.

If we were editing the stripe module inside the ecommerce.payments package, we would want, for example, to use the database package inside the parent package instead. This is easily done with two periods, as shown here:

from ..database import Database 

We can use more periods to go further up the hierarchy, but at some point, we have to acknowledge that we have too many packages. Of course, we can also go down one side and back up the other. The following would be a valid import from the ecommerce.contact package containing an email module if we wanted to import the send_mail function into our payments.stripe module:

from ..contact.email import send_mail

This import uses two periods indicating the parent of the payments.stripe package, and then uses the normal package.module syntax to go back down into the contact package to name the email module.

Relative imports aren't as useful as they might seem. As mentioned earlier, the Zen of Python (you can read it when you run import this) suggests "flat is better than nested". Python's standard library is relatively flat, with few packages and even fewer nested packages. If you're familiar with Java, the packages are deeply nested, something the Python community likes to avoid. A relative import is needed to solve a specific problem where module names are reused among packages. They can be helpful in a few cases. Needing more than two dots to locate a common parent-of-a-parent package suggests the design should be flattened out.

Packages as a whole

We can import code that appears to come directly from a package, as opposed to a module inside a package. As we'll see, there's a module involved, but it has a special name, so it's hidden. In this example, we have an ecommerce package containing two module files named database.py and products.py. The database module contains a db variable that is accessed from a lot of places. Wouldn't it be convenient if this could be imported as from ecommerce import db instead of from ecommerce.database import db?

Remember the __init__.py file that defines a directory as a package? This file can contain any variable or class declarations we like, and they will be available as part of the package. In our example, if the ecommerce/__init__.py file contained the following line:

from .database import db

We could then access the db attribute from main.py or any other file using the following import:

from ecommerce import db

It might help to think of the ecommerce/__init__.py file as if it were the ecommerce.py file. It lets us view the ecommerce package as having a module protocol as well as a package protocol. This can also be useful if you put all your code in a single module and later decide to break it up into a package of modules. The __init__.py file for the new package can still be the main point of contact for other modules using it, but the code can be internally organized into several different modules or subpackages.

We recommend not putting much code in an __init__.py file, though. Programmers do not expect actual logic to happen in this file, and much like with from x import *, it can trip them up if they are looking for the declaration of a particular piece of code and can't find it until they check __init__.py.

After looking at modules in general, let's dive into what should be inside a module. The rules are flexible (unlike other languages). If you're familiar with Java, you'll see that Python gives you some freedom to bundle things in a way that's meaningful and informative.

Organizing our code in modules

The Python module is an important focus. Every application or web service has at least one module. Even a seemingly "simple" Python script is a module. Inside any one module, we can specify variables, classes, or functions. They can be a handy way to store the global state without namespace conflicts. For example, we have been importing the Database class into various modules and then instantiating it, but it might make more sense to have only one database object globally available from the database module. The database module might look like this:

class Database:
    """The Database Implementation"""
    def __init__(self, connection: Optional[str] = None) -> None:
        """Create a connection to a database."""
        pass
database = Database("path/to/data")

Then we can use any of the import methods we've discussed to access the database object, for example:

from ecommerce.database import database 

A problem with the preceding class is that the database object is created immediately when the module is first imported, which is usually when the program starts up. This isn't always ideal, since connecting to a database can take a while, slowing down startup, or the database connection information may not yet be available because we need to read a configuration file. We could delay creating the database until it is actually needed by calling an initialize_database() function to create a module-level variable:

db: Optional[Database] = None
def initialize_database(connection: Optional[str] = None) -> None:
    global db
    db = Database(connection)

The Optional[Database] type hint signals to the mypy tool that this may be None or it may have an instance of the Database class. The Optional hint is defined in the typing module. This hint can be handy elsewhere in our application to make sure we confirm that the value for the database variable is not None.

The global keyword tells Python that the database variable inside initialize_database() is the module-level variable, outside the function. If we had not specified the variable as global, Python would have created a new local variable that would be discarded when the function exits, leaving the module-level value unchanged.

We need to make one additional change. We need to import the database module as a whole. We can't import the db object from inside the module; it might not have been initialized. We need to be sure database.initialize_database() is called before db will have a meaningful value. If we wanted direct access to the database object, we'd use database.db.

A common alternative is a function that returns the current database object. We could import this function everywhere we needed access to the database:

def get_database(connection: Optional[str] = None) -> Database:
    global db
    if not db:
        db = Database(connection) 
    return db

As these examples illustrate, all module-level code is executed immediately at the time it is imported. The class and def statements create code objects to be executed later when the function is called. This can be a tricky thing for scripts that perform execution, such as the main script in our e-commerce example. Sometimes, we write a program that does something useful, and then later find that we want to import a function or class from that module into a different program. However, as soon as we import it, any code at the module level is immediately executed. If we are not careful, we can end up running the first program when we really only meant to access a couple of functions inside that module.

To solve this, we should always put our startup code in a function (conventionally, called main()) and only execute that function when we know we are running the module as a script, but not when our code is being imported from a different script. We can do this by guarding the call to main inside a conditional statement, demonstrated as follows:

class Point:
    """
    Represents a point in two-dimensional geometric coordinates.
    """
    pass
def main() -> None:
    """
    Does the useful work.
    >>> main()
    p1.calculate_distance(p2)=5.0
    """
    p1 = Point()
    p2 = Point(3, 4)
    print(f"{p1.calculate_distance(p2)=}")
if __name__ == "__main__":
    main()

The Point class (and the main() function) can be reused without worry. We can import the contents of this module without any surprising processing happening. When we run it as a main program, however, it executes the main() function.

This works because every module has a __name__ special variable (remember, Python uses double underscores for special variables, such as a class' __init__ method) that specifies the name of the module when it was imported. When the module is executed directly with python module.py, it is never imported, so the __name__ is arbitrarily set to the "__main__" string.

So, methods go in classes, which go in modules, which go in packages. Is that all there is to it?

Actually, no. This is the typical order of things in a Python program, but it's not the only possible layout. Classes can be defined anywhere. They are typically defined at the module level, but they can also be defined inside a function or method, like this:

from typing import Optional
class Formatter:
    def format(self, string: str) -> str:
        pass
def format_string(string: str, formatter: Optional[Formatter] = None) -> str:
    """
    Format a string using the formatter object, which
    is expected to have a format() method that accepts
    a string.
    """
    class DefaultFormatter(Formatter):
        """Format a string in title case."""
        def format(self, string: str) -> str:
            return str(string).title()
    if not formatter:
        formatter = DefaultFormatter()
    return formatter.format(string)

We've defined a Formatter class as an abstraction to explain what a formatter class needs to have. We haven't used the abstract base class (abc) definitions (we'll look at these in detail in Chapter 6, Abstract Base Classes and Operator Overloading). Instead, we've provided the method with no useful body. It has a full suite of type hints, to make sure mypy has a formal definition of our intent.

Within the format_string() function, we created an internal class that is an extension of the Formatter class. This formalizes the expectation that our class inside the function has a specific set of methods. This connection between the definition of the Formatter class, the formatter parameter, and the concrete definition of the DefaultFormatter class assures that we haven't accidentally forgotten something or added something.

We can execute this function like this:

>>> hello_string = "hello world, how are you today?"
>>> print(f" input: {hello_string}")
 input: hello world, how are you today?
>>> print(f"output: {format_string(hello_string)}")
output: Hello World, How Are You Today?

The format_string function accepts a string and optional Formatter object and then applies the formatter to that string. If no Formatter instance is supplied, it creates a formatter of its own as a local class and instantiates it. Since it is created inside the scope of the function, this class cannot be accessed from anywhere outside of that function. Similarly, functions can be defined inside other functions as well; in general, any Python statement can be executed at any time.

These inner classes and functions are occasionally useful for one-off items that don't require or deserve their own scope at the module level, or only make sense inside a single method. However, it is not common to see Python code that frequently uses this technique.

We've seen how to create classes and how to create modules. With these core techniques, we can start thinking about writing useful, helpful software to solve problems. When the application or service gets big, though, we often have boundary issues. We need to be sure that objects respect each other's privacy and avoid confusing entanglements that make complex software into a spaghetti bowl of interrelationships. We'd prefer each class to be a nicely encapsulated ravioli. Let's look at another aspect of organizing our software to create a good design.