Creational design patterns deal with an object creation [j.mp/wikicrea]. The aim of a creational design pattern is to provide better alternatives for situations where a direct object creation (which in Python happens by the __init__() function [j.mp/divefunc], [Lott14, page 26]) is not convenient.

In the Factory design pattern, a client asks for an object without knowing where the object is coming from (that is, which class is used to generate it). The idea behind a factory is to simplify an object creation. It is easier to track which objects are created if this is done through a central function, in contrast to letting a client create objects using a direct class instantiation [Eckel08, page 187]. A factory reduces the complexity of maintaining an application by decoupling the code that creates an object from the code that uses it [Zlobin13, page 30].

Factories typically come in two forms: the Factory Method, which is a method (or in Pythonic terms, a function) that returns a different object per input parameter [j.mp/factorympat]; the Abstract Factory, which is a group of Factory Methods used to create a family of related products [GOF95, page 100], [j.mp/absfpat]

(For more resources related to this topic, see here.)

Factory Method

In the Factory Method, we execute a single function, passing a parameter that provides information about what we want. We are not required to know any details about how the object is implemented and where it is coming from.

A real-life example

An example of the Factory Method pattern used in reality is in plastic toy construction. The molding powder used to construct plastic toys is the same, but different figures can be produced using different plastic molds. This is like having a Factory Method in which the input is the name of the figure that we want (soldier and dinosaur) and the output is the plastic figure that we requested. The toy construction case is shown in the following figure, which is provided by www.sourcemaking.com [j.mp/factorympat].

python-design-patterns-depth-factory-pattern-img-0

A software example

The Django framework uses the Factory Method pattern for creating the fields of a form. The forms module of Django supports the creation of different kinds of fields (CharField, EmailField) and customizations (max_length, required) [j.mp/djangofacm].

Use cases

If you realize that you cannot track the objects created by your application because the code that creates them is in many different places instead of a single function/method, you should consider using the Factory Method pattern [Eckel08, page 187]. The Factory Method centralizes an object creation and tracking your objects becomes much more easier. Note that it is absolutely fine to create more than one Factory Method, and this is how it is typically done in practice. Each Factory Method logically groups the creation of objects that have similarities. For example, one Factory Method might be responsible for connecting you to different databases (MySQL, SQLite), another Factory Method might be responsible for creating the geometrical object that you request (circle, triangle), and so on.

The Factory Method is also useful when you want to decouple an object creation from an object usage. We are not coupled/bound to a specific class when creating an object, we just provide partial information about what we want by calling a function. This means that introducing changes to the function is easy without requiring any changes to the code that uses it [Zlobin13, page 30].

Another use case worth mentioning is related with improving the performance and memory usage of an application. A Factory Method can improve the performance and memory usage by creating new objects only if it is absolutely necessary [Zlobin13, page 28]. When we create objects using a direct class instantiation, extra memory is allocated every time a new object is created (unless the class uses caching internally, which is usually not the case). We can see that in practice in the following code (file id.py), it creates two instances of the same class A and uses the id() function to compare their memory addresses. The addresses are also printed in the output so that we can inspect them. The fact that the memory addresses are different means that two distinct objects are created as follows:

class A(object):

    pass

if __name__ == '__main__':

    a = A()

    b = A()

    print(id(a) == id(b))

    print(a, b)

Executing id.py on my computer gives the following output:
>> python3 id.py
False
<__main__.A object at 0x7f5771de8f60> <__main__.A object at 0x7f5771df2208>

Note that the addresses that you see if you execute the file are not the same as I see because they depend on the current memory layout and allocation. But the result must be the same: the two addresses should be different. There's one exception that happens if you write and execute the code in the Python Read-Eval-Print Loop (REPL) (interactive prompt), but that's a REPL-specific optimization which is not happening normally.

Implementation

Data comes in many forms. There are two main file categories for storing/retrieving data: human-readable files and binary files. Examples of human-readable files are XML, Atom, YAML, and JSON. Examples of binary files are the .sq3 file format used by SQLite and the .mp3 file format used to listen to music.

In this example, we will focus on two popular human-readable formats: XML and JSON. Although human-readable files are generally slower to parse than binary files, they make data exchange, inspection, and modification much more easier. For this reason, it is advised to prefer working with human-readable files, unless there are other restrictions that do not allow it (mainly unacceptable performance and proprietary binary formats).

In this problem, we have some input data stored in an XML and a JSON file, and we want to parse them and retrieve some information. At the same time, we want to centralize the client's connection to those (and all future) external services. We will use the Factory Method to solve this problem. The example focuses only on XML and JSON, but adding support for more services should be straightforward.

First, let's take a look at the data files. The XML file, person.xml, is based on the Wikipedia example [j.mp/wikijson] and contains information about individuals (firstName, lastName, gender, and so on) as follows:

<persons>

  <person>

    <firstName>John</firstName>

    <lastName>Smith</lastName>

    <age>25</age>

    <address>

      <streetAddress>21 2nd Street</streetAddress>

      <city>New York</city>

      <state>NY</state>

      <postalCode>10021</postalCode>

    </address>

    <phoneNumbers>

      <phoneNumber type="home">212 555-1234</phoneNumber>

      <phoneNumber type="fax">646 555-4567</phoneNumber>

    </phoneNumbers>

    <gender>

      <type>male</type>

    </gender>

  </person>

  <person>

    <firstName>Jimy</firstName>

    <lastName>Liar</lastName>

    <age>19</age>

    <address>

      <streetAddress>18 2nd Street</streetAddress>

      <city>New York</city>

      <state>NY</state>

      <postalCode>10021</postalCode>

    </address>

    <phoneNumbers>

      <phoneNumber type="home">212 555-1234</phoneNumber>

    </phoneNumbers>

    <gender>

      <type>male</type>

    </gender>

  </person>

  <person>

    <firstName>Patty</firstName>

    <lastName>Liar</lastName>

    <age>20</age>

    <address>

      <streetAddress>18 2nd Street</streetAddress>

      <city>New York</city>

      <state>NY</state>

      <postalCode>10021</postalCode>

    </address>

    <phoneNumbers>

      <phoneNumber type="home">212 555-1234</phoneNumber>

      <phoneNumber type="mobile">001 452-8819</phoneNumber>

    </phoneNumbers>

    <gender>

      <type>female</type>

    </gender>

  </person>

</persons>

The JSON file, donut.json, comes from the GitHub account of Adobe [j.mp/adobejson] and contains donut information (type, price/unit i.e. ppu, topping, and so on) as follows:

[

  {

    "id": "0001",

    "type": "donut",

    "name": "Cake",

    "ppu": 0.55,

    "batters": {

      "batter": [

        { "id": "1001", "type": "Regular" },

        { "id": "1002", "type": "Chocolate" },

        { "id": "1003", "type": "Blueberry" },

        { "id": "1004", "type": "Devil's Food" }

      ]

    },

    "topping": [

      { "id": "5001", "type": "None" },

      { "id": "5002", "type": "Glazed" },

      { "id": "5005", "type": "Sugar" },

      { "id": "5007", "type": "Powdered Sugar" },

      { "id": "5006", "type": "Chocolate with Sprinkles" },

      { "id": "5003", "type": "Chocolate" },

      { "id": "5004", "type": "Maple" }

    ]

  },

  {

    "id": "0002",

    "type": "donut",

    "name": "Raised",

    "ppu": 0.55,

    "batters": {

      "batter": [

        { "id": "1001", "type": "Regular" }

      ]

    },

    "topping": [

      { "id": "5001", "type": "None" },

      { "id": "5002", "type": "Glazed" },

      { "id": "5005", "type": "Sugar" },

      { "id": "5003", "type": "Chocolate" },

      { "id": "5004", "type": "Maple" }

    ]

  },

  {

    "id": "0003",

    "type": "donut",

    "name": "Old Fashioned",

    "ppu": 0.55,

    "batters": {

      "batter": [

        { "id": "1001", "type": "Regular" },

        { "id": "1002", "type": "Chocolate" }

      ]

    },

    "topping": [

      { "id": "5001", "type": "None" },

      { "id": "5002", "type": "Glazed" },

      { "id": "5003", "type": "Chocolate" },

      { "id": "5004", "type": "Maple" }

    ]

  }

]

We will use two libraries that are part of the Python distribution for working with XML and JSON: xml.etree.ElementTree and json as follows:

import xml.etree.ElementTree as etree

import json

The JSONConnector class parses the JSON file and has a parsed_data() method that returns all data as a dictionary (dict). The property decorator is used to make parsed_data() appear as a normal variable instead of a method as follows:

class JSONConnector:

    def __init__(self, filepath):

        self.data = dict()

        with open(filepath, mode='r', encoding='utf-8') as f:

            self.data = json.load(f)

 

    @property

    def parsed_data(self):

        return self.data

The XMLConnector class parses the XML file and has a parsed_data() method that returns all data as a list of xml.etree.Element as follows:

class XMLConnector:

    def __init__(self, filepath):

        self.tree = etree.parse(filepath)

    @property

   def parsed_data(self):

        return self.tree

The connection_factory() function is a Factory Method. It returns an instance of JSONConnector or XMLConnector depending on the extension of the input file path as follows:

def connection_factory(filepath):

    if filepath.endswith('json'):

        connector = JSONConnector

    elif filepath.endswith('xml'):

        connector = XMLConnector

    else:

        raise ValueError('Cannot connect to {}'.format(filepath))

    return connector(filepath)

The connect_to() function is a wrapper of connection_factory(). It adds exception handling as follows:

def connect_to(filepath):

    factory = None

    try:

        factory = connection_factory(filepath)

    except ValueError as ve:

        print(ve)

    return factory

The main() function demonstrates how the Factory Method design pattern can be used. The first part makes sure that exception handling is effective as follows:

def main():

    sqlite_factory = connect_to('data/person.sq3')

The next part shows how to work with the XML files using the Factory Method. XPath is used to find all person elements that have the last name Liar. For each matched person, the basic name and phone number information are shown as follows:

xml_factory = connect_to('data/person.xml')

    xml_data = xml_factory.parsed_data()

    liars = xml_data.findall
    (".//{person}[{lastName}='{}']".format('Liar'))

    print('found: {} persons'.format(len(liars)))

    for liar in liars:

        print('first name: 
        {}'.format(liar.find('firstName').text))

        print('last name: {}'.format(liar.find('lastName').text))

        [print('phone number ({}):'.format(p.attrib['type']), 
        p.text) for p in liar.find('phoneNumbers')]

The final part shows how to work with the JSON files using the Factory Method. Here, there's no pattern matching, and therefore the name, price, and topping of all donuts are shown as follows:

json_factory = connect_to('data/donut.json')

    json_data = json_factory.parsed_data

    print('found: {} donuts'.format(len(json_data)))

    for donut in json_data:

        print('name: {}'.format(donut['name']))

        print('price: ${}'.format(donut['ppu']))

        [print('topping: {} {}'.format(t['id'], t['type'])) for t 
        in donut['topping']]

For completeness, here is the complete code of the Factory Method implementation (factory_method.py) as follows:

import xml.etree.ElementTree as etree

import json

class JSONConnector:

    def __init__(self, filepath):

        self.data = dict()

        with open(filepath, mode='r', encoding='utf-8') as f:

            self.data = json.load(f)

    @property

    def parsed_data(self):

        return self.data

class XMLConnector:

    def __init__(self, filepath):

        self.tree = etree.parse(filepath)

    @property

    def parsed_data(self):

        return self.tree

def connection_factory(filepath):

    if filepath.endswith('json'):

        connector = JSONConnector

    elif filepath.endswith('xml'):

        connector = XMLConnector

    else:

        raise ValueError('Cannot connect to {}'.format(filepath))

    return connector(filepath)

def connect_to(filepath):

    factory = None

    try:

       factory = connection_factory(filepath)

    except ValueError as ve:

        print(ve)

    return factory

def main():

    sqlite_factory = connect_to('data/person.sq3')

    print()

    xml_factory = connect_to('data/person.xml')

    xml_data = xml_factory.parsed_data

    liars = xml_data.findall(".//{}[{}='{}']".format('person', 
    'lastName', 'Liar'))

    print('found: {} persons'.format(len(liars)))

    for liar in liars:

        print('first name: 
        {}'.format(liar.find('firstName').text))

        print('last name: {}'.format(liar.find('lastName').text))

        [print('phone number ({}):'.format(p.attrib['type']), 
        p.text) for p in liar.find('phoneNumbers')]

    print()

    json_factory = connect_to('data/donut.json')

    json_data = json_factory.parsed_data

    print('found: {} donuts'.format(len(json_data)))

    for donut in json_data:

    print('name: {}'.format(donut['name']))

    print('price: ${}'.format(donut['ppu']))

    [print('topping: {} {}'.format(t['id'], t['type'])) for t 
    in donut['topping']]

if __name__ == '__main__':

    main()

Here is the output of this program as follows:

>>> python3 factory_method.py
Cannot connect to data/person.sq3
found: 2 persons
first name: Jimy
last name: Liar
phone number (home): 212 555-1234
first name: Patty
last name: Liar
phone number (home): 212 555-1234
phone number (mobile): 001 452-8819
found: 3 donuts
name: Cake
price: $0.55
topping: 5001 None
topping: 5002 Glazed
topping: 5005 Sugar
topping: 5007 Powdered Sugar
topping: 5006 Chocolate with Sprinkles
topping: 5003 Chocolate
topping: 5004 Maple
name: Raised
price: $0.55
topping: 5001 None
topping: 5002 Glazed
topping: 5005 Sugar
topping: 5003 Chocolate
topping: 5004 Maple
name: Old Fashioned
price: $0.55
topping: 5001 None
topping: 5002 Glazed
topping: 5003 Chocolate
topping: 5004 Maple

Notice that although JSONConnector and XMLConnector have the same interfaces, what is returned by parsed_data() is not handled in a uniform way. Different codes must be used to work with each connector. Although it would be nice to be able to use the same code for all connectors, this is at most times not realistic unless we use some kind of common mapping for the data which is very often provided by external data providers. Assuming that you can use exactly the same code for handling the XML and JSON files, what changes are required to support a third format, for example, SQLite? Find an SQLite file or create your own and try it.

As is now, the code does not forbid a direct instantiation of a connector. Is it possible to do this? Try doing it (hint: functions in Python can have nested classes).