Modularity/Development/Coding Style

Python

Most of our code is written in Python, so this document will concentrate on it.

PEP 8: Official Python Style Guide

Fortunately, with PEP 8 there's an official Style Guide for Python Code. All new Python code you submit should conform to it, unless you have good reasons to deviate from it, for instance readability.

Keep It Simple

The code you write now probably needs to be touched by someone else down the road, and that someone else might be less experienced than you, or have a terrible headache and be under pressure of time. So while a particular construct may be a clever way of doing something, a simple way of doing the same thing can be and often is preferrable.

Python 2 and 3

Python comes in two major versions nowadays:

• The legacy version 2, of which the first release 2.0 came out in October 2000. The Python project will maintain its final minor release 2.7 until 2020.

• The current version 3, its first release 3.0 was published in December 2008. At the time of writing, the current minor release is version 3.5, to be superseded by 3.6 around the end of 2016.

Version 3 is not backwards compatible to version 2. While we mainly target "the future", there are some components we have to work with that haven't yet been ported over the Python 3, most notably koji. Additionally, we may also want to support the "user tools" we create on legacy systems, so we can't write code that uses all the latest features. Fortunately, many of the original Python 3 features have been back-ported to Python 2.7, so we can and should write code that is very close to writing idiomatic Python 3 but can still be run on version 2.7. Targeting older minor releases (Python 2.6 and earlier) is much more of a balancing act, so we won't aim for it.

The following sections cover areas that require some attention. The Python project itself has a great Porting Python 2 Code to Python 3 document which goes into much detail about the differences and is worth a read, even though it mainly addresses existing Python 2 code bases.

Absolute and relative imports

In Python 2, importing modules can be ambiguous when a module of that name exists in the same package and elsewhere in the module search path sys.path. To work around this ambiguity, programmers often resorted to adding paths private to the project to the beginning of sys.path to force loading modules from a project-internal location (which adds unwanted noise and can make e.g. testing code that isn't installed difficult). Python 3 introduces new syntax for import statements which makes both cases distinct, this is available since version 2.5 from the __future__ module:

from __future__ import absolute_import

# Import the sys module from the module search path
import sys

# Import the foo module from the same directory
from . import foo

# Import snafu from the bar module one directory above
from ..bar import snafu

Print function

Python 3 did away with print as a statement and introduced it as a function. In order to use it the same way in Python 2.7, add the following to the top of source code files where you use print:

from __future__ import print_function

Numbers

Python 2 has two integer types, int which is whatever integer-type is native to the system (which has certain maximal and minimal values and can overflow) and long which can store arbitrary integer numbers. Python 3 only the latter type, but it's called int.

Dividing integer numbers using / truncates the result to an integer in Python 2 by default, but yields a floating point number in Python 3. In order for code to do the same thing on either version, include the following line at the top of your source files where you divide numbers, and use / for normal divisions and // for divisions that should truncate the result:

from __future__ import division

Strings

Some consider this the main difference between Python 2 and 3: Both versions have a type for strings of bytes and strings of Unicode character points. They are called str and unicode in version 2 and bytes and str in version 3, respectively.

String Literals

Python 2 and 3 use different ways of marking literals of the different types by default. Byte strings can have no prefix or b in Python 2.7, but must be prefixed in Python 3, and text strings must have the u prefix in Python 2 which can be and usually is omitted in Python 3:

# a byte string in Python 2 and 3
string1 = b"abc"

# a byte string in Python 2, but a text string in Python 3
string2 = "def"

# a text string in Python 2 and 3
string3 = u"ghi"

In order to ease writing code that is compatible between the versions, you can switch Python 2 to treat unprefixed string literals as unicode, the text string type, by adding this snippet to the top of the relevant source code files:

from __future__ import unicode_literals

Explicit Encoding and Decoding

In Python 2, the byte and text string types are exchangeable in many places, taking the user's or system default locale into account (and sometimes failing, when the locale didn't match up with encoded data). Apart from the change in type names and how literals look like, Python 3 requires you to explicitly encode str and decode bytes objects if you need them cast into the respective other string type. It is good practice to exclusively use text strings for strings that represent text in a program and decode byte strings as early and encode text strings as late as possible at interfaces that produce or consume encoded data.

Implicit string type conversion in Python 2
Python 2 lets you attempt to replace a str substring in a unicode object (or vice versa) and would attempt to cast the one into the other by encoding or decoding on the fly as needed. This piece of code won't work in Python 3:

from __future__ import print_function
text_string = u"Hello, world!"
print(text_string.replace("world", "gang"))

Explicit string type conversion in Python 2 and 3
Python 3 requires explicit encoding/decoding to cast between byte and text strings. This also works in Python 2 and is preferred of course.

from __future__ import print_function, unicode_literals
text_string = "Hello, world!"
print(text_string.replace(b"world".decode('utf-8'), b"gang".decode('ascii')))

Old- and New-style Classes

Python 2 and earlier knows two types of classes, old-style which have no base class, and new-style which have object as the base class. Because their behavior is slightly different in some places, and some things can't be done with old-style classes, we want to stick to new-style classes wherever possible.

The syntactical difference is that new-style classes have to explicitly be derived from object or another new-style class.

# old-style classes
class OldFoo:
    pass

class OldBar(OldFoo):
    pass

# new-style classes
class NewFoo(object):
    pass

class NewBar(NewFoo):
    pass

Python 3 only knows new-style classes and the requirement to explicitly derive from object was dropped. In projects that will only ever run on Python 3, it's acceptable not to explicitly derive classes without parents from object, but if in doubt, do it just the same.

Idiomatic code

In Python, it's easy to inadvertently emulate idiomatic styles of other languages like C/C++ or Java. In cases where there are constructs "native" to the language, it's preferrable to use them.

Here are some examples:

Looping

Languages like C normally use incremented indices to loop over arrays:

float pixels[NUMBER_OF_PIXELS] = [...];

for (int i = 0; i < NUMBER_OF_PIXELS; i++)
{
    do_something_with_a_pixel(pixels[i]);
}

Looping C-style in Python
Please avoid looping over indices of sequences, rather than the sequences themselves in Python.

Implementing the loop like this would give away that you've programmed in C or a similar language before:

pixels = [...]

for i in range(len(pixels)):
    do_something_with_a_pixel(pixels[i])

Looping over iterables in Python
In Python, you can simply iterate over many non-scalar data types.

Here's the "native" way to implement the above loop:

pixels = [...]

for p in pixels:
    do_something_with_a_pixel(p)

Using enumerate()
If you need to keep track of the current count of looped-over items, use the enumerate() built-in.

It yields pairs of count and the current value like this:

pixels = [...]

for i, p in enumerate(pixels):
    print("Working on pixel no. {}".format(i + 1))
    do_something_with_a_pixel(p)

Properties rather than explicit accessor methods

In order to allow future changes in how object attributes (member variables) are set, some languages encourage always using getter and/or setter methods. This is unnecessary in Python, as you can intercept access to an attribute by wrapping it into a property. These allow having accessor methods without making the user of the class have to use them explicitly. This way you can validate values when an attribute is set, or translate back and forth between the interface used on the attribute and an internal representation.

Validating a value when setting an attribute

To ensure that an Employee object only has positive values for its salary attribute, you'd put a property in its place which checks values before storing them in an attribute called e.g. _salary:

class Employee(object):

    @property
    def salary(self):
        return self._salary

    @salary.setter
    def salary(self, salary):
        if salary <= 0:
            raise ValueError("Salary must be positive.")
        self._salary = salary

Avoid recursion
In order to avoid endless recursion, you must use a different attribute than the one using the property to store actual values.

Translating between attribute interface and internal representation

Take these classes of geometric primitives, Point and Circle:

class Point(object):
    def __init__(self, x, y):
        self.x = x
        self.y = y

class Circle(object):
    def __init__(self, point, radius):
        self.point = point
        self.radius = radius

If you wanted to add a diameter attribute, you can do so as a property which translates back and forth between it and the existing radius attribute:

...
class Circle(object):
    def __init__(self, point, radius=None, diameter=None):
        self.point = point
        if (radius is None) == (diameter is None):
            raise ValueError("Exactly one of radius or diameter must be set")
        if radius is not None:
            self.radius = radius
        else:
            self.diameter = diameter

    @property
    def diameter(self):
        return self.radius * 2

    @diameter.setter
    def diameter(self, diameter):
        self.radius = diameter / 2.0
...

Even setting self.diameter in the constructor goes by way of the property and therefore the setter method.