Python is a mature language, but it hasn't stopped evolving, and there are some issues to consider when coding Python, if you want your code to work with the latest version of Python five years from now. See PEP 3000 for more information.
- Keep in mind that only a little subset of existing Python 2.6 code is valid on Python 3.0. There is no guarantee that your code will run unmodified on Python 3.0, even if you follow the advices listed here.
See also the 2to3 tool, that helps translating Python 2.6 to Python 3.0 code. Some changes listed here can be automatically translated with this tool. Also, check the changes that can't be translated by the tool.
Automated testing is generally better than debugging, and is your safety-net for any kind of refactoring, not only when migrating to a newer Python. Use PyUnit (unittest), DocTest or any alternative (see UnitTests) to help you.
Write unit tests for all new or refactored code.
Since the beginning, Python has yielded an integer result when two integers are divided (e.g., 3/2 yields 1). While this is correct if we assume that dividing integers means integer division, it's not always obvious to beginners. This behaviour will change in a future Python version, so that a/b will yield a float as a result even if both a and b are integers, and a new floor division operator // will perform integer division. See See PEP 238 for more information.
Use true and floor division in new code.
You can enable the new behaviour in Python 2.x as follows:
>>> from __future__ import division # Enable the new behaviour >>> 3/2 1.5 >>> 3//2 1
In Python 2.x, imports are implicitly relative. For instance, if you're editing the file foo/__init__.py and want to import the module at foo/bar.py, you could use import bar.
In Python 3.0, this won't work, as all imports will be absolute by default. You should instead use from foo import bar; if you want to import a specific function or variable from bar, you can use relative imports, such as from .bar import myfunction). See PEP 328 for more information.
Don't use implicitly relative imports in new code.
New style classes
Currently, there are two kinds of classes in Python. The 'classic' or old style classes, and the new style classes. Old style classes will go away in Python 3.0, and while most code will still work when the default switches from old style to new style, there are some differences in semantics, and the new style classes have some extra features. See New-style Classes for more information.
Use new style classes in new code.
- Don't write
class X: pass
- Instead, write
class X(object): pass
Exception class inheritance
Starting from Python 3.0, all exceptions must be derived from BaseException, which will be the base class for KeyboardInterrupt, SystemExit and Exception from Python 2.5. See PEP 352 for more information.
When defining new exception classes, always inherit (directly or indirectly) from Exception.
class MyException(Exception): pass
Arguments for raise statement
The raise statement currently accepts different syntaxes. This will be consolidated so that only one argument is allowed for raise: the exception object, created like any other object. Other syntaxes will be deprecated.
Write the raise statement with the exception object as the only argument.
- Don't write:
raise MyException, "A nasty error" raise "Foo bar error: Invalid value"
- Instead, write:
raise MyException("A nasty error") raise FooBarError("Invalid value")
Comparison of incompatible types
In the future, x < y and friends (>, <=, >=) will raise an exception instead of an arbitrary result, if type(x) != type(y) unless the types explicitly define the behaviour for these comparisions.
Don't compare objects of different types with undefined comparison behaviour.
Deprecation of little-used alternative operators
One obvious way to test for inequality already exists: the != operator. The rarely-used <> will go away.
Use only != for inequality comparison.
The backtick syntax `foo`, which is equivalent to repr(foo), will not exist in Python 3.0. Almost all code currently uses repr(foo) for this purpose, and the backtick syntax is little-known and less readable.
Use only repr(foo) for the representation of an object.
All of the names with, as, nonlocal, True and False are becoming keywords. Code that attempts to assign to those names will generate an exception in Python 3.0.
Don't assign to names that will become keywords.
Here are some more tips on writing Python code to be future proof with Python 3.0.
Collin Winter has some tips in: http://oakwinter.com/code/porting-setuptools-to-py3k/
Fabio Zadrozny gives some tips for those interested in writing code that runs on Python 2 and Python 3 in: http://pydev.blogspot.com/2008/11/making-code-work-in-python-2-and-3.html
More changes in the future?
The following little code might be useful to run when you upgrade to a new Python version. It shows how some new features in Python are gradually becoming enabled. (This example is from Python 2.3.)
>>> import __future__ >>> for x in __future__.all_feature_names: ... print x, getattr(__future__, x) ... nested_scopes _Feature((2, 1, 0, 'beta', 1), (2, 2, 0, 'alpha', 0), 16) generators _Feature((2, 2, 0, 'alpha', 1), (2, 3, 0, 'final', 0), 4096) division _Feature((2, 2, 0, 'alpha', 2), (3, 0, 0, 'alpha', 0), 8192)
The second tuple (e.g. (3, 0, 0, 'alpha', 0) for division) shows when a new feature will become the default.