|Ned Batchelder : Blog | Code | Text | Site|
» Home : Blog : June 2008
Being a developer, I'm a sucker for rules to follow to improve my code, and for tools that can help me to follow them. Being a Python developer, I don't have a static type checking compiler to help me. Pylint aims to fill some of those gaps.
It examines your Python source code and reports on all sorts of things it doesn't like. Like most tools of this sort (its name is a reference to the classic lint tool for C), it can be annoyingly picky. Since its job is to flag things that might be a problem, it errs on the alarmist noisy side.
Pylint tries to apply light type checking to methods and variables, so it will complain about constructs simply because they interfere with that goal:
W0142: 26:MyClass.my_method: Used * or ** magic
Excuse me, but ** is not magic, it's a powerful language feature. Reading pylint's warnings, you get the feeling that it won't be completely happy until you are coding within the intersection of Python and Java.
But pylint's best feature is its configurability. In the settings file, you can disable individual messages:
# Disable the message(s) with the given id(s).
and also configure all sorts of other settings. This is important because pylint also natters on about style issues: valid names, line length, number of statements per method, and so on. Pylint also lets you disable a particular message in specific files, classes, or methods, which is extremely useful for overriding warnings about tricky cases, or simple misdiagnoses.
As with every other lint-like tool I've ever used, the first order of business is deciding what you really want pylint to tell you about. Initially, its reporting will be about things that just don't matter, and you'll disable a ton of messages. Then you'll get to the things that you'll agree are minor issues and you'll want to clean up, like unused imports.
The next rung of messages are helpful because they get you to think about the way you've written your code. For example, this code:
def my_method(self, arg1, extras=):
will get you this warning:
W0102:247:MyClass.my_method: Dangerous default value  as argument
Pylint warns about this because you could append to extras in the body of the method, and that would modify the single list object that is used as the default value for every invocation of the method, something you almost certainly didn't intend. Changing the code to this avoids the possibility and the warning:
def my_method(self, arg1, extras=None):
Whether you want to adopt this idiom uniformly, or stick with the more common extras= is something you'll have to decide. Pylint did you the favor of bringing it to your attention so that you could think through the issue and decide. In this case, you may be able to simply leave extras as None and use it as is in the body of the function, but you get the point.
Occasionally, you'll get unambiguous value from the pylint output. I ran pylint on a large actively developed code base, and it reported on an instance of an undefined variable. I looked, and sure enough, that code shouldn't work. Digging further, I looked at who called that code, and once I was done pulling on all of the threads I discovered, I had a couple hundred lines of code that were not used any more, and I could delete them.
I don't know whether I'll stick with pylint. It's a tricky balance to get it set properly so that it warns about things I genuinely believe to be issues.
The other minor downside to pylint is that you have to install three separate packages to get it to work. Logilab would do well to provide a single installer for pylint and its dependencies.