Wednesday 25 September 2013

PyCon UK 2013 Notes

Over the weekend, I was at PyCon UK 2013, which was fantastic! I took notes of some of the talks I went to, and the links to my Evernote notes are below:

For more PyCon UK follow-up goodness, check out the conference wiki.

Fixing pylint errors

What is

If you've engaged in any test-driven Python development, you've probably considered using nose as your test runner. One of my favourite parts of nose (among its many excellent parts) is the module. This not only contains a number of helpful tools (documented here), but also all of the assertion methods on unittest.TestCase converted to PEP-8 compatible functions (e.g. self.assertEqual becomes assert_equal).

These assertion functions are useful for a few reasons: readability, line length, PEP-8 compliance. Most importantly, though, is the fact that nose allows you to write test functions and test classes which don't subclass unittest.TestCase. The only way to access these assertions, therefore, is through

The assertion functions in are generated at runtime from unittest.TestCase and put in to the namespace. This has a couple of consequences: firstly, any new assertions in the standard library will immediately appear in (a word of warning: as nose doesn't implement the assertions, you need to pay attention to stdlib unittest assertion changes). Secondly, this means that they don't appear anywhere in the source code if you go looking for them. This will become relevant shortly.

What is pylint?

If you've engaged in any team Python development, you've probably considered pylint to help you maintain code quality. pylint uses static analysis to find problems in your code. Static analysis just means that it looks at your code as text, rather than executing it (which would be dynamic analysis). The problems it finds range from bad syntax to poor style; it can even work out when you've tried to import a non-existent symbol from another module.

What's the problem?

This last feature is really useful most of the time; it has saved me from pushing broken code innumerable times. However, it does cause one major headache: because the source code doesn't contain the PEP-8 assertion functions (remember, they're dynamically generated!), pylint's static analysis can't find them; every single time you import one of them you get a pylint error (E0611, to be precise). If you're using nose as part of your continuous integration to check your pylint errors, this can be a major source of pain. Luckily, pylint has a solution: disabling checks.

Fixing with disabled checks

pylint allows you to disable checks in a couple of ways: you can disable them globally (i.e. none of these warnings should ever be displayed), or based on scope (i.e. this module, class, function or line should never have warnings displayed). So, for example, an import line could look like:
from import assert_equal  # pylint: disable=E0611
Problem solved! Your pylint run will no longer flag this line up as problematic. There are a few issues with disabling checks, though. Firstly, you have to remember to do it. Secondly, it crufts up the screen when you're trying to read, and gives you more information that you need to process. Finally, and often most crucially, you can accidentally change the scope of your disabling. Consider the following lines at the top of a module:
# pylint: disable=E0611
from import assert_equal, assert_items_equal, assert_greater
This does solves the warning (and avoids violating PEP-8's line limit), but it has an unintentional consequence: E0611 is now disabled for the entire file. pylint will no longer catch this obvious error!
# pylint: disable=E0611
from import assert_equal, assert_items_equal, assert_greater
from mcok import Mock

Fixing with a pylint plugin

Luckily, there is a better way of solving this problem than disabling individual checks everywhere. pylint provides a plugin interface so that people can implement their own checks and plugins. Using this, we can also modify the way that the static analysis looks at the module. In this pylint plugin, I have implemented a function which transforms the abstract syntax tree which pylint uses when looking at to inject all of the functions which are generated at runtime:
from nose import tools

function_template = """
def {}(*args, **kwargs):

def nose_transform(module):
    funcs = ''.join(function_template.format(func_name)
                    for func_name in tools.__all__)
    fake = ASTNGBuilder(MANAGER).string_build(funcs)

    for func_name in tools.__all__:
        if func_name not in module.locals:
            module.locals[func_name] = fake[func_name]
Just follow the installation instructions in the gist, and all of your pylint problems will disappear!