Tutorial

When you want to force someone to write the code the way you want: you need to create a rule for that.

There are multiple options of how this can be done. This guide will walk trough all possible cases and cover every decision path.

Deciding what exactly to write

The most important thing is the question: what kind of rule do you want to create?

Depending on the answer you can end up with either:

  1. Creating a new flake8 plugin

  2. Creating a pair of new visitor and violation inside this plugin, and some checking logic to find problems with your code

  3. Just a new violation and checking logic to find problems with your code

What does it depend on internally?

Writing new plugin

First of all, you have to decide:

  1. Are you writing a separate plugin and adding it as a dependency?

  2. Are you writing a built-in extension to this styleguide?

How to make a decision?

Will this plugin be useful to other developers without this styleguide?

If so, it would be wise to create a separate flake8 plugin. Then you can add newly created plugin as a dependency. Our rules do not make any sense without each other.

It is also useful when you try to wrap an existing tool into flake8 API.

Real world examples of tools that are useful by them self:

Can this plugin be used with the existing checker?

flake8 has a very strict API about plugins. Here are some problems that you may encounter:

  • Some plugins are called once per file, some are called once per line

  • Plugins should define clear violation code / checker relation

  • It is impossible to use the same letter violation codes for several checkers

So, if you want a plugin to work with each logical line - you have to create a custom plugin.

Real world examples of plugins unsuitable for this checker:

Is this rule out off scope?

There are awesome tools that can not be added because they are just simply out of scope. This means that they cover very specific case or technology and not just good-old python.

Real world examples of plugins that are out of scope:

All these plugins should be installed individually to the end-user dependencies. And only when user really want it. So, it is up to the user to decide.

And these plugins while being awesome won’t be added to our project at all.

Conclusion

If you said “yes” to any of these question - write a plugin. Then possibly add it as a dependency to this project.

Writing new visitor

If you are still willing to write a builtin extension to this project, you will have to write a violation and/or visitor.

First of all, you have to decide what base class do you want to use?

There are several possibilities:

graph TD; BaseTokenVisitor BaseVisitor BaseNodeVisitor BaseFilenameVisitor NodeVisitor --> BaseNodeVisitor BaseVisitor --> BaseNodeVisitor BaseVisitor --> BaseFilenameVisitor BaseVisitor --> BaseTokenVisitor

When to choose what base class? Imagine that you have several ideas in mind:

  1. I want to lint module names not to contain numbers

  2. I want to lint code not to contain number 3

  3. I want to lint code to disallow multiplication of exactly two number

Each of these tasks will require different approaches.

  1. Will require to subclass a filename-based visitor

  2. Will require to subclass a tokenize-based visitor

  3. Will require to subclass a ast-based visitor

How to differ these cases by yourself?

  1. You need to read though the docs of ast and tokenize modules

  2. You can have a look at the existing visitors

But, you might not want to write a new visitor. You can reuse existing ones and write only a violation and checking logic.

Technical documentation about the Visitors API is available.

Writing new violation

The only thing you should care about is to select the correct base class for new violation.

graph TD; MaybeASTViolation TokenizeViolation ASTViolation _BaseASTViolation BaseViolation SimpleViolation BaseViolation --> _BaseASTViolation _BaseASTViolation --> ASTViolation _BaseASTViolation --> MaybeASTViolation BaseViolation --> TokenizeViolation BaseViolation --> SimpleViolation

It only depends on already selected visitor type, so you won’t have to make this decision twice.

Technical documentation about the Violations API is available.

Writing business logic

When you will have your visitor and violation it will be required to actually write some logic to raise a violation from visitor.

We do this inside the visitor, but we create protected methods and place logic there.

Consider this example:

class WrongComprehensionVisitor(BaseNodeVisitor):
    _max_ifs = 1

    def _check_ifs(self, node: ast.comprehension) -> None:
        if len(node.ifs) > self._max_ifs:
            # This will restrict to have more than 1 `if`
            # in your comprehensions:
            self.add_violation(MultipleIfsInComprehensionViolation(node))

    def visit_comprehension(self, node: ast.comprehension) -> None:
        self._check_ifs(node)
        self.generic_visit(node)

You may also end up using the same logic over and over again. In this case we can decouple it and move to logics/ package.

Then it would be easy to reuse something.

Writing tests

Writing end-to-end tests

In end-to-end tests we check that our visitor, violation and business logic work correctly together all the way from flake8 config file to its output.

To check all supported violations, we have two modules containing code which raises them: noqa.py and noqa_controlled.py. The first is for all possible violations while the second is only for those which may be tweaked using i_control_code option. If violation may be ignored (or istead, raised) with i_control_code, the appropriate piece of code should be added to both modules.

The next thing is test itself which should reside in tests/test_checker/test_noqa.py module. The main test functions are written already, so probably the only thing to do is to put the violation code into either SHOULD_BE_RAISED, or SHOULD_BE_RAISED_NO_CONTROL container, or into both. For example, if the violation is raised with i_control_code=True, it must be placed into SHOULD_BE_RAISED with value 1 and into SHOULD_BE_RAISED_NO_CONTROL with value 0. By doing this we check that the violation is raised in one situation and is not raised in opposite one. If the violation is ignored when i_control_code=True, swap 0 and 1 it containers. If the violation cannot be tweaked with i_control_code it should only be put into SHOULD_BE_RAISED container with appropriate value.