Transform plugins

Why write a plugin?

Pylint is a static analysis tool and Python is a dynamically typed language. So there will be cases where Pylint cannot analyze files properly (this problem can happen in statically typed languages also if reflection or dynamic evaluation is used).

The plugins are a way to tell Pylint how to handle such cases, since only the user would know what needs to be done. They are usually operating on the AST level, by modifying or changing it in a way which can ease its understanding by Pylint.

Example

Let us run Pylint on a module from the Python source: warnings.py and see what happens:

amitdev$ pylint -E Lib/warnings.py
E:297,36: Instance of 'WarningMessage' has no 'message' member (no-member)
E:298,36: Instance of 'WarningMessage' has no 'filename' member (no-member)
E:298,51: Instance of 'WarningMessage' has no 'lineno' member (no-member)
E:298,64: Instance of 'WarningMessage' has no 'line' member (no-member)

Did we catch a genuine error? Let's open the code and look at WarningMessage class:

class WarningMessage(object):

  """Holds the result of a single showwarning() call."""

  _WARNING_DETAILS = ("message", "category", "filename", "lineno", "file",
                      "line")

  def __init__(self, message, category, filename, lineno, file=None,
                  line=None):
    local_values = locals()
    for attr in self._WARNING_DETAILS:
      setattr(self, attr, local_values[attr])
    self._category_name = category.__name__ if category else None

  def __str__(self):
    ...

Ah, the fields (message, category etc) are not defined statically on the class. Instead they are added using setattr. Pylint would have a tough time figuring this out.

Enter Plugin

We can write a transform plugin to tell Pylint how to analyze this properly.

One way to fix our example with a plugin would be to transform the WarningMessage class, by setting the attributes so that Pylint can see them. This can be done by registering a transform function. We can transform any node in the parsed AST like Module, Class, Function etc. In our case we need to transform a class. It can be done so:

from typing import TYPE_CHECKING

import astroid

if TYPE_CHECKING:
    from pylint.lint import PyLinter


def register(linter: "PyLinter") -> None:
  """This required method auto registers the checker during initialization.

  :param linter: The linter to register the checker to.
  """
  pass

def transform(cls):
  if cls.name == 'WarningMessage':
    import warnings
    for f in warnings.WarningMessage._WARNING_DETAILS:
      cls.locals[f] = [astroid.ClassDef(f, None)]

astroid.MANAGER.register_transform(astroid.ClassDef, transform)

Let's go through the plugin. First, we need to register a class transform, which is done via the register_transform function in MANAGER. It takes the node type and function as parameters. We need to change a class, so we use astroid.ClassDef. We also pass a transform function which does the actual transformation.

transform function is simple as well. If the class is WarningMessage then we add the attributes to its locals (we are not bothered about type of attributes, so setting them as class will do. But we could set them to any type we want). That's it.

Note: We don't need to do anything in the register function of the plugin since we are not modifying anything in the linter itself.

Lets run Pylint with this plugin and see:

amitdev$ pylint -E --load-plugins warning_plugin Lib/warnings.py
amitdev$

All the false positives associated with WarningMessage are now gone. This is just an example, any code transformation can be done by plugins.

See astroid/brain for real life examples of transform plugins.