docstring-inheritance is a python package to avoid writing and maintaining duplicated python docstrings. The typical usage is to enable the inheritance of the docstrings from a base class such that its derived classes fully or partly inherit the docstrings.

Features

• Handle numpy and google docstring formats (i.e. sections based docstrings):
  • NumPy docstring format specification
  • Google docstring format specification
• Handle docstrings for functions, classes, methods, class methods, static methods, properties.
• Handle docstrings for classes with multiple or multi-level inheritance.
• Docstring sections are inherited individually, like methods.
• For docstring sections documenting signatures, the signature arguments are inherited individually.
• Minimum performance cost: the inheritance is performed at import time, not for each call.
• Compatible with rendering the documentation with Sphinx.
• Missing docstring sections for signature arguments are notified by warnings.
• Docstring sections can be compared to detect duplicated or similar contents that could be inherited.

Licenses

The source code is distributed under the MIT license. The documentation is distributed under the CC BY 4.0 license. The dependencies, with their licenses, are given in the CREDITS.md file.

Installation

Install with pip:

pip install docstring-inheritance

Or with conda:

conda install -c conda-forge docstring-inheritance

Basic Usage

Inheriting docstrings for classes

docstring-inheritance provides metaclasses to enable the docstrings of a class to be inherited from its base classes. This feature is automatically transmitted to its derived classes as well. The docstring inheritance is performed for the docstrings of the:

• class
• methods
• classmethods
• staticmethods
• properties

Use the NumpyDocstringInheritanceMeta metaclass to inherit docstrings in numpy format if __init__ method is documented in its own docstring. Otherwise, if __init__ method is documented in the class docstring, use the NumpyDocstringInheritanceInitMeta metaclass.

Use the GoogleDocstringInheritanceMeta metaclass to inherit docstrings in google format. if __init__ method is documented in its own docstring. Otherwise, if __init__ method is documented in the class docstring, use the GoogleDocstringInheritanceInitMeta metaclass.

from docstring_inheritance import NumpyDocstringInheritanceMeta


class Parent(metaclass=NumpyDocstringInheritanceMeta):
  def method(self, x, y=None):
    """Parent summary.

    Parameters
    ----------
    x:
       Description for x.
    y:
       Description for y.

    Notes
    -----
    Parent notes.
    """


class Child(Parent):
  def method(self, x, z):
    """
    Parameters
    ----------
    z:
       Description for z.

    Returns
    -------
    Something.

    Notes
    -----
    Child notes.
    """


# The inherited docstring is
Child.method.__doc__ == """Parent summary.

Parameters
----------
x:
   Description for x.
z:
   Description for z.

Returns
-------
Something.

Notes
-----
Child notes.
"""

Inheriting docstrings for functions

docstring-inheritance provides functions to inherit the docstring of a callable from a string. This is typically used to inherit the docstring of a function from another function.

Use the inherit_google_docstring function to inherit docstrings in google format.

Use the inherit_numpy_docstring function to inherit docstrings in numpy format.

from docstring_inheritance import inherit_google_docstring


def parent():
    """Parent summary.

    Args:
        x: Description for x.
        y: Description for y.

    Notes:
        Parent notes.
    """


def child():
    """
    Args:
        z: Description for z.

    Returns:
        Something.

    Notes:
        Child notes.
    """


inherit_google_docstring(parent.__doc__, child)

# The inherited docstring is
child.__doc__ == """Parent summary.

Args:
    x: Description for x.
    z: Description for z.

Returns:
    Something.

Notes:
    Child notes.
"""

Docstring inheritance specification

Sections order

The sections of an inherited docstring are sorted according to order defined in the NumPy docstring format specification:

• Summary
• Extended summary
• Parameters for the NumPy format or Args for the Google format
• Returns
• Yields
• Receives
• Other Parameters
• Attributes
• Methods
• Raises
• Warns
• Warnings
• See Also
• Notes
• References
• Examples
• sections with other names come next

This ordering is also used for the docstring written with the Google docstring format specification even though it does not define all of these sections.

Sections with items

Those sections are:

• Other Parameters
• Methods
• Attributes

The inheritance is done at the key level, i.e. a section of the inheritor will not fully override the parent one:

• the keys in the parent section and not in the child section are inherited,
• the keys in the child section and not in the parent section are kept,
• for keys that are both in the parent and child section, the child ones are kept.

This allows to only document the new keys in such a section of an inheritor. For instance:

from docstring_inheritance import NumpyDocstringInheritanceMeta


class Parent(metaclass=NumpyDocstringInheritanceMeta):
  """
  Attributes
  ----------
  x:
     Description for x
  y:
     Description for y
  """


class Child(Parent):
  """
  Attributes
  ----------
  y:
     Overridden description for y
  z:
     Description for z
  """


# The inherited docstring is
Child.__doc__ == """
Attributes
----------
x:
   Description for x
y:
   Overridden description for y
z:
   Description for z
"""

Here the keys are the attribute names. The description for the attribute y has been overridden and the description for the attribute z has been added. The only remaining description from the parent is for the attribute x.

Sections documenting signatures

Those sections are:

• Parameters (numpy format only)
• Args (google format only)

In addition to the inheritance behavior described above:

• the arguments not existing in the inheritor signature are removed,
• the arguments are sorted according the inheritor signature,
• the arguments with no description are provided with a dummy description.

from docstring_inheritance import GoogleDocstringInheritanceMeta


class Parent(metaclass=GoogleDocstringInheritanceMeta):
  def method(self, w, x, y):
    """
    Args:
        w: Description for w
        x: Description for x
        y: Description for y
    """


class Child(Parent):
  def method(self, w, y, z):
    """
    Args:
        z: Description for z
        y: Overridden description for y
    """


# The inherited docstring is
Child.method.__doc__ == """
Args:
    w: Description for w
    y: Overridden description for y
    z: Description for z
"""

Here the keys are the argument names. The description for the argument y has been overridden and the description for the argument z has been added. The only remaining description from the parent is for the argument w.

Advanced usage

Abstract base class

To create a parent class that both is abstract and has docstring inheritance, an additional metaclass is required:

import abc
from docstring_inheritance import NumpyDocstringInheritanceMeta


class Meta(abc.ABCMeta, NumpyDocstringInheritanceMeta):
  pass


class Parent(metaclass=Meta):
  pass

Detecting similar docstrings

Duplicated docstrings that could benefit from inheritance can be detected by setting the environment variable DOCSTRING_INHERITANCE_SIMILARITY_RATIO to a value between 0 and 1. When set, the docstring sections of a child and its parent are compared and warnings are issued when the docstrings are similar. The docstring sections are compared with difflib ratio from the standard library. If the ratio is higher or equal to the value of DOCSTRING_INHERITANCE_SIMILARITY_RATIO, the docstring sections are considered similar. Use a ratio of 1 to detect identical docstring sections. Use a ratio lower than 1 to detect similar docstring sections.

Similar projects

custom_inherit: docstring-inherit started as fork of this project before being re-written, we thank its author.