This extension allows you to use Python 3 annotations for documenting acceptable argument types and return value types of functions. See an example of the Sphinx render at the pyproject-api docs.
This allows you to use type hints in a very natural fashion, allowing you to migrate from this:
def format_unit(value, unit):
"""
Formats the given value as a human readable string using the given units.
:param float|int value: a numeric value
:param str unit: the unit for the value (kg, m, etc.)
:rtype: str
"""
return f"{value} {unit}"
to this:
from typing import Union
def format_unit(value: Union[float, int], unit: str) -> str:
"""
Formats the given value as a human readable string using the given units.
:param value: a numeric value
:param unit: the unit for the value (kg, m, etc.)
"""
return f"{value} {unit}"
First, use pip to download and install the extension:
pip install sphinx-autodoc-typehints
Then, add the extension to your conf.py
:
extensions = ["sphinx.ext.autodoc", "sphinx_autodoc_typehints"]
The following configuration options are accepted:
-
typehints_fully_qualified
(default:False
): ifTrue
, class names are always fully qualified (e.g.module.for.Class
). IfFalse
, just the class name displays (e.g.Class
) -
always_document_param_types
(default:False
): IfFalse
, do not add type info for undocumented parameters. IfTrue
, add stub documentation for undocumented parameters to be able to add type info. -
typehints_document_rtype
(default:True
): IfFalse
, never add an:rtype:
directive. IfTrue
, add the:rtype:
directive if no existing:rtype:
is found. -
typehints_use_rtype
(default:True
): Controls behavior whentypehints_document_rtype
is set toTrue
. IfTrue
, document return type in the:rtype:
directive. IfFalse
, document return type as part of the:return:
directive, if present, otherwise fall back to using:rtype:
. Use in conjunction with napoleon_use_rtype to avoid generation of duplicate or redundant return type information. -
typehints_defaults
(default:None
): IfNone
, defaults are not added. Otherwise, adds a default annotation:'comma'
adds it after the type, changing Sphinx’ default look to “param (int, default:1
) -- text”.'braces'
adds(default: ...)
after the type (useful for numpydoc like styles).'braces-after'
adds(default: ...)
at the end of the parameter documentation text instead.
-
simplify_optional_unions
(default:True
): IfTrue
, optional parameters of type "Union[...]" are simplified as being of type Union[..., None] in the resulting documentation (e.g. Optional[Union[A, B]] -> Union[A, B, None]). IfFalse
, the "Optional"-type is kept. Note: IfFalse
, any Union containingNone
will be displayed as Optional! Note: If an optional parameter has only a single type (e.g Optional[A] or Union[A, None]), it will always be displayed as Optional! -
typehints_formatter
(default:None
): If set to a function, this function will be called withannotation
as first argument andsphinx.config.Config
argument second. The function is expected to return a string with reStructuredText code orNone
to fall back to the default formatter. -
typehints_use_signature
(default:False
): IfTrue
, typehints for parameters in the signature are shown. -
typehints_use_signature_return
(default:False
): IfTrue
, return annotations in the signature are shown.
The extension listens to the autodoc-process-signature
and autodoc-process-docstring
Sphinx events. In the former,
it strips the annotations from the function signature. In the latter, it injects the appropriate :type argname:
and
:rtype:
directives into the docstring.
Only arguments that have an existing :param:
directive in the docstring get their respective :type:
directives
added. The :rtype:
directive is added if and only if no existing :rtype:
is found.
To use sphinx.ext.napoleon with sphinx-autodoc-typehints, make sure you load sphinx.ext.napoleon first, before sphinx-autodoc-typehints. See Issue 15 on the issue tracker for more information.
Sometimes functions or classes from two different modules need to reference each other in their type annotations. This creates a circular import problem. The solution to this is the following:
- Import only the module, not the classes/functions from it
- Use forward references in the type annotations (e.g.
def methodname(self, param1: 'othermodule.OtherClass'):
)
On Python 3.7, you can even use from __future__ import annotations
and remove the quotes.
If you're documenting code that needs to stay compatible with Python 2.7, you cannot use regular type annotations.
Instead, you must either be using Python 3.8 or later or have typed_ast
installed. The package extras type_comments
will pull in the appropriate dependencies automatically. Then you can add
type hint comments in the following manner:
def myfunction(arg1, arg2):
# type: (int, str) -> int
return 42
or alternatively:
def myfunction(
arg1, # type: int
arg2, # type: str
):
# type: (...) -> int
return 42