Python Docstring Using Sphinx and reStructuredText


Python is a dynamic script programming language that has no static type checking. For Python, documentation becomes very important for developers to understand and use Python components such as modules, classes, functions and methods.

reStructuredText (abbreviated as reST) is an extensible markup language that is used to write most Python documents. Sphinx is a widely-used documentation tool that uses the reST for Python project and code documentation. Sphinx can convert reST documents into many different formats including HTML, PDF or even plain text. Sphinx add many extensions to reST language to produce documents or document web sites. The Sphinx reST memo has a brief introduction of reST and Sphinx.

PyCharm has good support for the reST syntax and Sphinx extensions. Together, they make it easy to document Python project. However, there is no good Python code document tutorial that gives a clear and easy-to-follow instructions on how to document Python code. This article is an attempt to fill this gap. The focus is how one write docstrings to document a module, a function, a class and its methods using the reStructuredText markup language. These docstrings are used by Sphinx to automatically generate project documents.

DocString Primer

According to the PEP 257: “A docstring is a string literal that occurs as the first statement in a module, function, class, or method definition. Such a docstring becomes the doc special attribute of that object.”

Because Python docstrings are embedded in a Python source code file, it should be short, completed and helpful. Docstrings only use a subset of the reStructuredText features. For example, it is usually not a good idea to use section titles or external links in a source code file. Only features that are commonly used in code documentation are covered in the following sections.

It suggests that all modules should have docstrings. Functions, classes exported by a module should have docstrings. All public methods and the constructor (__init__) of an exported class should also have docstrings. A package may be documented using a docstring it its file.

To be consistent, it is suggested to use """triple double quotes""" in docstring. If there is any backslashes, use `r”"”raw triple double quotes”””.

There are two types of docstrings: one-line and multi-line.

One-line Docstring

As the name means, a one-line docstring is a triple double quote string in one line. Additionally, there is no blank line before or after the one-line docstring. It should be a phrase prescribing the effect of a function or a method. It should have a period at the end. Following is an example of a one-line docstring:

def sync():
    """Sync cache record to backend database."""

It should only be used in very simple cases that a function or a method that has no parameter and return value. There is no space between the triple double quotes and the inside text.

Multi-line Docstrings

A multi-line docstring starts with a summary line like the one-line docstring, i.e., it is a prescribing phrase ended with a period. Then there is a blank line followed by one or mor paragraphs. The end triple double quote should be the last line by itself in a multi-line docstring.

A module’s docstring should list the classes, exceptions and functions exported by the module. Because a class or a function are separated by two blank lines, there should be two blank lines after a module’s docstring.

A package’s docstring should list the modules and subpackages exported by the package.

A class’s docstring should list its public methods and instance variables. Additional interfaces used for subclasses should be document in a separate paragraph. The docstring should be following by a blank line because all methods are separated by a blank line. Use standard words such as “override” and “extend” to describe subclass methods with regard to the super class methods.

All exported functions and public methods should have docstrings that document the effect, parameters and return values. For complicated functions/methods, code samples showing how to use them are recommended.

reStructuredText Primer

reST is a simple markup language for tex files. Like HTML, reST has the concepts of block (contents separated by a new line or other symbols), inline (markups embedded inline) and cross references (internal or external links). Block level constructors include sections, paragraphs, lists, and tables. A section consists of one or more blocks.

Being a markup language, reST needs markers to give special meanings or process instructions to a text file. reST has two extensible markup mechanisms: Directive and Interpreted Text.

A directive marker begins with .. (two periods and a space), followed by the directive type and two colons, then optionally a space and a directive title at the same line. Directives are block-level markers. For example, .. note::, and .. code:: python two directives.

An interpreted text has a general syntax of :role:`intepreted text`. The role determines how to interpretation process. It is often used to format or create a link for the interpreted text. A interpreted text is enclosed a pair of single-backquotes. It is often used in docstring to refer a Python element.

Following are standard reST constructors commonly used in docstrings.


A text chunk separated by one or more blank lines ia a paragraph. The indentation is significant in reST. Quoted paragraphs are created by indentation.

Inline Markup

The standard reST provides some simple inline markups such as emphasis: *test*, boldface: **test**, and code samples: ``some code``. The format markups such as emphasis and boldface should not be used in a docstring. The code sample markup is often used in a docstring to mark a class name, a function name, a method name or a variable name.

List Markup

Like markdown, reST uses *, #. and numbers to make a list. It supports nested lists using indentation. Following are some examples:

* Bullet list item 1
* Bullet list item 2
        * Nested Item 1
        * Nested Item 2
* Bullet list item 3

1. Numbered item 1
2. Numbered item 2

#. Automatically numbered item 1
#. Automatically numbered item 2

Source Code Block

A source code block is defined by ending its previous paragraph with ::. A source code block must be indented from its previous paragraph. A source code is also called a literal block because it is not processed in any way except that the indentation is removed.

There should not be any whitespace in front of the :: marker. If there is any whitespace preceding the :: marker, it becomes a single colon.

Sphinx Python Roles

reST uses interpreted text roles” to generate cross-references or format **interpreted texts such as emphasis, code literal. A role name specifies that an enclosed text should be interpreted in a specific way, either by pre-defined roles or an extension application. This is an extension mechanism of reST to support new tags (roles). The general syntax is :rolename:`target`. The target will be the link’s text. If the “target” is prefixed with a “!”, no reference/hyperlink is created. If the “target” is prefixed with at “~”, only the last component of the target will be used as the link text. Following are Python role names defined by Sphinx:

  • :py:mod:: a module; a dotted name may be used.
  • :py:func:: a function, a dotted name may be used.
  • :py:data:: a module-level variable.
  • :py:const:: a defined constant.
  • :py:class:: a class; a dotted name may be used.
  • :py:meth:: a method. The role text can include type name and the method name; a dotted name may be used.
  • :py:attr:: an attribute of an object.
  • :py:exc:: an exception; a dotted name may be used.
  • :py:obj:: used as the default role to reference an object of unspecified type.

Actually the inline markups in the previous section are shortcuts for their corresponding text roles of :emphasis:, :strong: and :literal:.

Other common Sphinx roles include abbr, file, manpage, regexp, and samp. Following are some examples:

:abbr:`RFC(request for comments)`	
:samp:`cp {file} {target}`	


Like an interpreted text role, a directive is another extension mechanism of reST to make a block of content should be handled in a specific way. A directive starts with tow dots and a space “.. “, followed by a type (name) and two colons, and a white space. These are called “directive marker”. After a directive maker are arguments, options and content. Arguments and options must form a contiguous block that are indented from the second line. The content must be preceded by a blank line.

reST has some built-in directives such as note, warning, seealso, code, etc. Following is a note example:

.. note:: A note admonition
    The second line and after must be indented. These are usually 
    for options for other directives but not used by the note directive.
    The content starts after a blank line.
    More contents...

Sphinx Domain

The Sphinx document gives the following definition of a Sphinx domain:

A domain is a collection of markup (reStructuredText directives and roles) 
to describe and link to objects belonging together, e.g. elements of 
a programming language. Directive and role names in a domain have 
names like domain:name, e.g. py:function. Domains can also 
provide custom indices (like the Python Module Index).

Therefore it is a language-specific feature to define programming language elements in a language namespace. For example, .. py:funciton:: get_name(id) is a Python domain directive that describes a function. Actually the :py:func: described before is a role in the same Python domain (named as py).

The Python domain (named as py) defines the following directives:

  • .. py:module:: name
  • .. py:exception:: name
  • .. py:currentmodule:: name
  • .. py:data:: name
  • .. py:function:: name(parameters)
  • .. py:class:: name
  • .. py:class:: name(parameters)
  • .. py:attribute:: name
  • .. py:method:: name(parameters)
  • .. py:staticmethod:: name(parameters)
  • .. py:classmethod:: name(parameters)
  • .. py:decorator:: name
  • .. py:decorator:: name(parameters)
  • .. py:decoratormethod:: name
  • .. py:decoratormethod:: name(parameters)

Some of these are useful in writing package, module and class docstrings.

Default Domain

Sphinx uses primary_domain configuration parameter to set default domain. If not set, the default is py. A default domain can be selected using the default_domain directive.

Directives and roles without a domain name use the default domain. For example, if the default domain is py, the following directive and role are in the py domain:

.. function:: my_func(parameters)

    function description

:func: `my_func` 

Sphinx Info Field Lists

Inside Python object directives or docstrings, Sphinx uses the following field list to rend a code document.

  • :param parameter_name: parameter description
  • :type parameter_name: parameter type
  • :returns: return value description
  • :rtype: return type
  • :raises: :exc:`ExceptionType`
  • :var:variable_name: variable description

There are different field names and syntax to write field lists. PyCharm uses the field names and syntax listed above. Below is an example from Sphinx doc

.. py:function:: send_message(sender, recipient, message_body, [priority=1])

   Send a message to a recipient

   :param str sender: The person sending the message
   :param str recipient: The recipient of the message
   :param str message_body: The body of the message
   :param priority: The priority of the message, can be a number 1-5
   :type priority: integer or None
   :return: the message id
   :rtype: int
   :raises ValueError: if the message_body exceeds 160 characters
   :raises TypeError: if the message_body is not a basestring

Markups in Docstring

Combining the requirements of docstrings with the markups in reST and Sphinx, we should document Python code using the following simple rules:

  • Document all modules, classes, functions, and public methods as suggested by the PEP 257.
  • Use paragraphs, lists, inline code, code block, and Python roles in a docstring.
  • All exported functions and public methods should have complete description of their parameter types and return types.
Written on January 23, 2015