Creating reStructuredText Directives

Authors: Dethe Elza
David Goodger
Lea Wiemann
Date: 2012-01-03
Revision: 7302
Copyright: This document has been placed in the public domain.

Directives are the primary extension mechanism of reStructuredText. This document aims to make the creation of new directives as easy and understandable as possible. There are only a couple of reStructuredText-specific features the developer needs to know to create a basic directive.

The syntax of directives is detailed in the reStructuredText Markup Specification, and standard directives are described in reStructuredText Directives.

Directives are a reStructuredText markup/parser concept. There is no "directive" document tree element, no single element that corresponds exactly to the concept of directives. Instead, choose the most appropriate elements from the existing Docutils elements. Directives build structures using the existing building blocks. See The Docutils Document Tree and the docutils.nodes module for more about the building blocks of Docutils documents.

Table of Contents

The Directive Class

Directives are created by defining a directive class that inherits from docutils.parsers.rst.Directive:

from docutils.parsers import rst

class MyDirective(rst.Directive):


To understand how to implement the directive, let's have a look at the docstring of the Directive base class:

>>> from docutils.parsers import rst
>>> print rst.Directive.__doc__

    Base class for reStructuredText directives.

    The following attributes may be set by subclasses.  They are
    interpreted by the directive parser (which runs the directive

    - `required_arguments`: The number of required arguments (default:

    - `optional_arguments`: The number of optional arguments (default:

    - `final_argument_whitespace`: A boolean, indicating if the final
      argument may contain whitespace (default: False).

    - `option_spec`: A dictionary, mapping known option names to
      conversion functions such as `int` or `float` (default: {}, no
      options).  Several conversion functions are defined in the
      directives/ module.

      Option conversion functions take a single parameter, the option
      argument (a string or ``None``), validate it and/or convert it
      to the appropriate form.  Conversion functions may raise
      `ValueError` and `TypeError` exceptions.

    - `has_content`: A boolean; True if content is allowed.  Client
      code must handle the case where content is required but not
      supplied (an empty content list will be supplied).

    Arguments are normally single whitespace-separated words.  The
    final argument may contain whitespace and/or newlines if
    `final_argument_whitespace` is True.

    If the form of the arguments is more complex, specify only one
    argument (either required or optional) and set
    `final_argument_whitespace` to True; the client code must do any
    context-sensitive parsing.

    When a directive implementation is being run, the directive class
    is instantiated, and the `run()` method is executed.  During
    instantiation, the following instance variables are set:

    - ``name`` is the directive type or name (string).

    - ``arguments`` is the list of positional arguments (strings).

    - ``options`` is a dictionary mapping option names (strings) to
      values (type depends on option conversion functions; see
      `option_spec` above).

    - ``content`` is a list of strings, the directive content line by line.

    - ``lineno`` is the line number of the first line of the directive.

    - ``content_offset`` is the line offset of the first line of the content from
      the beginning of the current input.  Used when initiating a nested parse.

    - ``block_text`` is a string containing the entire directive.

    - ``state`` is the state which called the directive function.

    - ``state_machine`` is the state machine which controls the state which called
      the directive function.

    Directive functions return a list of nodes which will be inserted
    into the document tree at the point where the directive was
    encountered.  This can be an empty list if there is nothing to

    For ordinary directives, the list must contain body elements or
    structural elements.  Some directives are intended specifically
    for substitution definitions, and must return a list of `Text`
    nodes and/or inline elements (suitable for inline insertion, in
    place of the substitution reference).  Such directives must verify
    substitution definition context, typically using code like this::

        if not isinstance(state, states.SubstitutionDef):
            error = state_machine.reporter.error(
                'Invalid context: the "%s" directive can only be used '
                'within a substitution definition.' % (name),
                nodes.literal_block(block_text, block_text), line=lineno)
            return [error]


Option Conversion Functions

An option specification (Directive.option_spec) must be defined detailing the options available to the directive. An option spec is a mapping of option name to conversion function; conversion functions are applied to each option value to check validity and convert them to the expected type. Python's built-in conversion functions are often usable for this, such as int, float. Other useful conversion functions are included in the docutils.parsers.rst.directives package (in the module):

A further utility function, choice, is supplied to enable options whose argument must be a member of a finite set of possible values. A custom conversion function must be written to use it. For example:

from docutils.parsers.rst import directives

def yesno(argument):
    return directives.choice(argument, ('yes', 'no'))

For example, here is an option spec for a directive which allows two options, "name" and "value", each with an option argument:

option_spec = {'name': unchanged, 'value': int}

Error Handling

If your directive implementation encounters an error during processing, you should call self.error() inside the run() method:

if error_condition:
    raise self.error('Error message.')

The self.error() method will immediately raise an exception that will be caught by the reStructuredText directive handler. The directive handler will then insert an error-level system message in the document at the place where the directive occurred.

Instead of self.error, you can also use self.severe and self.warning for more or less severe problems.

If you want to return a system message and document contents, you need to create the system message yourself instead of using the self.error convenience method:

def run(self):
    # Create node(s).
    node = nodes.paragraph(...)
    # Node list to return.
    node_list = [node]
    if error_condition:
         # Create system message.
         error = self.reporter.error(
             'Error in "%s" directive: Your error message.' %,
             nodes.literal_block(block_text, block_text), line=lineno)
    return node_list

Register the Directive


For the most direct and accurate information, "Use the Source, Luke!". All standard directives are documented in reStructuredText Directives, and the source code implementing them is located in the docutils/parsers/rst/directives package. The module contains a mapping of directive name to module and function name. Several representative directives are described below.


Admonition directives, such as "note" and "caution", are quite simple. They have no directive arguments or options. Admonition directive content is interpreted as ordinary reStructuredText.

The resulting document tree for a simple reStructuredText line ".. note:: This is a note." looks as follows:

This is a note.

The directive class for the "note" directive simply derives from a generic admonition directive class:

class Note(BaseAdmonition):

    node_class = nodes.note

Note that the only thing distinguishing the various admonition directives is the element (node class) generated. In the code above, the node class is set as a class attribute and is read by the run() method of BaseAdmonition, where the actual processing takes place:

# Import Docutils document tree nodes module.
from docutils import nodes
# Import Directive base class.
from docutils.parsers.rst import Directive

class BaseAdmonition(Directive):

    required_arguments = 0
    optional_arguments = 0
    final_argument_whitespace = True
    option_spec = {}
    has_content = True

    node_class = None
    """Subclasses must set this to the appropriate admonition node class."""

    def run(self):
        # Raise an error if the directive does not have contents.
        text = '\n'.join(self.content)
        # Create the admonition node, to be populated by `nested_parse`.
        admonition_node = self.node_class(rawsource=text)
        # Parse the directive contents.
        self.state.nested_parse(self.content, self.content_offset,
        return [admonition_node]

Three things are noteworthy in the run() method above:

  • The admonition_node = self.node_class(text) line creates the wrapper element, using the class set by the specific admonition subclasses (as in note, node_class = nodes.note).
  • The call to state.nested_parse() is what does the actual processing. It parses the directive content and adds any generated elements as child elements of admonition_node.
  • If there was no directive content, the assert_has_content() convenience method raises an error exception by calling self.error() (see Error Handling above).


The "image" directive is used to insert a picture into a document. This directive has one argument, the path to the image file, and supports several options. There is no directive content. Here's an early version of the image directive class:

# Import Docutils document tree nodes module.
from docutils import nodes
# Import ``directives`` module (contains conversion functions).
from docutils.parsers.rst import directives
# Import Directive base class.
from docutils.parsers.rst import Directive

def align(argument):
    """Conversion function for the "align" option."""
    return directives.choice(argument, ('left', 'center', 'right'))

class Image(Directive):

    required_arguments = 1
    optional_arguments = 0
    final_argument_whitespace = True
    option_spec = {'alt': directives.unchanged,
                   'height': directives.nonnegative_int,
                   'width': directives.nonnegative_int,
                   'scale': directives.nonnegative_int,
                   'align': align,
    has_content = False

    def run(self):
        reference = directives.uri(self.arguments[0])
        self.options['uri'] = reference
        image_node = nodes.image(rawsource=self.block_text,
        return [image_node]

Several things are noteworthy in the code above:

  • The "image" directive requires a single argument, which is allowed to contain whitespace (final_argument_whitespace = True). This is to allow for long URLs which may span multiple lines. The first line of the run() method joins the URL, discarding any embedded whitespace.
  • The reference is added to the options dictionary under the "uri" key; this becomes an attribute of the nodes.image element object. Any other attributes have already been set explicitly in the reStructuredText source text.

The Pending Element

Directives that cause actions to be performed after the complete document tree has been generated can be implemented using a pending node. The pending node causes a transform to be run after the document has been parsed.

For an example usage of the pending node, see the implementation of the contents directive in