Docutils | Overview | About | Users | Reference | Developers

Docutils Hacker's Guide


Lea Wiemann


You have used reStructuredText and played around with the Docutils front-end tools before. Some (basic) Python knowledge is certainly helpful (though not necessary, strictly speaking).


This is the introduction to Docutils for all persons who want to extend Docutils in some way.

Overview of the Docutils Architecture

To give you an understanding of the Docutils architecture, we'll dive right into the internals using a practical example.

Consider the following reStructuredText file:

My *favorite* language is Python_.

.. _Python:

Using the rst2html front-end tool, you would get an HTML output which looks like this:

[uninteresting HTML code removed]
<div class="document">
<p>My <em>favorite</em> language is <a class="reference" href="">Python</a>.</p>

While this looks very simple, it's enough to illustrate all internal processing stages of Docutils. Let's see how this document is processed from the reStructuredText source to the final HTML output:

Reading the Document

The Reader reads the document from the source file and passes it to the parser (see below). The default reader is the standalone reader (docutils/readers/ which just reads the input data from a single text file. Unless you want to do really fancy things, there is no need to change that.

Since you probably won't need to touch readers, we will just move on to the next stage:

Parsing the Document

The Parser analyzes the the input document and creates a node tree representation. In this case we are using the reStructuredText parser:

from docutils import frontend, utils
from docutils.parsers.rst import Parser
settings = frontend.get_default_settings(Parser)
with open('test.txt', encoding='utf-8') as file:
    document = utils.new_document(, settings)
    Parser().parse(, document)

Let us now examine the node tree. With print(document.pformat()) we get:

<document source="test.txt">
         language is
        <reference name="Python" refname="python">
    <target ids="python" names="python" refuri="">

The top-level node is document. It has a source attribute whose value is text.txt. There are two children: A paragraph node and a target node. The paragraph in turn has children: A text node ("My "), an emphasis node, a text node (" language is "), a reference node, and again a Text node (".").

These node types (document, paragraph, emphasis, etc.) are all defined in docutils/ The node types are internally arranged as a class hierarchy (for example, both emphasis and reference have the common superclass Inline). To get an overview of the node class hierarchy, use epydoc (type epydoc and look at the class hierarchy tree.

Transforming the Document

In the node tree above, the reference node does not contain the target URI ( yet.

Assigning the target URI (from the target node) to the reference node is not done by the parser (the parser only translates the input document into a node tree).

Instead, it's done by a Transform. In this case (resolving a reference), it's done by the ExternalTargets transform in docutils/transforms/

The Transforms are applied after parsing. To see how the node tree has changed after applying the Transforms, we use the rst2pseudoxml tool [1]:

$ rst2pseudoxml test.txt
<document source="test.txt">
         language is
        <reference name="Python" **refuri=""**>
    <target ids="python" names="python" ``refuri=""``>

For our small test document, the only change is that the refname attribute of the reference has been replaced by a refuri attribute—the reference has been resolved.

While this does not look very exciting, transforms are a powerful tool to apply any kind of transformation on the node tree. In fact, there are quite a lot of Transforms, which do various useful things like creating the table of contents, applying substitution references or resolving auto-numbered footnotes. For details, see Docutils Transforms.

Writing the Document

To get an HTML document out of the node tree, we use a Writer, the HTML writer in this case (docutils/writers/

The writer receives the node tree and returns the output document. For HTML output, we can test this using the rst2html tool:

$ rst2html --link-stylesheet test.txt
<?xml version="1.0" encoding="utf-8" ?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "">
<html xmlns="" xml:lang="en" lang="en">
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<meta name="generator" content="Docutils 0.3.10:" />
<link rel="stylesheet" href="../docutils/writers/html4css1/html4css1.css" type="text/css" />
<div class="document">
<p>My <em>favorite</em> language is <a class="reference" href="">Python</a>.</p>

So here we finally have our HTML output. The actual document contents are in the fourth-last line. Note, by the way, that the HTML writer did not render the (invisible) target node—only the paragraph node and its children appear in the HTML output.

Extending Docutils

Now you'll ask, "how do I actually extend Docutils?"

First of all, once you are clear about what you want to achieve, you have to decide where to implement it—in the Parser (e.g. by adding a directive or role to the reStructuredText parser), as a Transform, or in the Writer. There is often one obvious choice among those three (Parser, Transform, Writer). If you are unsure, ask on the Docutils-develop mailing list.

In order to find out how to start, it is often helpful to look at similar features which are already implemented. For example, if you want to add a new directive to the reStructuredText parser, look at the implementation of a similar directive in docutils/parsers/rst/directives/.

Modifying the Document Tree Before It Is Written

You can modify the document tree right before the writer is called. One possibility is to use the publish_doctree and publish_from_doctree functions.

To retrieve the document tree, call:

document = docutils.core.publish_doctree(...)

Please see the docstring of publish_doctree for a list of parameters.

document is the root node of the document tree. You can now change the document by accessing the document node and its children—see The Node Interface below.

When you're done with modifying the document tree, you can write it out by calling:

output = docutils.core.publish_from_doctree(document, ...)

The Node Interface

As described in the overview above, Docutils' internal representation of a document is a tree of nodes. We'll now have a look at the interface of these nodes.

(To be completed.)

What Now?

This document is not complete. Many topics could (and should) be covered here. To find out with which topics we should write about first, we are awaiting your feedback. So please ask your questions on the Docutils-develop mailing list.