Developing a “TODO” extension

The objective of this tutorial is to create a more comprehensive extension than that created in Developing a “Hello world” extension. Whereas that guide just covered writing a custom directive, this guide adds multiple directives, along with custom nodes, additional config values and custom event handlers. To this end, we will cover a todo extension that adds capabilities to include todo entries in the documentation, and to collect these in a central place. This is similar the sphinxext.todo extension distributed with Sphinx.



To understand the design of this extension, refer to Important objects and Build Phases.

We want the extension to add the following to Sphinx:

  • A todo directive, containing some content that is marked with “TODO” and only shown in the output if a new config value is set. Todo entries should not be in the output by default.

  • A todolist directive that creates a list of all todo entries throughout the documentation.

For that, we will need to add the following elements to Sphinx:

  • New directives, called todo and todolist.

  • New document tree nodes to represent these directives, conventionally also called todo and todolist. We wouldn’t need new nodes if the new directives only produced some content representable by existing nodes.

  • A new config value todo_include_todos (config value names should start with the extension name, in order to stay unique) that controls whether todo entries make it into the output.

  • New event handlers: one for the doctree-resolved event, to replace the todo and todolist nodes, one for env-merge-info to merge intermediate results from parallel builds, and one for env-purge-doc (the reason for that will be covered later).


As with Developing a “Hello world” extension, we will not be distributing this plugin via PyPI so once again we need a Sphinx project to call this from. You can use an existing project or create a new one using sphinx-quickstart.

We assume you are using separate source (source) and build (build) folders. Your extension file could be in any folder of your project. In our case, let’s do the following:

  1. Create an _ext folder in source

  2. Create a new Python file in the _ext folder called

Here is an example of the folder structure you might obtain:

└── source
    ├── _ext
    │   └──
    ├── _static
    ├── somefolder
    ├── index.rst
    ├── somefile.rst
    └── someotherfile.rst

Writing the extension

Open and paste the following code in it, all of which we will explain in detail shortly:

  1from docutils import nodes
  2from docutils.parsers.rst import Directive
  4from sphinx.application import Sphinx
  5from sphinx.locale import _
  6from sphinx.util.docutils import SphinxDirective
  7from sphinx.util.typing import ExtensionMetadata
 10class todo(nodes.Admonition, nodes.Element):
 11    pass
 14class todolist(nodes.General, nodes.Element):
 15    pass
 18def visit_todo_node(self, node):
 19    self.visit_admonition(node)
 22def depart_todo_node(self, node):
 23    self.depart_admonition(node)
 26class TodolistDirective(Directive):
 27    def run(self):
 28        return [todolist('')]
 31class TodoDirective(SphinxDirective):
 32    # this enables content in the directive
 33    has_content = True
 35    def run(self):
 36        targetid = 'todo-%d' % self.env.new_serialno('todo')
 37        targetnode ='', '', ids=[targetid])
 39        todo_node = todo('\n'.join(self.content))
 40        todo_node += nodes.title(_('Todo'), _('Todo'))
 41        self.state.nested_parse(self.content, self.content_offset, todo_node)
 43        if not hasattr(self.env, 'todo_all_todos'):
 44            self.env.todo_all_todos = []
 46        self.env.todo_all_todos.append({
 47            'docname': self.env.docname,
 48            'lineno': self.lineno,
 49            'todo': todo_node.deepcopy(),
 50            'target': targetnode,
 51        })
 53        return [targetnode, todo_node]
 56def purge_todos(app, env, docname):
 57    if not hasattr(env, 'todo_all_todos'):
 58        return
 60    env.todo_all_todos = [todo for todo in env.todo_all_todos if todo['docname'] != docname]
 63def merge_todos(app, env, docnames, other):
 64    if not hasattr(env, 'todo_all_todos'):
 65        env.todo_all_todos = []
 66    if hasattr(other, 'todo_all_todos'):
 67        env.todo_all_todos.extend(other.todo_all_todos)
 70def process_todo_nodes(app, doctree, fromdocname):
 71    if not app.config.todo_include_todos:
 72        for node in doctree.findall(todo):
 73            node.parent.remove(node)
 75    # Replace all todolist nodes with a list of the collected todos.
 76    # Augment each todo with a backlink to the original location.
 77    env = app.builder.env
 79    if not hasattr(env, 'todo_all_todos'):
 80        env.todo_all_todos = []
 82    for node in doctree.findall(todolist):
 83        if not app.config.todo_include_todos:
 84            node.replace_self([])
 85            continue
 87        content = []
 89        for todo_info in env.todo_all_todos:
 90            para = nodes.paragraph()
 91            filename = env.doc2path(todo_info['docname'], base=None)
 92            description = _(
 93                '(The original entry is located in %s, line %d and can be found '
 94            ) % (filename, todo_info['lineno'])
 95            para += nodes.Text(description)
 97            # Create a reference
 98            newnode = nodes.reference('', '')
 99            innernode = nodes.emphasis(_('here'), _('here'))
100            newnode['refdocname'] = todo_info['docname']
101            newnode['refuri'] = app.builder.get_relative_uri(fromdocname, todo_info['docname'])
102            newnode['refuri'] += '#' + todo_info['target']['refid']
103            newnode.append(innernode)
104            para += newnode
105            para += nodes.Text('.)')
107            # Insert into the todolist
108            content.extend((
109                todo_info['todo'],
110                para,
111            ))
113        node.replace_self(content)
116def setup(app: Sphinx) -> ExtensionMetadata:
117    app.add_config_value('todo_include_todos', False, 'html')
119    app.add_node(todolist)
120    app.add_node(
121        todo,
122        html=(visit_todo_node, depart_todo_node),
123        latex=(visit_todo_node, depart_todo_node),
124        text=(visit_todo_node, depart_todo_node),
125    )
127    app.add_directive('todo', TodoDirective)
128    app.add_directive('todolist', TodolistDirective)
129    app.connect('doctree-resolved', process_todo_nodes)
130    app.connect('env-purge-doc', purge_todos)
131    app.connect('env-merge-info', merge_todos)
133    return {
134        'version': '0.1',
135        'parallel_read_safe': True,
136        'parallel_write_safe': True,
137    }

This is far more extensive extension than the one detailed in Developing a “Hello world” extension, however, we will will look at each piece step-by-step to explain what’s happening.

The node classes

Let’s start with the node classes:

 3class todo(nodes.Admonition, nodes.Element):
 4    pass
 7class todolist(nodes.General, nodes.Element):
 8    pass
11def visit_todo_node(self, node):
12    self.visit_admonition(node)

Node classes usually don’t have to do anything except inherit from the standard docutils classes defined in docutils.nodes. todo inherits from Admonition because it should be handled like a note or warning, todolist is just a “general” node.


Many extensions will not have to create their own node classes and work fine with the nodes already provided by docutils and Sphinx.


It is important to know that while you can extend Sphinx without leaving your, if you declare an inherited node right there, you’ll hit an unobvious PickleError. So if something goes wrong, please make sure that you put inherited nodes into a separate Python module.

For more details, see:

The directive classes

A directive class is a class deriving usually from docutils.parsers.rst.Directive. The directive interface is also covered in detail in the docutils documentation; the important thing is that the class should have attributes that configure the allowed markup, and a run method that returns a list of nodes.

Looking first at the TodolistDirective directive:

3class TodolistDirective(Directive):
4    def run(self):

It’s very simple, creating and returning an instance of our todolist node class. The TodolistDirective directive itself has neither content nor arguments that need to be handled. That brings us to the TodoDirective directive:

 2class TodoDirective(SphinxDirective):
 3    # this enables content in the directive
 4    has_content = True
 6    def run(self):
 7        targetid = 'todo-%d' % self.env.new_serialno('todo')
 8        targetnode ='', '', ids=[targetid])
10        todo_node = todo('\n'.join(self.content))
11        todo_node += nodes.title(_('Todo'), _('Todo'))
12        self.state.nested_parse(self.content, self.content_offset, todo_node)
14        if not hasattr(self.env, 'todo_all_todos'):
15            self.env.todo_all_todos = []
17        self.env.todo_all_todos.append({
18            'docname': self.env.docname,
19            'lineno': self.lineno,
20            'todo': todo_node.deepcopy(),
21            'target': targetnode,
22        })
24        return [targetnode, todo_node]

Several important things are covered here. First, as you can see, we’re now subclassing the SphinxDirective helper class instead of the usual Directive class. This gives us access to the build environment instance using the self.env property. Without this, we’d have to use the rather convoluted self.state.document.settings.env. Then, to act as a link target (from TodolistDirective), the TodoDirective directive needs to return a target node in addition to the todo node. The target ID (in HTML, this will be the anchor name) is generated by using env.new_serialno which returns a new unique integer on each call and therefore leads to unique target names. The target node is instantiated without any text (the first two arguments).

On creating admonition node, the content body of the directive are parsed using self.state.nested_parse. The first argument gives the content body, and the second one gives content offset. The third argument gives the parent node of parsed result, in our case the todo node. Following this, the todo node is added to the environment. This is needed to be able to create a list of all todo entries throughout the documentation, in the place where the author puts a todolist directive. For this case, the environment attribute todo_all_todos is used (again, the name should be unique, so it is prefixed by the extension name). It does not exist when a new environment is created, so the directive must check and create it if necessary. Various information about the todo entry’s location are stored along with a copy of the node.

In the last line, the nodes that should be put into the doctree are returned: the target node and the admonition node.

The node structure that the directive returns looks like this:

| target node        |
| todo node          |
     | admonition title   |
     | paragraph          |
     | ...                |

The event handlers

Event handlers are one of Sphinx’s most powerful features, providing a way to do hook into any part of the documentation process. There are many events provided by Sphinx itself, as detailed in the API guide, and we’re going to use a subset of them here.

Let’s look at the event handlers used in the above example. First, the one for the env-purge-doc event:

1def purge_todos(app, env, docname):
2    if not hasattr(env, 'todo_all_todos'):
3        return
5    env.todo_all_todos = [todo for todo in env.todo_all_todos if todo['docname'] != docname]

Since we store information from source files in the environment, which is persistent, it may become out of date when the source file changes. Therefore, before each source file is read, the environment’s records of it are cleared, and the env-purge-doc event gives extensions a chance to do the same. Here we clear out all todos whose docname matches the given one from the todo_all_todos list. If there are todos left in the document, they will be added again during parsing.

The next handler, for the env-merge-info event, is used during parallel builds. As during parallel builds all threads have their own env, there’s multiple todo_all_todos lists that need to be merged:

1    if not hasattr(env, 'todo_all_todos'):
2        env.todo_all_todos = []
3    if hasattr(other, 'todo_all_todos'):
4        env.todo_all_todos.extend(other.todo_all_todos)

The other handler belongs to the doctree-resolved event:

 1    if not app.config.todo_include_todos:
 2        for node in doctree.findall(todo):
 3            node.parent.remove(node)
 5    # Replace all todolist nodes with a list of the collected todos.
 6    # Augment each todo with a backlink to the original location.
 7    env = app.builder.env
 9    if not hasattr(env, 'todo_all_todos'):
10        env.todo_all_todos = []
12    for node in doctree.findall(todolist):
13        if not app.config.todo_include_todos:
14            node.replace_self([])
15            continue
17        content = []
19        for todo_info in env.todo_all_todos:
20            para = nodes.paragraph()
21            filename = env.doc2path(todo_info['docname'], base=None)
22            description = _(
23                '(The original entry is located in %s, line %d and can be found '
24            ) % (filename, todo_info['lineno'])
25            para += nodes.Text(description)
27            # Create a reference
28            newnode = nodes.reference('', '')
29            innernode = nodes.emphasis(_('here'), _('here'))
30            newnode['refdocname'] = todo_info['docname']
31            newnode['refuri'] = app.builder.get_relative_uri(fromdocname, todo_info['docname'])
32            newnode['refuri'] += '#' + todo_info['target']['refid']
33            newnode.append(innernode)
34            para += newnode
35            para += nodes.Text('.)')
37            # Insert into the todolist
38            content.extend((
39                todo_info['todo'],
40                para,
41            ))
43        node.replace_self(content)

The doctree-resolved event is emitted at the end of phase 3 (resolving) and allows custom resolving to be done. The handler we have written for this event is a bit more involved. If the todo_include_todos config value (which we’ll describe shortly) is false, all todo and todolist nodes are removed from the documents. If not, todo nodes just stay where and how they are. todolist nodes are replaced by a list of todo entries, complete with backlinks to the location where they come from. The list items are composed of the nodes from the todo entry and docutils nodes created on the fly: a paragraph for each entry, containing text that gives the location, and a link (reference node containing an italic node) with the backreference. The reference URI is built by which creates a suitable URI depending on the used builder, and appending the todo node’s (the target’s) ID as the anchor name.

The setup function

As noted previously, the setup function is a requirement and is used to plug directives into Sphinx. However, we also use it to hook up the other parts of our extension. Let’s look at our setup function:

 1def setup(app: Sphinx) -> ExtensionMetadata:
 2    app.add_config_value('todo_include_todos', False, 'html')
 4    app.add_node(todolist)
 5    app.add_node(
 6        todo,
 7        html=(visit_todo_node, depart_todo_node),
 8        latex=(visit_todo_node, depart_todo_node),
 9        text=(visit_todo_node, depart_todo_node),
10    )
12    app.add_directive('todo', TodoDirective)
13    app.add_directive('todolist', TodolistDirective)
14    app.connect('doctree-resolved', process_todo_nodes)
15    app.connect('env-purge-doc', purge_todos)
16    app.connect('env-merge-info', merge_todos)
18    return {
19        'version': '0.1',
20        'parallel_read_safe': True,
21        'parallel_write_safe': True,
22    }

The calls in this function refer to the classes and functions we added earlier. What the individual calls do is the following:

  • add_config_value() lets Sphinx know that it should recognize the new config value todo_include_todos, whose default value should be False (this also tells Sphinx that it is a boolean value).

    If the third argument was 'html', HTML documents would be full rebuild if the config value changed its value. This is needed for config values that influence reading (build phase 1 (reading)).

  • add_node() adds a new node class to the build system. It also can specify visitor functions for each supported output format. These visitor functions are needed when the new nodes stay until phase 4 (writing). Since the todolist node is always replaced in phase 3 (resolving), it doesn’t need any.

  • add_directive() adds a new directive, given by name and class.

  • Finally, connect() adds an event handler to the event whose name is given by the first argument. The event handler function is called with several arguments which are documented with the event.

With this, our extension is complete.

Using the extension

As before, we need to enable the extension by declaring it in our file. There are two steps necessary here:

  1. Add the _ext directory to the Python path using sys.path.append. This should be placed at the top of the file.

  2. Update or create the extensions list and add the extension file name to the list

In addition, we may wish to set the todo_include_todos config value. As noted above, this defaults to False but we can set it explicitly.

For example:

import os
import sys


extensions = ['todo']

todo_include_todos = False

You can now use the extension throughout your project. For example:

Hello, world

.. toctree::

Hello world. Below is the list of TODOs.

.. todolist::

Some intro text here...

.. todo:: Fix this

Some more text here...

.. todo:: Fix that

Because we have configured todo_include_todos to False, we won’t actually see anything rendered for the todo and todolist directives. However, if we toggle this to true, we will see the output described previously.

Further reading

For more information, refer to the docutils documentation and Sphinx Extensions API.