Generating Code Documentation for Polyglot Projects

At OddBird we believe that undocumented code is unfinished code. The act of documenting clarifies what we are building, and the resulting documentation guides consistency in how we do things in different parts of a project built by different team members at different times.

In a typical project, one of our documentation deliverables is a living style guide which serves as documentation of the visual elements used in the app as well as guidelines for and examples of usage. The style guide is built automatically whenever we build the code, and it is delivered as a static HTML website which can be served alongside the app. For example, the style guide for this website can be found here:

We are currently building our style guides using a tool called SassDoc which compiles the documentation based on special comments written inline in our stylesheets. For example, the triple-slash commented lines in the following Sass:

// Selection
// ---------
/// Selected text is highlighted in orange,
/// with any text-shadows removed.
/// @group typography
::selection {
  @include contrasted('select');
  text-shadow: none;
}

is rendered in the style guide like this:

We are working on our own theme for Sassdoc, called Herman, which provides extra tools for rendering samples of things like colors, fonts, and icons. (Alas, while we are using it to produce documentation, it is not yet very well-documented itself.)

Generating documentation from inline comments like this is ideal for developer-oriented documentation (that is, documentation of how the system is built rather than how to use it). It keeps the documentation right next to the code that it documents, which both makes it more accessible and increases the chances that it will be kept up to date.

But SassDoc only knows how to read comments from Sass. And the projects we work on implement patterns using multiple languages working together: Sass for stylesheets, Nunjucks/Jinja2 for markup, JavaScript for interactivity, Python on the backend. How can we generate documentation for a pattern that involves multiple languages, without giving up on the goal of writing documentation inline?

The Python documentation tool Sphinx has a nice pattern for this: rather than only fetching documentation from one particular kind of code source file, it allows for writing free-form documentation organized in whatever way makes sense, but with the ability to use “autodoc” directives to pull in documentation from inline source comments wherever makes sense. For example, this directive would add documentation generated from the code and comments in the rstblog module:

.. automodule:: rstblog

But this approach still suffers from the single-language problem! Sphinx’s autodoc extension is focused on Python code. And while it is extensible, there is a challenge in creating good autodoc extensions for other languages: different languages use different syntaxes, so need to be parsed by a tool that understands the language. But often a high-quality parser of a particular language is not available in the Python ecosystem. So Sphinx autodoc extensions to pull in inline documentation from other languages are not consistently available or well-maintained.

In order to provide more flexibility, I propose tackling this challenge using a decoupled architecture: a central documentation formatter that parses source code using separate processes.

The central formatter would work similarly to Sassdoc or Sphinx: read a file that specifies the overall structure of the documentation and look for directives that ask to include automatic documentation from other source files.

A parsing utility would have the limited responsibility of reading a source file of one particular type. The central formatter would run the parser as a separate process and output a JSON representation of the code structure and comments for the central formatter to make use of. This way the parsing utility can be written in whatever language best supports parsing the source language.

As a proof of concept, in the near future we intend to add a feature to Herman to automatically include documentation of macros from Nunjucks templates. Stay tuned!

If you have thoughts about how to make documentation better, or if you’d like to hire us to help set up a living style guide or other tools, please get in touch.