Choosing RST#
Status#
Provisional
A major concern with accepting this decision is whether we can overcome the RST and GitHub barrier to entry by investing in training and proper documentation. This decision should be re-evaluated once we’ve put the relevant quickstarts, how-tos and references through their paces.
Context#
Now that we agree on the doc:general goal <0001-purpose-of-this-repo> of a root documentation site, we need to figure out how we will write the words for the site. Per OEP-19 developer documentation is written in RST. However since the audience for all documentation is not developers it is worth re-considering and reviewing the alternatives and actively make a decision here rather than passively choose what already exists.
Decision#
We will use Sphinx-flavored RST for the root docs.openedx.org site and any documentation that is written within the repositories where the relevant features exist. In particular we will use the Sphinx tool to manage our RST content and publish it using ReadTheDocs.org
Nuance#
This does not mean that all documentation needs to live in repositories nor that all of it must be RST. There are many good reasons to write documentation on the wiki, in GitHub Issues, or as discussion forum posts. Especially early on when there might be a lot of iteration.
Summary of Findings#
Markdown often looks like the right choice when looking at it from the perspective of individual documents but quickly stops being a good solution as we expand the scope of what we’re trying to solve for in a larger project. For this reason, Markdown is often going to come up as a suggested replacement for Sphinx Flavored RST. However, while it has a pretty low barrier to entry, it’s overall capabilities are not yet mature enough to be a good replacement for a project as large and as complex as the Open edX Platform.
We choose Sphinx RST as our tooling of choice because:
It allows us to easily write words while spending minimal effort writing basic markup.
It has tooling to help us build translations for documentation.
It lets us use more advanced components using a straight forward directive syntax that’s consistent and can be extended to add more features.
It can be deployed easily using off the shelf tools like readthedocs.org
It can be stored in version control which gives us the ability to build review and preview process around changes to the documentation.
It allows us to deploy multiple versions of the documentation so that it’s easy to access documentation for previous named releases.
RST allows for easy componentization of documents so that they can be re-used. This is especially useful for user facing How-Tos. For example, we often need to do the same initial steps to setup a course. Making this re-usable allows us reduce boilerplate in our documentation and ease updates.
Developer documentation is in RST already, choosing a different technology requires us to revisit that decision as well which may be more costly.
Concerns Considered#
These were some notes on specific concerns we considered when making our decision. For each concern, we provide some notes on how we will address them.
Git and GitHub are a barrier to entry for non-technical writers.
We acknowledge that we will have to do training and write good documentation to mitigate this for less technical users in our community.
Investments by GitHub as well as effort we’ll be making in terms of quickstarts and how-tos will help
RST is a barrier to entry.
While RST can become very complicated, the basics of it are not much different from Markdown and other simple markup languages. A small opinionated quickreference should help mitigate the major issues.
Translating RST docs is difficult.
The process and tooling around translations is not robust, the translations working group has had issues getting everything to work as expected. However, in spite of this Sphinx seems to be more mature than the alternatives we looked at.
Rejected Alternatives#
Basic Markdown#
The basics of Markdown are extremely simple and easy to learn. Because Markdown was built for the web, it’s often supported in web based editors and many people are already quite familiar with it from other uses.
However, there are currently over 30 implementations of the original Markdown reference specification. The basic specification does not meet the needs of our project as it lacks many of the capabilities that we would want to build a large documentation project.
It doesn’t allow for relative document links, which makes it difficult to have multiple builds (local development vs production) of the same docs and have the links point to local files locally and remote files when deployed.
Complex items such as videos, panels/cards, and generated documentation from other sources such as code must be added manually using raw HTML. This means we lose Markdown’s advantage when we want to do anything more advanced than raw text and images.
Basic RST#
The basic RST specification has many of the same problems as the basic Markdown specification.
Like basic Markdown, basic RST also doesn’t allow for relative document links.
Basic RST also lacks support for complex items such as videos, panels/cards and other ways of generating documentation from code.
We reject this for much the same reason as we reject basic Markdown.
MkDocs Flavored Markdown#
The MkDocs tool powered by the Python-Markdown flavor of Markdown solves many of the problems with pure Markdown by introducing the ability to add extensions that add more functionality. Both the issues noted above in the Basic Markdown are solved via extensions in MkDocs. However, the extension mechanism is not consistent so each extension essentially adds its own new syntax to learn in order to make use of the new capabilities.
Sphinx has similar extension capabilities but provides a much more consistent syntax with which to make use of complex capabilities in the form of directives.
So while there are ways to overcome the shortcomings of basic Markdown, they add undesired complexity which reduces flavored Markdown’s advantage over flavored RST.
Other reasons that we reject this option:
Currently no good translations tooling exists to ease the process of translating the Markdown content.
MDX Flavored Markdown#
Another popular way of writing Markdown is MDX, a flavor of Markdown that combines Markdown syntax with JSX syntax to let you write React web pages. While it is possible to create a documentation site this way, we do not believe it is the right tool for the audience as it adds even further barriers to entry for non-technical documentation writers.
Other reasons that we reject this option:
Requires us to re-develop many capabilities that the other options already have available. From basic things like Admonitions to more complex things like code and video blocks.
Currently no good translations tooling exists to ease the process of translating the Markdown content.
Increased development and maintenance burden as we would have to deploy the site ourselves rather than using an existing industry standard tool such as ReadTheDocs.org
No good tooling exists to deploy and maintain multiple versions of the same documentation.
Confluence Wiki#
One of the major drawbacks of both the RST and Markdown approaches is that both require us to use git and GitHub to coordinate and make changes. We currently believe that this is a barrier that can be overcome with sufficient guidance and documentation for newcomers. However an alternative that was considered was to use the existing Confluence wiki as the root documentation site.
We rejected this option for the following reasons:
It introduces a much more complex permissioning system for who can edit what documentation.
It lacks a good search mechanism which can’t easily be replaced with standard search engines because of its complex permissioning system.
It is harder to guard against malicious changes since Confluence does not provide a review workflow for content changes.
Keeping multiple versions of documentation available is not easily possible.
There is no workflow for maintaining translations of documentation.