Documenting is hard
A non-trivial aspect of my job for the past year at CloudBees has been communication. To claim that this is a new change in my career would be to fundamentally mis-attribute the vast majority of what makes good Software Engineers and Engineering Managers good. Communication in my job as a “Technical Evangelist” (or as my business card states: “Community Concierge”) is many orders of magnitude more involved than it was an Engineering Manager, and what makes it very challenging is the size of the audience. As an Engineering Manager the audience is typically less than 20 people throughout an organization where the spoken-word is the primary means of communication. By conservative estimates more than a million people use and interact with Jenkins as part of their work, the primary way to reach them being written English in some medium or another.
It is an embarrassing admission to say that I have never really spent much time worrying about documentation for any of the software I have worked on over the past lotsofyears. Considering one of my goals is to communicate about common patterns, best practices, and new technologies to the vast expanse of the Jenkins user-base, I care more now about documentation than at any previous point in my career.
And writing documentation is hard.
(it’s no wonder Jenkins has been so poorly documented in the past!)
I wouldn’t consider myself much better than I was twelve months ago at writing documentation, but I have improved, and want to share a few thoughts on how.
Writing as a craft
Most of my “documentation” in the past has been code comments, class and method annotations, for which my audience is typically a future version of myself who is annoyed with the mistakes of a past version. Basically, if I consider myself the typical developer who will be reading my documentation, I would write the documentation for somebody like me, with all the understanding and mental model that I possess.
Always code as if the guy who ends up maintaining your code will be a violent psychopath who knows where you live. Code for readability (John Woods)
High quality documentation cannot make those same assumptions of the audience. It typically must address both novice and advanced readers within the same body of work. The structure and style of the FreeBSD Handbook has been a major source of inspiration for my own work on the Jenkins Handbook. Having used FreeBSD for over a decade, I still find myself periodically referring back to sections of the FreeBSD Handbook as a reference guide, but I can still remember a time when I started at the beginning and found it useful.
Having read back through the Handbook with an eye towards how one might write such a practical set of documentation, a few general characteristics can be identified:
User and Task-focused
The starting set of documents assume no prior knowledge or existing FreeBSD installation and thoughtfully guide the user through the installation process. As if targeted at a user who has sat down at an empty computer screen and wishes to go from zero to a usable system, the Handbook then progresses through the basics of configuring a FreeBSD system including installing third party applications and setting up the X Windowing system. For a user who knows what they’re looking for, many of these sections will be ignored, but for the beginner, they carefully guide the reader through progressively more complex and important tasks. Which leads me to structure.
Most documentation I have seen written by developers tends to be one-off. A developer adds Feature X, and updates the README or some other document with a paragraph explaining Feature X. In doing so, usually ignores the forest for the trees (to butcher a colloquialism). A colleague, who shall go unnamed, referred to this resulting documentation as “info vomit.” Yes there is documentation here, but it has been spewed into an unordered, seemingly un-curated, series of unrelated paragraphs which are eminently confusing wall of text to read through.
The structure of high quality documentation, like the FreeBSD Handbook, is thoughtful and incrementally builds upon previous sections. Compared to, for example, the Gradle User Guide which is also very useful, but is structured more haphazardly. Its chapters and sections don’t cleanly build upon knowledge gained in the previous section which also has an unfortunate side effect of forcing the reader who might just be starting out with Gradle to hunt through the haystack for the appropriate sections to start practically using Gradle.
An easy to overlook characteristic of high quality documentation is also consistency, which entails consistency of terms, formatting, and of course style.
Leading up to the Jenkins 2.0 launch in early 2016, a few members of the project like Chris Orr made efforts to stamp out inconsistency verbiage in Jenkins, a crucial prerequisite to consistency in documentation. As an experienced Jenkins user, I understand what somebody means when they interchangeably mix terms like “VM”, “slave”, “node”, “machine”, and “agent.” A beginner doesn’t have that same built up mental model, so mixed terms will cause confusion.
Consistent formatting is also an important, but odious, task. With any visual presentation of content, differences in style and formatting between different documents results in additional cognitive overhead, forcing the reader to consider more of the document than they may need to. When all headings look like so, each section can be expected to have a preamble, source code blocks are consistently marked up, and tips/notes (admonitions) are consistently formatted the reader can much more easily scan the document for the relevant information.
I certainly have my own opinions of what the right formatting is for technical documentation, but at the end of the day, so long as it is consistent across a corpus of documentation, that suffices.
This is deceptively obvious and not as subjective as I would have thought a couple of years ago. There are numerous texts on writing well (“On Writing Well”, “The Elements of Style”, and “Simple & Direct” are my recommendations) which should be followed or taken into consideration when writing documentation. Poorly written English documentation is still preferred to unwritten documentation, but still makes the content difficult to read and understand. A few guidelines that I consider important to keep in mind are:
- Don’t be clever: nobody cares about your expansive vocabulary, most times your audience is both English-native and non-native speakers. Strive for the simplest, most concise, sentence structure while still flowing (it shouldn’t read like it was written by a child).
- Avoid parenthetical statements: Developers seem to have stack-based thoughts and will tend to write deeply nested statements which can be confusing to parse and understand. Practically all parenthetical statements I have seen warrant a proper sentence.
- Use spell-check.
- Err on the side of proper nouns in place of pronouns: I seen this more often with documentation written by non-native speakers, but use of pronouns in statements should be avoided. For example “X is a tool which integrates with this, that, and the other. It runs on Linux.” would be clearer with “X runs on Linux” to avoid any confusion, however temporary, as to what the “it” refers to. Clarity is king!
In working with documentation, I often encounter developers who advocate for Markdown as a format for documentation.
They are wrong, er, misguided.
Markdown is a highly useful subset of HTML for writing GitHub Issues and shitposting on Stack Overflow. It lacks however practically all the functionality needed to make it useful for documentation, compared to formats like AsciiDoc or reStructuredText.
My preference is towards AsciiDoc, by way of AsciiDoctor, which the Jenkins project uses very heavily for all our new documentation needs. This includes formatting and functionality such as:
- Various block delimiters such as source code or quote blocks
- Footnote support
- Inline image support
- Defining inline anchors for:
- Inter-document and intra-document cross-referencing
- Document attributes for defining document meta-data and properties
- Automatic tables of contents and bibliographies
- Rendering PDF, DocBook, HTML, Man pages, and other output formats
- Extension support for defining custom markup/tags
Without a good documentation tool like AsciiDoctor, I do not believe that we (the Jenkins project) would be able to create consistent, quality documentation.
Finally, an “outside-in-approach” is what I believe to be another key to writing better documentation. This might be more subjective than the previous two sections in that I tend to organize my thoughts in outline form regardless of medium.
As far as documentation, what has helped more than anything has been to start at a high level and progressively work inward, for example:
- Who is the audience? Jenkins users.
- What are the common things that Jenkins users need to do or know? Install Jenkins, Configure Jenkins, Configure their workloads on Jenkins (and many more topics).
- (For example) What does Installing Jenkins entail? Depends on the platform, there’s Linux packages, installers, and containers available.
- What do I need to know to get Jenkins running on Linux? Add these repositories, and some other bits.
By putting my thoughts down for each level before moving onto the next, I rapidly identify misplaced ideas or topics, or in the best-case scenario, find useless topics which needn’t be documented to begin with. This approach has an added benefit of forcing an ordering more akin to how a beginner would approach Jenkins, instead of me brain-dumping (or info vomiting) 8 years of knowledge into a document.
Frankly, I hope you consider contributing quality documentation to any open source project you use, since it’s always in short-supply, for what I hope are now obvious reasons.
In short, documenting is hard.