:love_letter: Things I've learned about writing good READMEs.
Where does the term "README" come from?
The nomenclature dates back to at least the 1970s and the PDP-10, though it may even harken back to the days of informative paper notes placed atop stacks of punchcards, "READ ME!" scrawled on them, describing their use.
A reader1 suggested that the title README may be a playful nudge toward Lewis Carroll's Alice's Adventures in Wonderland, which features a potion and a cake labelled "DRINK ME" and "EAT ME", respectively.
The pattern of README appearing in all-caps is a consistent facet throughout history. In addition to the visual strikingness of using all-caps, UNIX systems would sort capitals before lower case letters, conveniently putting the README before the rest of the directory's content2.
The intent is clear: "This is important information for the user to read before proceeding." Let's explore together what constitutes "important information" in this modern age.
This is an article about READMEs. About what they do, why they are an absolute necessity, and how to craft them well.
This is written for module creators, for as a builder of modules, your job is to create something that will last. This is an inherent motivation, even if the author has no intent of sharing their work. Once 6 months pass, a module without documentation begins to look new and unfamiliar.
This is also written for module consumers, for every module author is also a module consumer. Node has a very healthy degree of interdependency: no one lives at the bottom of the dependency tree.
Despite being focused on Node, the author contends that its lessons apply equally well to other programming ecosystems, as well.
The Node ecosystem is powered by its modules. npm is the magic that makes it all go. In the course of a week, Node developers evaluate dozens of modules for inclusion in their projects. This is a great deal of power being churned out on a daily basis, ripe for the plucking, just as fast as one can write
Like any ecosystem that is extremely accessible, the quality bar varies. npm does its best to nicely pack away all of these modules and ship them far and wide. However, the tools found are widely varied: some are shining and new, others broken and rusty, and still others are somewhere in between. There are even some that we don't know what they do!
For modules, this can take the form of inaccurate or unhelpful names (any guesses what the
fudgemodule does?), no documentation, no tests, no source code comments, or incomprehensible function names.
Many don't have an active maintainer. If a module has no human available to answer questions and explain what a module does, combined with no remnants of documentation left behind, a module becomes a bizarre alien artifact, unusable and incomprehensible by the archaeologist-hackers of tomorrow.
For those modules that do have documentation, where do they fall on the quality spectrum? Maybe it's just a one-liner description:
"sorts numbers by their hex value". Maybe it's a snippet of example code. These are both improvements upon nothing, but they tend to result in the worst-case scenario for a modern day module spelunker: digging into the source code to try and understand how it actually works. Writing excellent documentation is all about keeping the users out of the source code by providing instructions sufficient to enjoy the wonderful abstractions that your module brings.
Node has a "wide" ecosystem: it's largely made up of a very long list of independent do-one-thing-well modules flying no flags but their own. There are exceptions, but despite these minor fiefdoms, it is the single-purpose commoners who, given their larger numbers, truly rule the Node kingdom.
This situation has a natural consequence: it can be hard to find quality modules that do exactly what you want.
This is okay. Truly. A low bar to entry and a discoverability problem is infinitely better than a culture problem, where only the privileged few may participate.
Plus, discoverability -- as it turns out -- is easier to address.
The Node community has responded to the challenge of discoverability in different ways.
Some experienced Node developers band together to create curated lists of quality modules. Developers leverage their many years examining hundreds of different modules to share with newcomers the crème de la crème: the best modules in each category. This might also take the form of RSS feeds and mailing lists of new modules deemed to be useful by trusted community members.
How about the social graph? This idea spurred the creation of node-modules.com, a npm search replacement that leverages your GitHub social graph to find modules your friends like or have made.
Of course there is also npm's built-in search functionality: a safe default, and the usual port of entry for new developers.
No matter your approach, regardless whether a module spelunker enters the module underground at npmjs.org, github.com, or somewhere else, this would-be user will eventually end up staring your README square in the face. Since your users will inevitably find themselves here, what can be done to make their first impressions maximally effective?
A README is a module consumer's first -- and maybe only -- look into your creation. The consumer wants a module to fulfill their need, so you must explain exactly what need your module fills, and how effectively it does so.
Your job is to
This is your job. It's up to the module creator to prove that their work is a shining gem in the sea of slipshod modules. Since so many developers' eyes will find their way to your README before anything else, quality here is your public-facing measure of your work.
The lack of a README is a powerful red flag, but even a lengthy README is not indicative of there being high quality. The ideal README is as short as it can be without being any shorter. Detailed documentation is good -- make separate pages for it! -- but keep your README succinct.
It is said that those who do not study their history are doomed to make its mistakes again. Developers have been writing documentation for quite some number of years. It would be wasteful to not look back a little bit and see what people did right before Node.
Perl, for all of the flak it receives, is in some ways the spiritual grandparent of Node. Both are high-level scripting languages, adopt many UNIX idioms, fuel much of the internet, and both feature a wide module ecosystem.
It so turns out that the monks of the Perl community indeed have a great deal of experience in writing quality READMEs. CPAN is a wonderful resource that is worth reading through to learn more about a community that wrote consistently high-calibre documentation.
No README means developers will need to delve into your code in order to understand it.
The Perl monks have wisdom to share on the matter:
Your documentation is complete when someone can use your module without ever having to look at its code. This is very important. This makes it possible for you to separate your module's documented interface from its internal implementation (guts). This is good because it means that you are free to change the module's internals as long as the interface remains the same.
Remember: the documentation, not the code, defines what a module does. -- Ken Williams
Once a README is located, the brave module spelunker must scan it to discern if it matches the developer's needs. This becomes essentially a series of pattern matching problems for their brain to solve, where each step takes them deeper into the module and its details.
Name -- self-explanatory names are best.
collide-2d-aabb-aabbsounds promising, though it assumes I know what an "aabb" is. If the name sounds too vague or unrelated, it may be a signal to move on.
One-liner -- having a one-liner that describes the module is useful for getting an idea of what the module does in slightly greater detail.
Determines whether a moving axis-aligned bounding box (AABB) collides with other AABBs.
Awesome: it defines what an AABB is, and what the module does. Now to gauge how well it'd fit into my code:
Usage -- rather than starting to delve into the API docs, it'd be great to see what the module looks like in action. I can quickly determine whether the example JS fits the desired style and problem. People have lots of opinions on things like promises/callbacks and ES6. If it does fit the bill, then I can proceed to greater detail.
API -- the name, description, and usage of this module all sound appealing to me. I'm very likely to use this module at this point. I just need to scan the API to make sure it does exactly what I need and that it will integrate easily into my codebase. The API section ought to detail the module's objects and functions, their signatures, return types, callbacks, and events in detail. Types should be included where they aren't obvious. Caveats should be made clear.
Installation -- if I've read this far down, then I'm sold on trying out the module. If there are nonstandard installation notes, here's where they'd go, but even if it's just a regular
npm install, I'd like to see that mentioned, too. New users start using Node all the time, so having a link to npmjs.org and an install command provides them the resources to figure out how Node modules work.
License -- most modules put this at the very bottom, but this might actually be better to have higher up; you're likely to exclude a module VERY quickly if it has a license incompatible with your work. I generally stick to the MIT/BSD/X11/ISC flavours. If you have a non-permissive license, stick it at the very top of the module to prevent any confusion.
The ordering of the above was not chosen at random.
Module consumers use many modules, and need to look at many modules.
Once you've looked at hundreds of modules, you begin to notice that the mind benefits from predictable patterns.
You also start to build out your own personal heuristic for what information you want, and what red flags disqualify modules quickly.
Thus, it follows that in a README it is desirable to have:
You don't need to use this format, but try to be consistent to save your users precious cognitive cycles.
The ordering presented here is lovingly referred to as "cognitive funneling," and can be imagined as a funnel held upright, where the widest end contains the broadest more pertinent details, and moving deeper down into the funnel presents more specific details that are pertinent for only a reader who is interested enough in your work to have reached that deeply in the document. Finally, the bottom can be reserved for details only for those intrigued by the deeper context of the work (background, credits, biblio, etc.).
Once again, the Perl monks have wisdom to share on the subject:
The level of detail in Perl module documentation generally goes from less detailed to more detailed. Your SYNOPSIS section should contain a minimal example of use (perhaps as little as one line of code; skip the unusual use cases or anything not needed by most users); the DESCRIPTION should describe your module in broad terms, generally in just a few paragraphs; more detail of the module's routines or methods, lengthy code examples, or other in-depth material should be given in subsequent sections.
Ideally, someone who's slightly familiar with your module should be able to refresh their memory without hitting "page down". As your reader continues through the document, they should receive a progressively greater amount of knowledge. -- fromperlmodstyle
Awesome; the ordering of these key elements should be decided by how quickly they let someone 'short circuit' and bail on your module.
This sounds bleak, doesn't it? But think about it: your job, when you're doing it with optimal altruism in mind, isn't to "sell" people on your work. It's to let them evaluate what your creation does as objectively as possible, and decide whether it meets their needs or not -- not to, say, maximize your downloads or userbase.
This mindset doesn't appeal to everyone; it requires checking your ego at the door and letting the work speak for itself as much as possible. Your only job is to describe its promise as succinctly as you can, so module spelunkers can either use your work when it's a fit, or move on to something else that does.
Go forth, brave module spelunker, and make your work discoverable and usable through excellent documentation!
Outside of the key points of the article, there are other practices you can follow (or not follow) to raise your README's quality bar even further and maximize its usefulness to others:
Aggressively linkify! If you talk about other modules, ideas, or people, make that reference text a link so that visitors can more easily grok your module and the ideas it builds on. Few modules exist in a vacuum: all work comes from other work, so it pays to help users follow your module's history and inspiration.
Include information on types of arguments and return parameters if it's not obvious. Prefer convention wherever possible (
cbprobably means callback function,
numprobably means a
Include the example code in Usage as a file in your repo -- maybe as
example.js. It's great to have README code that users can actually run if they clone the repository.
Be judicious in your use of badges. They're easy to abuse. They can also be a breeding ground for bikeshedding and endless debate. They add visual noise to your README and generally only function if the user is reading your Markdown in a browser online, since the images are often hosted elsewhere on the internet. For each badge, consider: "what real value is this badge providing to the typical viewer of this README?" Do you have a CI badge to show build/test status? This signal would better reach important parties by emailing maintainers or automatically creating an issue. Always consider the audience of the data in your README and ask yourself if there's a flow for that data that can better reach its intended audience.
API formatting is highly bikesheddable. Use whatever format you think is clearest, but make sure your format expresses important subtleties:
a. which parameters are optional, and their defaults
b. type information, where it is not obvious from convention
optsobject parameters, all keys and values that are accepted
d. don't shy away from providing a tiny example of an API function's use if it is not obvious or fully covered in the Usage section. However, this can also be a strong signal that the function is too complex and needs to be refactored, broken into smaller functions, or removed altogether
e. aggressively linkify specialized terminology! In markdown you can keep footnotes at the bottom of your document, so referring to them several times throughout becomes cheap. Some of my personal preferences on API formatting can be found here
If your module is a small collection of stateless functions, having a Usage section as a Node REPL session of function calls and results might communicate usage more clearly than a source code file to run.
If your module provides a CLI (command line interface) instead of (or in addition to) a programmatic API, show usage examples as command invocations and their output. If you create or modify a file,
catit to demonstrate the change before and after.
Don't forget to use
package.jsonkeywords to direct module spelunkers to your doorstep.
The more you change your API, the more work you need to exert updating documentation -- the implication here is that you should keep your APIs small and concretely defined early on. Requirements change over time, but instead of front-loading assumptions into the APIs of your modules, load them up one level of abstraction: the module set itself. If the requirements do change and 'do-one-concrete-thing' no longer makes sense, then simply write a new module that does the thing you need. The 'do-one-concrete-thing' module remains a valid and valuable model for the npm ecosystem, and your course correction cost you nothing but a simple substitution of one module for another.
Finally, please remember that your version control repository and its embedded README will outlive your repository host and any of the things you hyperlink to -- especially images -- so inline anything that is essential to future users grokking your work.
Not coincidentally, this is also the format used by common-readme, a set of README guidelines and handy command-line generator. If you like what's written here, you may save some time writing READMEs with
common-readme. You'll find real module examples with this format, too.
You may also enjoy standard-readme, which is a more structured, lintable take on a common README format.
Theory is well and good, but what do excellent READMEs look like? Here are some that I think embody the principles of this article well:
A helpful checklist to gauge how your README is coming along:
This little project began back in May in Berlin at squatconf, where I was digging into how Perl monks write their documentation and also lamenting the state of READMEs in the Node ecosystem. It spurred me to create common-readme. The "README Tips" section overflowed with tips though, which I decided could be usefully collected into an article about writing READMEs. Thus, Art of README was born!
You can reach me at
See The Jargon File. However, most systems today will not sort capitals before all lowercase characters, reducing this convention's usefulness to just the visual strikingness of all-caps.
Thank you to the following awesome readers for noticing errors and sending me PRs :heart: :
Thank you to @qihaiyan for translating Art of README to Chinese! The following users also made contributions:
Thank you to @lennonjesus for translating Art of README to Brazilian Portuguese! The following users also made contributions:
Thank you to @jabiinfante for translating Art of README to Spanish!
Thank you to @Ryuno-Ki for translating Art of README to German! The following users also made contributions:
Finally, thanks for all of the feedback! Please share your comments as an issue!