August 16, 2017 Johan Sigfred Abildskov   Techblog

#NoDocs

How to make your software obvious

Developers hate wasting time on documentation. We all wish there was less of it. This article shows you how to cut documentation down to the absolute minimum by following a few key development principles.

Broken Code or Broken Docs

You know the frustration. You want to run a simple command, but it fails with some obscure AC35BA882 error code. Or worse, you just wanted to install a program and now you find yourself running through a thirteen-page Word document. You might even be in your fourth Stack Overflow post hoping to find that link to the official documentation that solved exactly your case the last time you had this issue.

All this is obviously a failure on the part of the developer. But how can we avoid leading our users on wild-goose chases?

In this blogpost we will outline a set of principles that you can apply to your own software and tooling.

It’s about being there for the user

Let’s get one thing right. Our mission is to provide value to our customers. If we don’t do that, we have failed.

So, when our software breaks, or is complex, it is our divine duty to make it as frictionless as possible for the user to get back on track and get value out of our products.

Whether this means helpful error messages, command completion or ample configuration examples, it is an important and often neglected task!

The roadblocks that are in the way of our users should be our first priority, even if we just enjoy tinkering with our code.

Let’s also accept the failures of our software. We are not perfect. The software is not perfect. It will never be perfect. Let’s own our failures and accept responsibility rather than going on the defensive.

It’s not (always) the customer’s fault our software broke.

Do not blame the user for your assumptions

I have a pet peeve. My soul dies a little bit every time a piece of software or a script crashes with a useless exception because some arbitrary folder it apparently expected is missing. It should not be an exceptional situation. The script knows exactly what the circumstances, problem and resolution are. So, what is your excuse for not letting the user know what is going on and how to resolve it? Or, better yet, ask if you should fix the situation for them.

This behavior leads to much frustration in users because we have a feeling that we did nothing wrong and yet the product crashed.

Script what you can

There is one thing I know with 100% certainty. If I have to do a manual task, sooner or later I will fuck it up!

If I have to create seven directories that need to be in very specific locations, ask if you can do it for me. I’m all for delegating.

Should the configuration be complex it must be accompanied by sufficient documentation and samples, but an interactive prompt that generates the configuration will be a big win.

And please make sure your software does something sane in the default configuration.

Contain yourself!

I love data and I can’t get enough of random statistics. But I’m also into developing debug information and user help. If information is not immediately helping the user to fix the issue on his own, then it is hindering the user.

Hopefully it is very rare that we at any point in our software do not know what went wrong. So use that same laser focus to clearly present the cause and resolution to the user. This empowers the user and provides for the most seamless experience.

Generally Git is great at this. The laser focus part is not always present, but after running git status and actually reading the output, you seldom find yourself in doubt of what your actions should be.

If it looks like a duck

Whether you are naming a function or describing a feature, make it obvious - not clever - what the purpose is.

When we are obvious about our purpose, of our intent, we give the user context. This helps the user to reason about the expected behavior of the item at investigation.

A configuration should not be named TimeoutRetry. As a user, I do not know if I should use a boolean in order to have the software retry if something times out, or if I should give it a number. If I give it a number is it then a number of retries, or the timeout before a retry in what unit of time? A configuration could be much more clearly named “RetryAfterTimeout”.

Walk a mile in their boots

I’m a tools guy, so I live in the command line. We live by the idea that “the terminal is king”, and “nix is the only serious OS”. Sometimes I forget that this is not necessarily the world of our customers. Of the people that I am providing value to.

I might love doing my Git work in the terminal because that is the context that I am working in. But, if I’m not investigating the context the users are experiencing, I’m failing.

Hell, I’m a toolsmith. I’ll write that Jenkins plugin, build that Visual Studio integration, or create the alias so my users will feel awesome.

Burn the papers

Aim to make your documentation obsolete. Not in the sense that it becomes outdated over time, but in the sense that the users only in the rarest of cases need to consult it. This means your software should be obvious, resilient to errors, and self-documenting.

Consulting the manual is the first step towards making a support request, so see it as an investment in minimizing your support budget needs.

Conclusion


  • Give the user what they need in the context they’re in
  • Handle exceptions
  • Avoid information overload
  • Follow the user’s path…

User's path