Save 45% for a limited time.

Get access to all courses and lessons on egghead today.

Autoplay

    Elminate Illegal States in Reason with Variants

    Nik GrafNik Graf

    The ability to express multiple options that are exclusive as a data structure is a powerful concept. To leverage it's full power it's requires some rethinking on how to model our data.

    We are going use a simple UI example to rethink how to approach representing our state.

    This lesson was highly inspired by Yaron Minky's advice to "make illegal states unrepresentable" in his talk Effective ML as well as Richard Feldman's talk Making impossible states impossible with Elm.

    reasonReason
    Code

    Code

    Become a Member to view code

    You must be a Member to view code

    Access all courses and lessons, track your progress, gain confidence and expertise.

    Become a Member
    and unlock code for this lesson
    Transcript

    Transcript

    Instructor: 00:00 To demonstrate how to make illegal states impossible, we're going to build a simplified version of an app fetching and rendering the current user info. We start out using pseudo code to define the UI states. We need three of them.

    00:15 First, we want to print a loading text while we're waiting for the data. Next, in case of a successful request, we want to print, "Your name is" followed by the name. In case of an error, we print, "Something went wrong."

    00:30 Let's start by declaring your record type request. First, we add a field loading of type Boolean to indicate if the request is currently loading. We do the same for error. Finally, we add the field name of type string.

    00:48 Right away, we create the first legal state. While loading is set to true, error is set to false, and name an empty string. The print logic is pretty straightforward. If my request is currently loading, we return loading. In case it fails, we print something is wrong. The only other option is name is available, and we print the name.

    01:16 The second legal status, when the request finished loading but failed, works as expected. Last but not least, the legal state for loading finished, and the field name is filled. We covered all three cases and fulfilled the requirements.

    01:39 Wait a minute. Did you notice there's a lot that possibly could go wrong? By accident or on purpose, someone could create a state that should never be possible. For example, we could set loading to true and error to true.

    01:58 When reevaluating our if/else expression, it's really interesting. It renders loading but by definition, this is a state that should never happen. Due to the way our data is structured, it opens up the question, what should our UI code do in such a case? Maybe show the error instead of loading? Maybe ignore this at all, because it should never happen? Clearly, it can.

    02:22 It might not be you, but someone else might come in who didn't know about the initial requirement and create states that aren't legal. What's even more troublesome is not one combination, but even in this simple example, there are more illegal states. For example, loading true, error false, but the name set. Or loading false, error true, and the name set.

    02:47 There is one case I'm especially concerned about if the request succeeded but the assignment of the name didn't happen and therefore stayed empty. In this case, the user would see a broken UI, a bug that leads to a bad user experience.

    03:03 Of course, this is in a simple example, and you probably would have identified the issue quickly, but even this simple example requires additional cognitive overhead for developers and in worst case leads to bugs like the one I just described. Believe me or not, but it doesn't get better the more complex your state becomes.

    03:22 What now? There is a better way how to deal with this problem. This is exactly where variants come in. Let's start from scratch and recreate the type request with a variant. Each constructor represents one of the states, loading, error, success. Success was an additional string for the name.

    03:42 We set our state to the loading constructor and write a switch expression for a text to be rendered. We can change the state and rerun our switch expression. All cases work as expected. What's really fascinating about this, with a request variant, we can never have loading, error a success at the same time.

    04:19 In addition, using the switch expression in combination with the variant, we can't forget a state. For example, if we forget about the loading state, the compiler will warn us. There is one more edge case, though.

    04:32 What if you can't trust the backend and, for some users, it might return an empty string? This would lead to the same broken UI as I described earlier as a bug. Fortunately, since we can match for exact values, we can even match for an empty string and return an appropriate case.

    04:51 Isn't it fascinating? One additional concept to represent states, and suddenly, we eliminated a whole range of bugs. I don't know about you, but I was super excited and mind blown the first time I realized this. To me, this is one of the most critical reasons why I'm using Reason.

    05:09 Last move on because I want to demo you one more thing. In a real-world case, the constructor, success, most likely will contain another data structure like another variant, record, or list. For example, a user response record that would contain the ID, name, and age of a user. This is exactly what we can use in a success constructor. Pretty cool.

    05:45 Before we move on to the next lesson, let me remind you that a type system doesn't magically eliminate bugs. Properly leveraged, though, it points out the unhandled conditions and asks you to cover them. The ability to model this or that correctly is crucial to avoid bugs.

    Discuss

    Discuss