DEV Community

Fernando Martín Ortiz
Fernando Martín Ortiz

Posted on

Unidirectional Architectures and time traveling in Swift: I

Time traveling in Swift is possible. This is a series that describe how Unidirectional architectures can be applied in Swift and how we can do very impressive things by enforcing them.

This is the first article in the series, but before deep diving into Swift specifics. I'd like to give you an introduction to the general concepts that make all of this possible.

What is time traveling?

Of course this is not my idea. Time Traveling has been discussed in the Javascript world several times. However, there are not so many examples of it in Swift.

When we talk about time traveling in the context of a Frontend application, this is what we have in mind.

We can record user sessions, and then reproduce the actions, go back, go forward, and "travel" in time, because all the actions are reproducible.

The main components

A word in the previous description is important: Actions. An Action is an object that describes something that happened in the system.

Did the user change the text in a textfield? That is an Action.

Did the user tap the button to create a To Do? That is an Action.

Did the user complete a To Do? That is also an Action.

Given that, if we want to describe something that happened in the system, anything, we need to create an Action that describes the event.

arch-001

In order to time travel, we need to be able to get a certain State for the system by applying the Action objects, sequentially. Nothing else can possibly affect or modify the State. If anything else was able to mutate the State other than an action, then we couldn't reproduce old states accordingly.

unidirectional-005

In the diagram above, I illustrate how a sequence of actions resulted in a state with an empty text attribute and an array with a single todo object that is completed. If we removed the toggleTodo action, we could apply all the other actions and get a state which is in turn the previous state. It is an undo event.

unidirectional-006

We can also store the deleted actions in a separate stack, so we can bring those back to the main actions stack and calculate the state again. It is a redo event.

unidirectional-004

So, we apply a set of Action objects, and we get a resulting application State. So far, so good. But, we can think of actions as an enum with a number of cases with associated values, and the State of a struct or a class with all the values inside, and that's ok. However, something is missing: how is state mutated? Where is that mutation described?

The missing piece: The reducer

A reducer is a pure function. That means that its output is only calculated based on its input. It is a function in the traditional mathematical sense of the concept. Each element in its domain has only one element in its image.

Untitled-001

Image from Math is fun

The reducer is the only function which is allowed to perform mutations in the State. Ok, it doesn't mutate the State. It actually produces new states based on the previous State and an Action. It does so because it should only depend on its input. If it mutated the current state, it would also depend in a variable from the outside.

unidirectional-008

For each Action we get in the system, we send that action to the reducer along with the current State. The reducer returns the new State, so we store that value and we are prepared for the next Action that we will eventually get.

unidirectional-007

Due to the functional purity of the reducer, and given that all the events that could possibly mutate the State are expressed through an Action, each step is reproducible. We can travel in time by storing all the actions and calculate previous and next State values by applying all the Action values sequentially using the reducer function.

f(State) = UI

For all of this to work, we need to enforce functional purity in another transformation: the view should be a pure function of the application state.

unidirectional-009

Changes in the application states will produce modifications in the View to happen. This may sound as a lot of work, but keep in mind to things:

  1. The power of programming depends more on the things you can't do than on the things you are able to do. Applying restrictions to the things you can do is a way of getting properties in your system that allow you to do this kind of things.
  2. SwiftUI already does this for you for free. Views in SwiftUI are rendered depending only on the state you have on them and on your ObservableObjects. This is taking that approach one step forward, and centralizing all your state in a single, big State object that can be mutated only by a function called reducer.

The store

There is another component that hasn't been discussed so far. The Store is an object, that can be implemented in SwiftUI as a ObservableObject, that holds our State, our reducer, and exposes a function called dispatch(action:), which takes an Action and replaces the current State for the result of the reducer function.

unidirectional-11_(1)

To sum up

As said, The power of programming depends more on the things you can't do than on the things you are able to do. If you enforce functional purity in your application, and express all mutations through actions, the state then becomes reproducible, you can go back and forward as you want, you can record user sessions, and many many many other things we havent covered in this article.

In the next article, we will translate all of this in actual Swift/SwiftUI code, so you will get a more practical understanding on how all of this fits in a real app.

If you want to learn an iOS architecture that applies all of this and is production ready, I would HIGHLY recommend The Composable Architecture.

Top comments (0)