Introduction #

This article illustrates how we can build a Redux library in Rust. This library is thread safe and asynchronous (using Tokio). The middleware and subscribers will be run in asynchronously via Tokio tasks. But the reducer functions will be run in sequence (not in separate Tokio tasks).

💡 Learn more about:

  • Redux from the official docs. You can get familiar w/ the store, reducer functions, async middleware, and subscribers. Along w/ the idea of finite state machines as an effective way to manage your application’s state.
  • Tokio from our article. You can get familiar with the async programming model and the Tokio runtime. And get some insights into writing macros that help you write async code.

📜 The source code for the finished Redux library can be found here.

🌟 Please star the r3bl-open-core repo. You can get lots of useful Rust command line apps from the r3bl-cmdr crate in this repo:

  • 🐱giti: run interactive git commands with confidence in your terminal
  • 🦜edi: edit Markdown with style in your terminal
📦 You can install them using cargo install r3bl-cmdr

giti in action edi in action

If you would like to get involved in an open source project and like Rust crates, we welcome your contributions to the r3bl-open-core repo.

This article is getting us ready to building more complex TUI apps next using crates like termion and tui.

For more information on general Rust type system design (functional approach rather than object oriented), please take a look at this paper by Will Crichton demonstrating Typed Design Patterns with Rust.

Architecture #

Before we start, let’s get familiar with the Redux architecture.

  1. The store is a central place to manage the state of your application. It is the only place where you can change the state of your application. In order to change the state, you need to dispatch an action. State is immutable. So you can’t change it once it is created. The only way to change the state is to dispatch an action, which will generate a new and also immutable state. This is the key to Redux’s unidirectional data flow.
  2. An action describes your desired change and includes any data (in the form of arguments that will end up going to middleware or reducer functions).
  3. Once an action has been dispatched to the store, the reducer functions are responsible for updating the state based on these incoming actions. Reducer functions must also be pure functions, ie, no side effects are allowed (can’t generate unique ids, perform async function calls, use random number generators, or even use the current time). This is severely limiting, since these side effects are where most important things in an application’s business logic actually happen. This is where middleware functions come in.
  4. The middleware is a function that takes an action and returns a new action. This is where all kinds of async operations can be performed. Ultimately, each middleware function call asynchronously resolves into a new action. This action is then dispatched to the store (where the pure reducer function will use it to generate a new state).
  5. The subscriber is a function that is called whenever the state of the store changes. This is where you would normally perform UI updates, but they don’t have to be restricted to just UI. Any kind of data can be observed by the subscriber.

In our implementation of Redux, we will make the middleware and subscriber functions asynchronous. And we will use the Tokio runtime that allows us to run them concurrently and in parallel. However, we will run the reducer functions in sequence.

Here are some of our assumptions about the performance characteristics of these components.

  1. A reducer function is pure and is really fast. It isn’t allowed to do any async work, so its very limited in what it can do. It is best to run these in sequence.
  2. A middleware function is async and can be really slow (doing computations or waiting around for IO). These are heavy functions and its good that we can run them asynchronously using Tokio.
  3. A Subscriber function can be really slow as well, since it can render some really complex UI. However, since this is separate from state changes, we can run them asynchronously using Tokio.

💡 Please read our article on Tokio to learn more about the terms: Tokio runtime, synchronous, asynchronous, concurrent, parallel.

To make things even more flexible, we will provide 2 ways of dispatching actions:

  1. A spawning dispatch function. This frees the main thread from waiting around for the middleware, reducer, and subscriber functions to complete. Even the reducer functions will not block the calling thread. A new Tokio task is spawned inside of which the reducers are run, the async middleware and async subscribers are also run.
  2. A regular dispatch function. This does not spawn a new Tokio task. Instead, it runs the reducer functions on the calling thread. However, the middleware and subscriber functions will be run asynchronously in other Tokio tasks. You can actually await the results of all of them to complete.

Of “things” and their “managers” #

This brings us into the implementation of the Redux library. The first thing we will need is a Redux store that is shareable between threads / tasks and also allows thread safe interior mutability. We will also need other structures that have these same requirements (wrapped functions / lambdas, lists, etc).

💡 Please read our article on shared ownership and interior mutability.

Before starting our implementation, let’s take a close look at the following pattern we will heavily lean on which to make this “shareability” happen, which is:

  1. Wrap a “thing” that we want to be shareable inside of a Mutex or RwLock, and then wrap that inside of an Arc. The naming convention that we apply to this is prefixing the word Safe before the “thing” (e.g. SafeThing).
    • The “thing” in this case is Vec<T>.
    • And the Safe “thing” is Arc<Mutex<Vec<T>>> or Arc<RwLock<Vec<T>>>.
  2. So far, we’ve just made some type aliases. Where is all this instantiated, what holds memory? Let’s make a struct which has a single property (named my_arc) that holds the Safe “thing”. We will call this struct a Manager of the Safe “thing”, (e.g. SafeThingManager). In other words your code will work w/ the “manager” and not the “thing” directly.
    • The Manager of the Safe “thing” (or just “manager”) is what your code will work with. And not directly with the Safe “thing”.
    • The “manager” is the struct that occupies memory.
  3. The “manager” will:
    • Provide a constructor method new() that allows you to instantiate it.
    • Allow shared ownership by providing a get() method that simply returns a clone of the underlying Arc which we call my_arc.

💡 This is a fantastic cheat sheet of pointers and ownership in Rust. Here’s more info on vtable, dyn Trait, dynamically sized types:

  1. discussion
  2. book - wide pointers
  3. book - trait objects
  4. book - trait object safety
  5. article - super traits

Here is an example 🎉.

use std::sync::Arc;
use tokio::sync::RwLock;

pub type Thing<T> = Vec<T>;

pub type SafeThing<T> = Arc<RwLock<Thing<T>>>;

pub struct SafeThingManager<T>
where
  T: Sync + Send + 'static,
{
  my_arc: SafeThing<T>,
}

impl<T> SafeThingManager<T>
where
  T: Sync + Send + 'static,
{
  pub fn new() -> SafeThing<T> {
    Self {
      my_arc: Arc::new(RwLock::new(Thing::new())),
    }
  }

  pub fn get(&self) -> SafeThing<T> {
    self.my_arc.clone()
  }
}

You might be asking yourself: Ok, how do we access the "thing" inside the "manager"? 🤔

Excellent question 👍. The pattern we will use to do this is also repeated throughout the rest of the library codebase. The logic goes something like this:

  1. Call get() on the “manager”.
    • This returns a cloned Arc.
  2. Call .lock() or read() or write() on it (depending on whether you used Mutex or RwLock).
    • This returns a MutexGuard, RwLockReadGuard or RwLockWriteGuard). Which we then have to .await (since these are Tokio locks).
  3. Finally we have access to the underlying “thing” that we can now use 🎉.

As you can see these steps can be tedious and repetitive. One way to handle this repetition is via the use of macros. We can use macro_rules! to create some macros for this common pattern across some of our structs. Here are two examples of such macros.

💡 Please take a look at our Tokio article to learn more about macros and how they can be used w/ this “manager” and “thing” pattern.

/// The `$lambda` expression is not `async`.
macro_rules! iterate_over_vec_with {
  ($this:ident, $locked_list_arc:expr, $lambda:expr) => {
    let locked_list = $locked_list_arc.get();
    let list_r = locked_list.read().await;
    for item_fn in list_r.iter() {
      $lambda(&item_fn);
    }
  };
}

/// The `$lambda` expression is `async`.
macro_rules! iterate_over_vec_with_async {
  ($this:ident, $locked_list_arc:expr, $lambda:expr) => {
    let locked_list = $locked_list_arc.get();
    let list_r = locked_list.read().await;
    for item_fn in list_r.iter() {
      $lambda(&item_fn).await;
    }
  };
}

pub(crate) use iterate_over_vec_with;
pub(crate) use iterate_over_vec_with_async;

Here’s a snippet of how this ends up being used in the Redux library. This snippet might not make sense at this point, its just there to give you a sense of how this macro is used. We will get into the details of all of this in the next sections.

pub async fn actually_dispatch_action<'a>(
  &mut self,
  action: &A,
) {
  // Run reducers.
  {
    iterate_over_vec_with!(
      self,
      self.reducer_manager,
      |reducer_fn: &'a ReducerFnWrapper<S, A>| {
        let new_state = reducer_fn.invoke(&self.state, &action);
        self.update_history(&new_state);
        self.state = new_state;
      }
    );
  }
}

With this, we can now dive into the Redux library implementation 🚀.

📦 There are ways to automate the generation of all this boilerplate in Rust using procedural macros. To see a real world example of using code generation to express this pattern, please check out the manager_of_things.rs file in our r3bl_rs_utils crate. You can also look at the tests to see how this macro is used.

🌟 Please star the r3bl-open-core repo on github if you like it 🙏.

Using the Redux library #

🚀 You can find a CLI app that uses this Redux library to manage an address book here called address_book_with_redux. Please clone that repo on github and run it using cargo run to play with this library and see what it can do. If you type help when the CLI starts up, it will give you a list of commands you can use. Try add-async and add-sync to see what happens.

The Store struct is the heart of the Redux library. It is the “thing” that we will use to manage our shared application state. It allows for a few things:

  1. Store creation.
  2. Dispatching actions.
  3. Getting the current state.
  4. Accessing history of state changes.
  5. Managing middleware.
  6. Managing reducers.
  7. Managing subscribers.

Any functions or blocks that you write which uses the Redux library will have to be marked async as well. And you will have to spawn the Tokio runtime by using the #[tokio::main] macro. If you use the default runtime then Tokio will use multiple threads and its task stealing implementation to give you parallel and concurrent behavior. You can also use the single threaded runtime; its really up to you.

📦 You can use this Redux library today by adding r3bl_rs_utils crate as a dependency in your Cargo.toml.

🌟 Please star the r3bl-open-core repo on github if you like it 🙏.

This is what the API looks like when we use it our app. Let’s say we have the following action enum, and state struct.

/// Action enum.
#[derive(Debug, PartialEq, Eq, Hash, Clone)]
pub enum Action {
  Add(i32, i32),
  AddPop(i32),
  Clear,
  MiddlewareCreateClearAction,
}

/// State.
#[derive(Clone, Default, PartialEq, Debug, Hash)]
pub struct State {
  pub stack: Vec<i32>,
}

Here’s an example of the reducer function.

// Reducer function (pure).
let reducer_fn = |state: &State, action: &Action| match action {
  Action::Add(a, b) => {
    let sum = a + b;
    State { stack: vec![sum] }
  }
  Action::AddPop(a) => {
    let sum = a + state.stack[0];
    State { stack: vec![sum] }
  }
  Action::Clear => State { stack: vec![] },
  _ => state.clone(),
};

Here’s an example of an async subscriber function (which are run in parallel after an action is dispatched). The following example uses a lambda that captures a shared object. This is a pretty common pattern that you might encounter when creating subscribers that share state in your enclosing block or scope.

// This shared object is used to collect results from the subscriber function
// & test it later.
let shared_object = Arc::new(Mutex::new(Vec::<i32>::new()));
// This subscriber function is curried to capture a reference to the shared object.
let subscriber_fn = with(shared_object.clone(), |it| {
  let curried_fn = move |state: State| {
    let mut stack = it.lock().unwrap();
    stack.push(state.stack[0]);
  };
  curried_fn
});

Here are two types of async middleware functions. One that returns an action (which will get dispatched once this middleware returns), and another that doesn’t return anything (like a logger middleware that just dumps the current action to the console). Note that both these functions share the shared_object reference from above.

// This middleware function is curried to capture a reference to the shared object.
let mw_returns_none = with(shared_object.clone(), |it| {
  let curried_fn = move |action: Action| {
    let mut stack = it.lock().unwrap();
    match action {
      Action::Add(_, _) => stack.push(-1),
      Action::AddPop(_) => stack.push(-2),
      Action::Clear => stack.push(-3),
      _ => {}
    }
    None
  };
  curried_fn
});

// This middleware function is curried to capture a reference to the shared object.
let mw_returns_action = with(shared_object.clone(), |it| {
  let curried_fn = move |action: Action| {
    let mut stack = it.lock().unwrap();
    match action {
      Action::MiddlewareCreateClearAction => stack.push(-4),
      _ => {}
    }
    Some(Action::Clear)
  };
  curried_fn
});

Here’s how you can setup a store with the above reducer, middleware, and subscriber functions.

// Setup store.
let mut store = Store::<State, Action>::new();
store
  .add_reducer(ReducerFnWrapper::new(reducer_fn))
  .await
  .add_subscriber(SafeSubscriberFnWrapper::new(subscriber_fn))
  .await
  .add_middleware(SafeMiddlewareFnWrapper::new(mw_returns_none))
  .await;

Finally here’s an example of how to dispatch an action in a test. You can dispatch actions asynchronously using dispatch_spawn() which is “fire and forget” meaning that the caller won’t block or wait for the dispatch_spawn() to return. Then you can dispatch actions synchronously if that’s what you would like using dispatch().

// Test reducer and subscriber by dispatching Add and AddPop actions asynchronously.
store.dispatch_spawn(Action::Add(10, 10)).await;
store.dispatch(&Action::Add(1, 2)).await;
assert_eq!(shared_object.lock().unwrap().pop(), Some(3));
store.dispatch(&Action::AddPop(1)).await;
assert_eq!(shared_object.lock().unwrap().pop(), Some(21));
store.clear_subscribers().await;

// Test `async` middleware: mw_returns_action.
shared_object.lock().unwrap().clear();
store
  .add_middleware(SafeMiddlewareFnWrapper::new(mw_returns_action))
  .dispatch(&Action::MiddlewareCreateClearAction)
  .await;
assert_eq!(store.get_state().stack.len(), 0);
assert_eq!(shared_object.lock().unwrap().pop(), Some(-4));

The Store struct #

The Store is actually a “manager” for the “thing” which is SafeStoreStateMachineWrapper. The “thing” that it manages actually has all the good stuff like the following:

  1. The current state.
  2. The history of states.
  3. The subscriber manager (which itself is a “manager” for the “thing” that is a function).
  4. The middleware manager (which itself is a “manager” for the “thing” that is a function).
  5. The reducer manager (which itself is a “manager” for the “thing” that is a function).

Here’s a code snippet that shows how to use the Store struct. In terms of “manager” and “thing” here’s the breakdown:

  1. “manager”: Store is a “manager” for the “thing” that is SafeStoreStateMachineWrapper.
  2. “thing”: SafeStoreStateMachineWrapper is a “thing” that holds the current state, the history of states, the subscriber manager, the middleware manager, and the reducer manager.
pub struct StoreStateMachine<S, A>
where
  S: Sync + Send + 'static,
  A: Sync + Send + 'static,
{
  pub state: S,
  pub history: Vec<S>,
  pub subscriber_manager: SubscriberManager<S>,
  pub middleware_manager: MiddlewareManager<A>,
  pub reducer_manager: ReducerManager<S, A>,
}

The Store itself has quite a few methods that allows to add/remove/dispatch/get/clear/etc. Here are some of these methods.

  1. dispatch and dispatchSpawn which allow us to dispatch action objects synchronously or asynchronously.
  2. add_subscriber and clear_subscribers which allow us to add/remove subscriber functions.
  3. add_middleware and clear_middlewares which allow us to add/remove middleware functions.
  4. add_reducer which allows us to add a reducer function.

Reducer functions #

With 0.7.12 of the library, the codebase has been refactored to make all the things async. Please read the README for details on these changes and how to build your own reducers. Instead of using function pointers, the new implementation uses async trait objects (which are much easier to reason about and create, and also can be made async).

These live in the SafeStoreStateMachineWrapper struct. The reducer functions are managed by a ReducerManager which is just a type alias for SafeListManager. Here’s the breakdown of this in terms of “manager” and “thing”.

  1. “manager”: SafeListManager which holds a list of functions.
  2. “thing”: a ReducerFnWrapper function wrapper that is safe to share (it’s actually a “manager” for the actual function that is the “thing”).
// Reducer manager.
pub type ReducerManager<S, A> = SafeListManager<ReducerFnWrapper<S, A>>;

💡 This SafeListManager is used all over the place. It is used to manage a vector of things, where the things could be other managers.

The reducer function itself (which is a “thing”) is wrapped in a ReducerFnWrapper which is a “manager”. Here’s the breakdown:

  1. “manager”: ReducerFnWrapper which manages a function.
  2. “thing”: SafeFnSafeReducerFn which wraps a function that takes an action and state and returns a new state.

The function wrapper itself is a “manager” for the “thing” that is a lambda that accepts an action and returns a new state.

/// Reducer function.
pub type ReducerFn<S, A> = dyn Fn(&S, &A) -> S;
pub type SafeReducerFn<S, A> = Arc<Mutex<dyn Fn(&S, &A) -> S + Send + Sync + 'static>>;

#[derive(Clone)]
pub struct ReducerFnWrapper<S, A>
where
  S: Sync + Send + 'static,
  A: Sync + Send + 'static,
{
  fn_mut: SafeReducerFn<S, A>,
}

When an action is dispatched to the store, each reducer function is called one after another, and the state is updated. These happen sequentially (not in separate Tokio tasks). Before the reducers are run the middleware functions are run.

Middleware functions #

With 0.7.12 of the library, the codebase has been refactored to make all the things async. Please read the README for details on these changes and how to build your own reducers. Instead of using function pointers, the new implementation uses async trait objects (which are much easier to reason about and create, and also can be made async).

These live in the SafeStoreStateMachineWrapper struct. The middleware functions are managed by a MiddlewareManager which is just a type alias for SafeListManager. Here’s the breakdown of this in terms of “manager” and “thing”.

  1. “manager”: SafeListManager which holds a list of functions.
  2. “thing”: a SafeMiddlewareFnWrapper function wrapper that is safe to share (it’s actually a “manager” for the actual function that is the “thing”).

The middleware function itself (which is a “thing”) is wrapped in a SafeMiddlewareFnWrapper which is a “manager”. Here’s the breakdown:

  1. “manager”: SafeMiddlewareFnWrapper which manages an async function.
  2. “thing”: SafeMiddlewareFn which wraps a function that takes an action and returns an option that can hold an action. This means that it can return None or Some containing an action.
use std::{
  marker::{Send, Sync},
  sync::Arc,
};
use tokio::{task::JoinHandle, sync::RwLock};

pub type SafeMiddlewareFn<A> = Arc<RwLock<dyn FnMut(A) -> Option<A> + Sync + Send>>;
//                             ^^^^^^^^^^                             ^^^^^^^^^^^
//                             Safe to pass      Declare`FnMut` has thread safety
//                             around.           requirement to rust compiler.

#[derive(Clone)]
pub struct SafeMiddlewareFnWrapper<A> {
  fn_mut: SafeMiddlewareFn<A>,
}

💡 Here are some excellent resources on lifetimes and returning references:

  1. https://stackoverflow.com/questions/59442080/rust-pass-a-function-reference-to-threads
  2. https://stackoverflow.com/questions/68547268/cannot-borrow-data-in-an-arc-as-mutable
  3. https://willmurphyscode.net/2018/04/25/fixing-a-simple-lifetime-error-in-rust/

When an action is dispatched, each middleware is run in its own Tokio task. Then the store waits for all of these to finish. If any actions have been created then these are queued up and passed to the reducer functions (which actually generate new states). These reducers are run sequentially. When the final state is ready, it is then passed to each subscriber function; each subscriber function is run in a separate Tokio task).

Subscriber functions #

With 0.7.12 of the library, the codebase has been refactored to make all the things async. Please read the README for details on these changes and how to build your own reducers. Instead of using function pointers, the new implementation uses async trait objects (which are much easier to reason about and create, and also can be made async).

These live in the SafeStoreStateMachineWrapper struct. The subscriber functions are managed by a SubscriberManager which is just a type alias for SafeListManager. Here’s the breakdown of this in terms of “manager” and “thing”.

  1. “manager”: SafeListManager which holds a list of functions.
  2. “thing”: a SafeSubscriberFnWrapper function wrapper that is safe to share (it’s actually a “manager” for the actual function that is the “thing”).

The middleware function itself (which is a “thing”) is wrapped in a SafeSubscriberFnWrapper which is a “manager”. Here’s the breakdown:

  1. “manager”: SafeSubscriberFnWrapper which manages an async function.
  2. “thing”: SafeSubscriberFn which wraps a function that takes a state and returns nothing.
use std::sync::Arc;
use tokio::{task::JoinHandle, sync::RwLock};

/// Subscriber function.
pub type SafeSubscriberFn<S> = Arc<RwLock<dyn FnMut(S) + Sync + Send>>;

#[derive(Clone)]
pub struct SafeSubscriberFnWrapper<S> {
  fn_mut: SafeSubscriberFn<S>,
}

The last stage of action dispatch is to call each subscriber function. These happen in their own Tokio tasks.

Wrapping up #

🚀 You can find a CLI app that uses this Redux library to manage an address book here called address_book_with_redux. Please clone that repo on github and run it using cargo run to play with this library and see what it can do. If you type help when the CLI starts up, it will give you a list of commands you can use. Try add-async and add-sync to see what happens.

📜 The source code for the finished Redux library can be found here. You can always find the most up to date information on this library in it’s README file.

🌟 Please star the r3bl-open-core repo. You can get lots of useful Rust command line apps from the r3bl-cmdr crate in this repo:

  • 🐱giti: run interactive git commands with confidence in your terminal
  • 🦜edi: edit Markdown with style in your terminal
📦 You can install them using cargo install r3bl-cmdr

giti in action edi in action

If you would like to get involved in an open source project and like Rust crates, we welcome your contributions to the r3bl-open-core repo.

Related Posts