Update: We have created an open-source npm package called ngx-signal-state that can you can use in all your Angular projects. This is only relevant if you are on Angular 16 or higher.

What is state management? This is a very popular discussion topic in most teams. What is state? When do we manage it? What state do we need to manage? What is managing state? This article is about component state, which is the state that we keep inside our component.

This article is a follow-up article of the previous articles (newest to oldest):

Current way of handling ViewModels

In the last article, we learned how we could create a ViewModel for Angular by calculating the input state in a one-liner:

export class PagerComponent {
    @Input() public itemsPerPage: number = 0;
    @Input() public total: number = 0;
    @Input() public pageIndex: number = 0;

    // One-liner!
    @InputState() private readonly inputState$!: Observable<PagerInputState>;

    public readonly vm$: Observable<ViewModel> = this.inputState$.pipe(
        map(({ itemsPerPage, total, pageIndex }) => calculateVm)
    );
}

While this is convenient, it is not perfect just yet… There are still some issues:

  • When we would set the @Input() properties from within this class, the reactive input state would not update when setting them.
  • We have a way to handle the input state reactively but the component usually also has other states (in many situations these are handled with a number of BehaviorSubjects or regular object properties).
  • It would be great to have one Observable state instance that lives along with this component and gets destroyed when this component gets destroyed. (Not having to implement the ngOnDestroy() life cycle hook ourselves.)
  • There is no way to get a current snapshot of the states (other than subject$$.value which can feel hacky).

Component state

The component state does not need to be compared to an application-wide or even module-wide state. The state doesn’t need to be shared between components or even libs in our codebase.

When we think about it, every component has state:

  • A property when a pane is collapsed or not
  • An observable holding a list of users coming from the backend
  • A FormControl containing some kind of value
  • The current page index of a pager
  • The params of the ActivatedRoute that we will use to determine the outcome of the application

Definition component state: Component state contains everything that determines the outcome of your page.

Definition ViewModel: A ViewModel is a reactive object that contains all properties and only the properties that the template of a component needs, it is a calculation based on the component state.

Every time an Angular (or React, Vue, Qwik, …) application is written, the components have state. It’s just a matter of how we manage it. Some (uncontrolled and not opinionated) ways are:

  • Managing plain objects in private and public properties, some read-only, some not.
  • Managing ReplaySubjects, BehaviorSubjects and other kind of Observables in private and public properties.
  • Extracting all the state or parts of the state into stores and making sure these states get invalidated at the right time.
  • In more complex components this can result in complex RxJS logic.

In this article we suggest an opinionated and structured way that works for us:

  • All the component state is kept in an observable state object.
  • That observable state object holds a BehaviorSubject behind the scenes so we can always take snapshots when we like.
  • That observable state is provided on the component and will thus get destroyed if the component gets destroyed.
  • That observable state will automatically be fed by the Input state we were talking about earlier.
  • Setting state and reading state should always be done through this reactive state object:
    • Future proof if we want to go zoneless later
    • Consistent
    • Clear
    • Separation of concerns

In this article, we will focus on component state for ui components. If you don’t know the difference between smart and ui components you could read this article.

Why not use an existing library for this?

There are some libraries out there that give us some of the functionality that we are looking for. After doing some research we still decided to write it ourselves. One of the best practices we teach at our training is: Be careful which and how many dependencies you install and especially be careful of Angular dependencies. Every dependency is a tiny marriage, a tiny contract that makes you attached to the dependency in some way. We are suggesting a different approach today: Learn why we are using this approach by reading these articles, Copy-paste the code (it is really small), and maintain it yourself. We could open-source this but we don’t have time to maintain it, nor will we add features for you. We will update these articles though.

This is why we wrote it ourselves:

  • It’s not that much code
  • It’s easy to write
  • Copy-paste it and maintain it ourselves. Our code is free for you to use and it’s not that complex
  • Don’t be blocked by new Angular versions. Dependencies are not always up to date with the latest Angular versions
  • Stay close to Angular, to the standard.
  • Add features ourselves when we feel we need them.
  • Do we want a third-party dependency to infect every component in our codebase?

The pager component

The pager we have written in the previous article is fully functional, but let’s add some local state to introduce our first observable state. In the following example, we have added a button that can show a dropdown of items per page. Whether this dropdown is shown or not is considered local state and listens to the moreOptions property which defaults to false. (dropdown is hidden by default) Another piece of state is the itemsPerPageOptions property which contains a list of numbers that the dropdown can contain.

export class PagerComponent {
    public moreOptions = false; // Local state
    public itemsPerPageOptions = [10,20,50,100]; // Local state

    // Input state calculated by the inputs
    @InputState() private readonly inputState$!: Observable<PagerInputState>;

    @Input() public itemsPerPage: number = 0;
    @Input() public total: number = 0;
    @Input() public pageIndex: number = 0;
    @Output() public readonly pageIndexChange = new EventEmitter<number>();
    @Output() public readonly itemsPerPageChange = new EventEmitter<number>();

    public readonly vm$: Observable<ViewModel> = this.inputState$.pipe(...);

    ...

    public next(vm: ViewModel): void {
        this.pageIndexChange.emit(vm.pageIndex + 1);
    }

    public previous(vm: ViewModel): void {
        this.pageIndexChange.emit(vm.pageIndex - 1);
    }

    public goToEnd(vm: ViewModel): void {
        this.pageIndexChange.emit(Math.ceil(vm.total / vm.itemsPerPage) - 1);
    }

    // Toggle the local state
    public toggleMoreOptions(): void {
        this.moreOptions = !this.moreOptions;
    }
}

You can check the Stackblitz example here

Some issues

Before we continue to improve this code let’s pinpoint some issues:

  • Setting @Input() properties will not update the inputState and thus will not update the ViewModel.
  • To make the moreOptions part of the ViewModel we have to create a BehaviorSubject otherwise it won’t be reactive.
  • We have a partially observable state (input state) and partially plain objects (not observable).
  • The template should only consume state from the ViewModel so the moreOptions and itemsPerPageOptions properties should not be public members on the class instance but be exposed through the ViewModel.
  • The toggleMoreOptions() implementation would not work in a zone-less solution since it is not reactive.
  • We want one state object that we can talk to (one single source of truth regarding state).
  • The next, previous and goToEnd all get passed the ViewModel as a parameter so they would get access to the updated properties and we even put the pageIndex and itemsPerPage on the ViewModel for convenience reasons, even though the template doesn’t need those.

Refactor the pager component

We will dive into the implementation of the observable state later but we already know we want to provide an instance of it on the PagerComponent and we want to inject it as well. That way we can implement automatic teardown logic since the observable state instance will get destroyed together with the instance of the PagerComponent. Let’s update the ViewModel and PagerInputState types as well and let’s create a new type called PagerState which extends the PagerInputState (input state will always be part of the state).

type ViewModel = {
    itemFrom: number;
    itemTo: number;
    total: number;
    previousDisabled: boolean;
    nextDisabled: boolean;
    moreOptions: boolean; // new property
    itemsPerPageOptions: number[]; // new property
    // pageIndex: number; (shouldn't be part of ViewModel)
    // itemsPerPage: number; (shouldn't be part of ViewModel)
};

// Covered in previous article
type PagerInputState = {
    itemsPerPage: number;
    total: number;
    pageIndex: number;
};

// Combination of PagerInputState and local component state
type PagerState = PagerInputState & {
    moreOptions: boolean;
    itemsPerPageOptions: number[];
}

The next thing we want to do is initialize the state in the constructor. Since we want to only talk to the state in the future we don’t want to set Input() properties ourselves anymore. Since the state will be a BehaviorSubject behind the scenes, we will need to pass it an initial state. The initialize() function of our component will take the initial state and an optional input state observable:

// extend from ObservableState
export class PagerComponent extends ObservableState<PagerState> {
    // Created in previous article
    @InputState() private readonly inputState$!: Observable<PagerInputState>;
    
    // we can update them through this.patch({})
    @Input() public itemsPerPage: number = 0;
    @Input() public total: number = 0;
    @Input() public pageIndex: number = 0;

    ...

    constructor(){
        super();
        const {itemsPerPage, total, pageIndex} = this;
        // Determine the initial state
        const initialState: PagerState = {
            moreOptions: false, // Local state
            itemsPerPageOptions: [10,20,50,100], // Local state
            itemsPerPage,
            total,
            pageIndex
        }
        this.initialize(initialState, this.inputState$)
    }
    ...
}

The next thing we want to do is tackle the calculation of the ViewModel. This will not be calculated on this.inputState$ anymore but it will get calculated on this.state$, which will hold all the state of the component. One single source of truth of state for this component only. We can also see that the itemsPerPage and pageIndex aren’t needed anymore on the ViewModel (they are available through the observable state and they were never meant to be part of the ViewModel anyway since the template never needed access to it). On the other hand, 2 more options are added to our ViewModel: moreOptions and itemsPerPageOptions. Since they are part of the state they are reactive by default and could be updated through a patch() method.

public readonly vm$: Observable<ViewModel> = this.state$.pipe(
    map(({ itemsPerPage, total, pageIndex, moreOptions,itemsPerPageOptions }) => {
        return {
        total,
        // itemsPerPage, (not needed anymore)
        // pageIndex, (not needed anymore)
        previousDisabled: pageIndex === 0,
        nextDisabled: pageIndex >= Math.ceil(total / itemsPerPage) - 1,
        itemFrom: pageIndex * itemsPerPage + 1,
        itemTo:
            pageIndex < Math.ceil(total / itemsPerPage) - 1
            ? pageIndex * itemsPerPage + itemsPerPage
            : total,
        moreOptions, // New
        itemsPerPageOptions // New
        };
    })
);

The template does not need to pass vm to the methods next(), previous() and goToEnd() anymore. The class instance now has access to the component state. The observable state gives us a convenient snapshot that we can consume at any time. The component is reactive but at the same time can give us a snapshot so we don’t have to subscribe all the time:

public next(): void {
    // Take what we need from the snapshot
    const {pageIndex} = this.snapshot;
    this.pageIndexChange.emit(pageIndex + 1);
}

public previous(): void {
    // Take what we need from the snapshot
    const {pageIndex} = this.snapshot;
    this.pageIndexChange.emit(pageIndex - 1);
}

public goToEnd(): void {
    // Take what we need from the snapshot
    const {total, itemsPerPage} = this.snapshot;
    this.pageIndexChange.emit(Math.ceil(total / itemsPerPage) - 1);
}

Since the moreOptions and itemsPerPageOptions are now living on the ViewModel we can consume them easily in a reactive way:

<button ...>
    {{vm.moreOptions? 'Less': 'More'}}
</button>
<select ...
    *ngIf="vm.moreOptions">
    <option 
        ...
        *ngFor="let option of vm.itemsPerPageOptions">...
    </option>
</select>

The last thing we need is to implement the toggleMoreOptions() function. For that we can use the patch() method that takes an object that will update only the parts of the state that need to get updated:

public toggleMoreOptions(): void {
    const {moreOptions} = this.snapshot;
    this.patch({moreOptions: !moreOptions});
}

What did we solve?

Before continuing on to the actual implementation of the observable state, let’s summarize what we have fixed.

  • We have a reactive state model with a snapshot that is updated every time an input changes.
  • The reactive state model can be patched manually.
  • We have a clear ViewModel that does not contain anything but the data our template needs.
  • The template does not need to pass the ViewModel back to the class instance.
  • We have a single source of truth when it comes to handling state.
  • We can consume the input state observable easily in the initialize() method.
  • Our local state is also reactive and consumable through the ViewModel
  • We separated ViewModel logic from state logic completely
  • We don’t have to worry about implementing the ngOnDestroy() life cycle hook.

Implementing the Observable state

In this specific article, we will have a look at a limited version of the observable state. You can find the entire implementation here.

Let’s summarize what kind of functionality this observable state already holds for us:

  • It should clean itself up, when the instance that provides the ObservableState gets destroyed it should destroy the ObservableState as well.
  • It should hold a BehaviorSubject behind the scenes that is exposed as an observable that only emits when the state is initialized.
  • It should distinctUntilChanged() on all the properties and thus only emit when one of the values has actually changed.
  • It should expose a snapshot of the current state.
  • The intialize() method should take the default state but should also take an input state observable as a second argument and feed the state accordingly in a safe way (no memory leaks).
  • It should be completely type-safe.
  • It should be possible to patch values as a partial object so that there is only one emission on the state object every time we call patch() and not for every key.

Note: It’s important that we always update @Input() properties through the patch function since updating the inputs directly will not result in InputState updates. Making them readonly is not an option since this will prevent the parent component from using this component.

What we will not implement today

The implementation of this article is limited to the set of features listed above and will not contain:

  • connect({users: users$}): a method to connect multiple observables to our observable state.
  • onlySelect(['foo', 'bar']): a method that will return an observable that only is updated with and when one of the passed keys of our state gets updated.

const filterAndCastToT = <T>() =>
  pipe(
    filter((v: T | null) => v !== null),
    map((v) => v as T)
  );

// we need this dirty fix because of an issue with BehaviorSubject
// queuescheduler didn't cut it
export class StateSubject<T> extends BehaviorSubject<T> {
  public readonly syncState = this.asObservable().pipe(map(() => this.value)) as Observable<T>
}

@Injectable()
export class ObservableState<T extends Record<string, unknown>>
  implements OnDestroy {
  private readonly notInitializedError =
    'Observable state is not initialized yet, call the initialize() method';
  private readonly destroy$$ = new Subject<void>();
  private readonly state$$ = new StateSubject<T | null>(null);
  /**
   * Return the entire state as an observable
   * Only use this if you want to be notified on every update. For better optimization
   * use the onlySelectWhen() method
   * where we can pass keys on when to notify.
   */
  public readonly state$ = this.state$$.syncState.pipe(
    filterAndCastToT<T>(),
    distinctUntilChanged((previous: T, current: T) =>
      Object.keys(current).every(
        (key: string) => current[key as keyof T] === previous[key as keyof T]
      )
    ),
    takeUntil(this.destroy$$)
  );

  /**
   * Get a snapshot of the current state. This method is needed when we want to fetch the
   * state in functions. We don't have to use withLatestFrom if we want to keep it simple.
   */
  public get snapshot(): T {
    if (!this.state$$.value) {
      throw new Error(this.notInitializedError);
    }
    return this.state$$.value as T;
  }

  /**
   * Observable state doesn't work without initializing it first. Our state always needs
   * an initial state. You can pass the @InputState() as an optional parameter.
   * Passing that @InputState() will automatically feed the state with the correct values
   * @param state
   * @param inputState$
   */
  public initialize(state: T, inputState$?: Observable<Partial<T>>): void {
    this.state$$.next(state); // pass initial state
    // Feed the state when the input state gets a new value
    inputState$?.pipe(
      takeUntil(this.destroy$$)
    ).subscribe((res: Partial<T>) => this.patch(res));
  }

  /**
   * Patch a partial of the state. It will loop over all the properties of the passed
   * object and only next the state once.
   * @param object
   */
  public patch(object: Partial<T>): void {
    if (!this.state$$.value) {
      throw new Error(this.notInitializedError);
    }
    let newState: T = { ...this.state$$.value };
    Object.keys(object).forEach((key: string) => {
      newState = { ...newState, [key]: object[key as keyof T] };
    });
    this.state$$.next(newState);
  }

  public ngOnDestroy(): void {
    this.destroy$$.next();
    this.destroy$$.complete();
  }
}

Here is an updated Stackblitz example that contains everything you need for these kinds of components.

Conclusion

In this article, the observable state implementation is not yet complete, but we will update it in the next article where we will use it for more complex state management in smart components. Stay tuned! Check out this article if you want to learn some Angular State management Best practices

Update: We have created an open-source npm package called ngx-signal-state that can you can use in all your Angular projects.

If you like to learn directly from me, check out my Angular Training and Angular Coaching