Modularizing React Applications with Established UI Patterns

Apart from the user interface

React itself doesn’t care much about where to put calculation or business logic, which is fair as it’s only a library for building user interfaces. And beyond that view layer, a frontend application has other parts as well. To make the application work, you will need a router, local storage, cache at different levels, network requests, 3rd-party integrations, 3rd-party login, security, logging, performance tuning, etc.

With all this extra context, trying to squeeze everything into React components or hooks is generally not a good idea. The reason is mixing concepts in one place generally leads to more confusion. At first, the component sets up some network request for order status, and then there is some logic to trim off leading space from a string and then navigate somewhere else. The reader must constantly reset their logic flow and jump back and forth from different levels of details.

Packing all the code into components may work in small applications like a Todo or one-form application. Still, the efforts to understand such application will be significant once it reaches a certain level. Not to mention adding new features or fixing existing defects.

If we could separate different concerns into files or folders with structures, the mental load required to understand the application would be significantly reduced. And you only have to focus on one thing at a time. Luckily, there are already some well-proven patterns back to the pre-web time. These design principles and patterns are explored and discussed well to solve the common user interface problems - but in the desktop GUI application context.

Martin Fowler has a great summary of the concept of view-model-data layering.

On the whole I've found this to be an effective form of modularization for many applications and one that I regularly use and encourage. It's biggest advantage is that it allows me to increase my focus by allowing me to think about the three topics (i.e., view, model, data) relatively independently.

-- Martin Fowler

Layered architectures have been used to cope the challenges in large GUI applications, and certainly we can use these established patterns of front-end organization in our “React applications”.

Single Component Application

It can be called pretty much a Single Component Application:

Figure 1: Single Component Application

But soon, you realise one single component requires a lot of time just to read what is going on. For example, there is logic to iterate through a list and generate each item. Also, there is some logic for using 3rd-party components with only a few configuration code, apart from other logic.

Multiple Component Application

You decided to split the component into several components, with these structures reflecting what’s happening on the result HTML is a good idea, and it helps you to focus on one component at a time.

Figure 2: Multiple Component Application

And as your application grows, apart from the view, there are things like sending network requests, converting data into different shapes for the view to consume, and collecting data to send back to the server. And having this code inside components doesn’t feel right as they’re not really about user interfaces. Also, some components have too many internal states.

State management with hooks

It’s a better idea to split this logic into a separate places. Luckily in React, you can define your own hooks. This is a great way to share these state and the logic of whenever states change.

Figure 3: State management with hooks

That’s awesome! You have a bunch of elements extracted from your single component application, and you have a few pure presentational components and some reusable hooks that make other components stateful. The only problem is that in hooks, apart from the side effect and state management, some logic doesn’t seem to belong to the state management but pure calculations.

Business models emerged

So you’ve started to become aware that extracting this logic into yet another place can bring you many benefits. For example, with that split, the logic can be cohesive and independent of any views. Then you extract a few domain objects.

These simple objects can handle data mapping (from one format to another), check nulls and use fallback values as required. Also, as the amount of these domain objects grows, you find you need some inheritance or polymorphism to make things even cleaner. Thus you applied many design patterns you found helpful from other places into the front-end application here.

Figure 4: Business models

Layered frontend application

The application keeps evolving, and then you find some patterns emerge. There are a bunch of objects that do not belong to any user interface, and they also don’t care about whether the underlying data is from remote service, local storage or cache. And then, you want to split them into different layers. Here is a detailed explanation about the layer splitting Presentation Domain Data Layering.

Figure 5: Layered frontend application

The above evolution process is a high-level overview, and you should have a taste of how you should structure your code or at least what the direction should be. However, there will be many details you need to consider before applying the theory in your application.

In the following sections, I’ll walk you through a feature I extracted from a real project to demonstrate all the patterns and design principles I think useful for big frontend applications.

The problem with the initial implementation

The first issue I would like to address is how busy the component is. By that, I mean Payment deals with different things and makes the code difficult to read as you have to switch context in your head as you read.

In order to make any changes you have to comprehend how to initialise network request , how to map the data to a local format that the component can understand , how to render each payment method , and the rendering logic for Payment component itself .

src/Payment.tsx…

  export const Payment = ({ amount }: { amount: number }) => {
    const [paymentMethods, setPaymentMethods] = useState<LocalPaymentMethod[]>(
      []
    );
  
    useEffect(() => {
      const fetchPaymentMethods = async () => {
        const url = "https://online-ordering.com/api/payment-methods";
  
        const response = await fetch(url);
        const methods: RemotePaymentMethod[] = await response.json();
  
        if (methods.length > 0) {
          const extended: LocalPaymentMethod[] = methods.map((method) => ({
            provider: method.name,
            label: `Pay with ${method.name}`,
          }));
          extended.push({ provider: "cash", label: "Pay in cash" });
          setPaymentMethods(extended);
        } else {
          setPaymentMethods([]);
        }
      };
  
      fetchPaymentMethods();
    }, []);
  
    return (
      <div>
        <h3>Payment</h3>
        <div>
          {paymentMethods.map((method) => (
            <label key={method.provider}>
              <input
                type="radio"
                name="payment"
                value={method.provider}
                defaultChecked={method.provider === "cash"}
              />
              <span>{method.label}</span>
            </label>
          ))}
        </div>
        <button>${amount}</button>
      </div>
    );
  };

It's not a big problem at this stage for this simple example. However, as the code gets bigger and more complex, we'll need to refactoring them a bit.

It’s good practice to split view and non-view code into separate places. The reason is, in general, views are changing more frequently than non-view logic. Also, as they deal with different aspects of the application, separating them allows you to focus on a particular self-contained module that is much more manageable when implementing new features.

The split of view and non-view code

In React, we can use a custom hook to maintain state of a component while keeping the component itself more or less stateless. We can use Extract Function to create a function called usePaymentMethods (the prefix use is a convention in React to indicate the function is a hook and handling some states in it):

src/Payment.tsx…

  const usePaymentMethods = () => {
    const [paymentMethods, setPaymentMethods] = useState<LocalPaymentMethod[]>(
      []
    );
  
    useEffect(() => {
      const fetchPaymentMethods = async () => {
        const url = "https://online-ordering.com/api/payment-methods";
  
        const response = await fetch(url);
        const methods: RemotePaymentMethod[] = await response.json();
  
        if (methods.length > 0) {
          const extended: LocalPaymentMethod[] = methods.map((method) => ({
            provider: method.name,
            label: `Pay with ${method.name}`,
          }));
          extended.push({ provider: "cash", label: "Pay in cash" });
          setPaymentMethods(extended);
        } else {
          setPaymentMethods([]);
        }
      };
  
      fetchPaymentMethods();
    }, []);
  
    return {
      paymentMethods,
    };
  };

This returns a paymentMethods array (in type LocalPaymentMethod) as internal state and is ready to be used in rendering. So the logic in Payment can be simplified as:

src/Payment.tsx…

  export const Payment = ({ amount }: { amount: number }) => {
    const { paymentMethods } = usePaymentMethods();
  
    return (
      <div>
        <h3>Payment</h3>
        <div>
          {paymentMethods.map((method) => (
            <label key={method.provider}>
              <input
                type="radio"
                name="payment"
                value={method.provider}
                defaultChecked={method.provider === "cash"}
              />
              <span>{method.label}</span>
            </label>
          ))}
        </div>
        <button>${amount}</button>
      </div>
    );
  };

This helps relieve the pain in the Payment component. However, if you look at the block for iterating through paymentMethods, it seems a concept is missing here. In other words, this block deserves its own component. Ideally, we want each component to focus on, only one thing.

Split the view by extracting sub component

Also, if we can make a component a pure function - meaning given any input, the output is certain - that would help us a lot in writing tests, understanding the code and even reusing the component elsewhere. After all, the smaller a component, the more likely it will be reused.

We can use Extract Function again (maybe we should call it “Extract Component”, but in React, a component is a function anyway).

src/Payment.tsx…

  const PaymentMethods = ({
    paymentMethods,
  }: {
    paymentMethods: LocalPaymentMethod[];
  }) => (
    <>
      {paymentMethods.map((method) => (
        <label key={method.provider}>
          <input
            type="radio"
            name="payment"
            value={method.provider}
            defaultChecked={method.provider === "cash"}
          />
          <span>{method.label}</span>
        </label>
      ))}
    </>
  );

The Payment component can use the PaymentMethods directly and thus be simplified as below:

src/Payment.tsx…

  export const Payment = ({ amount }: { amount: number }) => {
    const { paymentMethods } = usePaymentMethods();
  
    return (
      <div>
        <h3>Payment</h3>
        <PaymentMethods paymentMethods={paymentMethods} />
        <button>${amount}</button>
      </div>
    );
  };

Note that PaymentMethods is a pure function (a pure component) that doesn’t have any state. It’s basically a string formatting function.

Data modelling to encapsulate logic

So far, the changes we have made are all about splitting view and non-view code into different places. It works well. The hook handles data fetching and reshaping. Both Payment and PaymentMethods are relatively small and easy to understand.

However, if you look closely, there is still room for improvement. To start with, in the pure function component PaymentMethods, we have a bit of logic to check if a payment method should be checked by default:

src/Payment.tsx…

  const PaymentMethods = ({
    paymentMethods,
  }: {
    paymentMethods: LocalPaymentMethod[];
  }) => (
    <>
      {paymentMethods.map((method) => (
        <label key={method.provider}>
          <input
            type="radio"
            name="payment"
            value={method.provider}
            defaultChecked={method.provider === "cash"}
          />
          <span>{method.label}</span>
        </label>
      ))}
    </>
  );

These test statements in a view can be considered a logic leak, and gradually they can be scatted in different places and make modification harder.

Another point of potential logic leakage is in the data conversion where we fetch data:

src/Payment.tsx…

  const usePaymentMethods = () => {
    const [paymentMethods, setPaymentMethods] = useState<LocalPaymentMethod[]>(
      []
    );
  
    useEffect(() => {
      const fetchPaymentMethods = async () => {
        const url = "https://online-ordering.com/api/payment-methods";
  
        const response = await fetch(url);
        const methods: RemotePaymentMethod[] = await response.json();
  
        if (methods.length > 0) {
          const extended: LocalPaymentMethod[] = methods.map((method) => ({
            provider: method.name,
            label: `Pay with ${method.name}`,
          }));
          extended.push({ provider: "cash", label: "Pay in cash" });
          setPaymentMethods(extended);
        } else {
          setPaymentMethods([]);
        }
      };
  
      fetchPaymentMethods();
    }, []);
  
    return {
      paymentMethods,
    };
  };

Note the anonymous function inside methods.map does the conversion silently, and this logic, along with the method.provider === “cash” above can be extracted into a class.

We could have a class PaymentMethod with the data and behaviour centralised into a single place:

src/PaymentMethod.ts…

  class PaymentMethod {
    private remotePaymentMethod: RemotePaymentMethod;
  
    constructor(remotePaymentMethod: RemotePaymentMethod) {
      this.remotePaymentMethod = remotePaymentMethod;
    }
  
    get provider() {
      return this.remotePaymentMethod.name;
    }
  
    get label() {
      if(this.provider === 'cash') {
        return `Pay in ${this.provider}`
      }
      return `Pay with ${this.provider}`;
    }
  
    get isDefaultMethod() {
      return this.provider === "cash";
    }
  }

With the class, I can define the default cash payment method:

const payInCash = new PaymentMethod({ name: "cash" });

And during the conversion - after the payment methods are fetched from the remote service - I can construct the PaymentMethod object in-place. Or even extract a small function called convertPaymentMethods:

src/usePaymentMethods.ts…

  const convertPaymentMethods = (methods: RemotePaymentMethod[]) => {
    if (methods.length === 0) {
      return [];
    }
  
    const extended: PaymentMethod[] = methods.map(
      (method) => new PaymentMethod(method)
    );
    extended.push(payInCash);
  
    return extended;
  };

Also, in the PaymentMethods component, we don’t use the method.provider === “cash”to check anymore, and instead call the getter:

src/PaymentMethods.tsx…

  export const PaymentMethods = ({ options }: { options: PaymentMethod[] }) => (
    <>
      {options.map((method) => (
        <label key={method.provider}>
          <input
            type="radio"
            name="payment"
            value={method.provider}
            defaultChecked={method.isDefaultMethod}
          />
          <span>{method.label}</span>
        </label>
      ))}
    </>
  );

Now we’re restructuring our Payment component into a bunch of smaller parts that work together to finish the work.

Figure 7: Refactored Payment with more parts that can be composed easily

The benefits of the new structure

• Having a class encapsulates all the logic around a payment method. It’s a domain object and doesn’t have any UI-related information. So testing and potentially modifying logic here is much easier than when embedded in a view.
• The new extracted component PaymentMethods is a pure function and only depends on a domain object array, which makes it super easy to test and reuse elsewhere. We might need to pass in a onSelect callback to it, but even in that case, it’s a pure function and doesn’t have to touch any external states.
• Each part of the feature is clear. If a new requirement comes, we can navigate to the right place without reading all the code.

I have to make the example in this article sufficiently complex so that many patterns can be extracted. All these patterns and principles are there to help simplify our code's modifications.

Internal state: agree to donation

To make these changes in Payment, we need a boolean state agreeToDonate to indicate whether a user selected the checkbox on the page.

src/Payment.tsx…

  const [agreeToDonate, setAgreeToDonate] = useState<boolean>(false);

  const { total, tip } = useMemo(
    () => ({
      total: agreeToDonate ? Math.floor(amount + 1) : amount,
      tip: parseFloat((Math.floor(amount + 1) - amount).toPrecision(10)),
    }),
    [amount, agreeToDonate]
  );

The function Math.floor will round the number down so we can get the correct amount when the user selects agreeToDonate, and the difference between the rounded-up value and the original amount will be assigned to tip.

And for the view, the JSX will be a checkbox plus a short description:

src/Payment.tsx…

  return (
    <div>
      <h3>Payment</h3>
      <PaymentMethods options={paymentMethods} />
      <div>
        <label>
          <input
            type="checkbox"
            onChange={handleChange}
            checked={agreeToDonate}
          />
          <p>
            {agreeToDonate
              ? "Thanks for your donation."
              : `I would like to donate $${tip} to charity.`}
          </p>
        </label>
      </div>
      <button>${total}</button>
    </div>
  );

With these new changes, our code starts handling multiple things again. It’s essential to stay alert for potential mixing of view and non-view code. If you find any unnecessary mixing, look for ways to split them.

Note that it's not a set-in-stone rule. Keep things all together nice and tidy for small and cohesive components, so you don't have to look in multiple places to understand the overall behaviour. Generally, you should be aware to avoid the component file growing too big to comprehend.

Extract a hook to the rescue

Here we need an object to calculate the tip and amount, and whenever a user changes their mind, the object should return the updated amount and tip.

So it sounds like we need an object that:

• takes the original amount as input
• returns total and tip whenever agreeToDonate changed.

It sounds like a perfect place for a custom hook again, right?

src/hooks/useRoundUp.ts…

  export const useRoundUp = (amount: number) => {
    const [agreeToDonate, setAgreeToDonate] = useState<boolean>(false);
  
    const {total, tip} = useMemo(
      () => ({
        total: agreeToDonate ? Math.floor(amount + 1) : amount,
        tip: parseFloat((Math.floor(amount + 1) - amount).toPrecision(10)),
      }),
      [amount, agreeToDonate]
    );
  
    const updateAgreeToDonate = () => {
      setAgreeToDonate((agreeToDonate) => !agreeToDonate);
    };
  
    return {
      total,
      tip,
      agreeToDonate,
      updateAgreeToDonate,
    };
  };

And in the view, we can call this hook with the initial amount and have all these states defined externally. The updateAgreeToDonate function can update the value in the hook and trigger a re-render.

src/components/Payment.tsx…

  export const Payment = ({ amount }: { amount: number }) => {
    const { paymentMethods } = usePaymentMethods();
  
    const { total, tip, agreeToDonate, updateAgreeToDonate } = useRoundUp(amount);
  
    return (
      <div>
        <h3>Payment</h3>
        <PaymentMethods options={paymentMethods} />
        <div>
          <label>
            <input
              type="checkbox"
              onChange={updateAgreeToDonate}
              checked={agreeToDonate}
            />
            <p>{formatCheckboxLabel(agreeToDonate, tip)}</p>
          </label>
        </div>
        <button>${total}</button>
      </div>
    );
  };

Note that We can also extract the message formatting part into the helper function formatCheckboxLabel to simplify the code in component.

const formatCheckboxLabel = (agreeToDonate: boolean, tip: number) => {
  return agreeToDonate
    ? "Thanks for your donation."
    : `I would like to donate $${tip} to charity.`;
};

And the Payment component can be simplified a lot - the states are now fully managed in hook useRoundUp.

You can imagine a hook as a state machine behind a view whenever some change happens in the UI, say, a checkbox change event. The event will be sent to the state machine to generate a new state, and the new state will trigger a re-render.

So the pattern here is that we should move state management away from a component and try to make it a presentational function (so it can be easily tested and reused just like these humble utility functions). The React hook was designed to share reusable logic from different components, but I find it beneficial even when there is only one use as it helps you to focus on rendering in a component and keeping state and data in hooks.

As the donation checkbox becomes more independent, we can move it into its own pure function component.

src/components/DonationCheckbox.tsx…

  const DonationCheckbox = ({
    onChange,
    checked,
    content,
  }: DonationCheckboxProps) => {
    return (
      <div>
        <label>
          <input type="checkbox" onChange={onChange} checked={checked} />
          <p>{content}</p>
        </label>
      </div>
    );
  };

While in Payment, thanks to the declarative UI in React, it’s pretty straightforward to read the code like a humble piece of HTML.

src/components/Payment.tsx…

  export const Payment = ({ amount }: { amount: number }) => {
    const { paymentMethods } = usePaymentMethods();
  
    const { total, tip, agreeToDonate, updateAgreeToDonate } = useRoundUp(amount);
  
    return (
      <div>
        <h3>Payment</h3>
        <PaymentMethods options={paymentMethods} />
        <DonationCheckbox
          onChange={updateAgreeToDonate}
          checked={agreeToDonate}
          content={formatCheckboxLabel(agreeToDonate, tip)}
        />
        <button>${total}</button>
      </div>
    );
  };

And at this point, our code structure starts to resemble something like the diagram below. Note how different parts focus on their own tasks and come together to make the process work.

Figure 9: Refactored Payment with donation

More changes about round-up logic

The round-up looks good so far, and as the business expands to other countries, it comes with new requirements. The same logic doesn’t work in Japan market as 0.1 Yen is too small as a donation, and it needs to round up to the nearest hundred for the Japanese currency. And for Denmark, it needs to round up to the nearest tens.

It sounds like an easy fix. All I need is a countryCode passed into the Payment component, right?

<Payment amount={3312} countryCode="JP" />;

And because all of the logic is now defined in the useRoundUp hook, I can also pass the countryCode through to the hook.

const useRoundUp = (amount: number, countryCode: string) => {
  //...

  const { total, tip } = useMemo(
    () => ({
      total: agreeToDonate
        ? countryCode === "JP"
          ? Math.floor(amount / 100 + 1) * 100
          : Math.floor(amount + 1)
        : amount,
      //...
    }),
    [amount, agreeToDonate, countryCode]
  );
  //...
};

You will notice that the if-else can go on and on as a new countryCode is added in the useEffect block. And for the getTipMessage, we need the same if-else checks as a different country may use other currency sign (instead of a dollar sign by default):

const formatCheckboxLabel = (
  agreeToDonate: boolean,
  tip: number,
  countryCode: string
) => {
  const currencySign = countryCode === "JP" ? "¥" : "$";

  return agreeToDonate
    ? "Thanks for your donation."
    : `I would like to donate ${currencySign}${tip} to charity.`;
};

One last thing we also need to change is the currency sign on the button:

<button>
  {countryCode === "JP" ? "¥" : "$"}
  {total}
</button>;

The shotgun surgery problem

This scenario is the famous “shotgun surgery” smell we see in many places (not particularly in React applications). This essentially says that we'll have to touch several modules whenever we need to modify the code for either a bug fixing or adding a new feature. And indeed, it’s easier to make mistakes with this many changes, especially when your tests are insufficient.

Figure 10: The shotgun surgery smell

As illustrated above, the coloured lines indicate branches of country code checks that cross many files. In views, we’ll need to do separate things for different country code, while in hooks, we’ll need similar branches. And whenever we need to add a new country code, we’ll have to touch all these parts.

For example, if we consider Denmark as a new country the business is expanding to, we’ll end up with code in many places like:

const currencySignMap = {
  JP: "¥",
  DK: "Kr.",
  AU: "$",
};

const getCurrencySign = (countryCode: CountryCode) =>
  currencySignMap[countryCode];

One possible solution for the problem of having branches scattered in different places is to use polymorphism to replace these switch cases or table look-up logic. We can use Extract Class on those properties and then Replace Conditional with Polymorphism.

Polymorphism to the rescue

The first thing we can do is examine all the variations to see what need to be extracted into a class. For example, different countries have different currency signs, so getCurrencySign can be extracted into a public interface. Also ,countries might have different round-up algorithms, thus getRoundUpAmount and getTip can go to the interface.

export interface PaymentStrategy {
  getRoundUpAmount(amount: number): number;

  getTip(amount: number): number;
}

A concrete implementation of the strategy interface would be like following the code snippet: PaymentStrategyAU.

export class PaymentStrategyAU implements PaymentStrategy {
  get currencySign(): string {
    return "$";
  }

  getRoundUpAmount(amount: number): number {
    return Math.floor(amount + 1);
  }

  getTip(amount: number): number {
    return parseFloat((this.getRoundUpAmount(amount) - amount).toPrecision(10));
  }
}

Note here the interface and classes have nothing to do with the UI directly. This logic can be shared in other places in the application or even moved to backend services (if the backend is written in Node, for example).

We could have subclasses for each country, and each has the country specific round-up logic. However, as function is first-class citizen in JavaScript, we can pass in the round-up algorithm into the strategy implementation to make the code less overhead without subclasses. And becaues we have only one implementation of the interface, we can use Inline Class to reduce the single-implementation-interface.

src/models/CountryPayment.ts…

  export class CountryPayment {
    private readonly _currencySign: string;
    private readonly algorithm: RoundUpStrategy;
  
    public constructor(currencySign: string, roundUpAlgorithm: RoundUpStrategy) {
      this._currencySign = currencySign;
      this.algorithm = roundUpAlgorithm;
    }
  
    get currencySign(): string {
      return this._currencySign;
    }
  
    getRoundUpAmount(amount: number): number {
      return this.algorithm(amount);
    }
  
    getTip(amount: number): number {
      return calculateTipFor(this.getRoundUpAmount.bind(this))(amount);
    }
  }

As illustrated below, instead of depend on scattered logic in components and hooks, they now only rely on a single class PaymentStrategy. And at runtime, we can easily substitute one instance of PaymentStrategy for another (the red, green and blue square indicates different instances of PaymentStrategy class).

Figure 11: Extract class to encapsulate logic

And the useRoundUp hook, the code could be simplified as:

src/hooks/useRoundUp.ts…

  export const useRoundUp = (amount: number, strategy: PaymentStrategy) => {
    const [agreeToDonate, setAgreeToDonate] = useState<boolean>(false);
  
    const { total, tip } = useMemo(
      () => ({
        total: agreeToDonate ? strategy.getRoundUpAmount(amount) : amount,
        tip: strategy.getTip(amount),
      }),
      [agreeToDonate, amount, strategy]
    );
  
    const updateAgreeToDonate = () => {
      setAgreeToDonate((agreeToDonate) => !agreeToDonate);
    };
  
    return {
      total,
      tip,
      agreeToDonate,
      updateAgreeToDonate,
    };
  };

In the Payment component, we pass the strategy from props through to the hook:

src/components/Payment.tsx…

  export const Payment = ({
    amount,
    strategy = new PaymentStrategy("$", roundUpToNearestInteger),
  }: {
    amount: number;
    strategy?: PaymentStrategy;
  }) => {
    const { paymentMethods } = usePaymentMethods();
  
    const { total, tip, agreeToDonate, updateAgreeToDonate } = useRoundUp(
      amount,
      strategy
    );
  
    return (
      <div>
        <h3>Payment</h3>
        <PaymentMethods options={paymentMethods} />
        <DonationCheckbox
          onChange={updateAgreeToDonate}
          checked={agreeToDonate}
          content={formatCheckboxLabel(agreeToDonate, tip, strategy)}
        />
        <button>{formatButtonLabel(strategy, total)}</button>
      </div>
    );
  };

And I then did a bit clean up to extract a few helper functions for generating the labels:

src/utils.ts…

  export const formatCheckboxLabel = (
    agreeToDonate: boolean,
    tip: number,
    strategy: CountryPayment
  ) => {
    return agreeToDonate
      ? "Thanks for your donation."
      : `I would like to donate ${strategy.currencySign}${tip} to charity.`;
  };

I hope you have noticed that we’re trying to directly extract non-view code into separate places or abstract new mechanisms to reform it to be more modular.

You can think of it this way: the React view is only one of the consumers of your non-view code. For example, if you would build a new interface - maybe with Vue or even a command line tool - how much code can you reuse with your current implementation?

Push the design a bit further: extract a network client

If I keep this “Separation of Concerns” mindset (for spliting view and non-view logic, or more broadly split different responsibility into its own funciton/class/object), the next step is to do something to relieve the mixing in usePaymentMethods hook.

At the moment, that hook doesn’t have much code. If I add things like error handling and retries, it can easily bloat. Also hooks are a React concept, and you cannot reuse it directly in your next fancy Vue view, right?

src/hooks/usePaymentMethods.ts…

  export const usePaymentMethods = () => {
    const [paymentMethods, setPaymentMethods] = useState<PaymentMethod[]>(
      []
    );
  
    useEffect(() => {
      const fetchPaymentMethods = async () => {
        const url = "https://online-ordering.com/api/payment-methods";
  
        const response = await fetch(url);
        const methods: RemotePaymentMethod[] = await response.json();
  
        setPaymentMethods(convertPaymentMethods(methods));
      };
  
      fetchPaymentMethods();
    }, []);
  
    return {
      paymentMethods,
    };
  };

I have extracted convertPaymentMethods here as a global function. I'd like to move the fetching logic into a separate function so I can use library like React Query to handle all the network-related headaches for me.

src/hooks/usePaymentMethods.ts…

  const fetchPaymentMethods = async () => {
    const response = await fetch("https://5a2f495fa871f00012678d70.mockapi.io/api/payment-methods?countryCode=AU");
    const methods: RemotePaymentMethod[] = await response.json();
  
    return convertPaymentMethods(methods)
  }

This small class does two things, fetch and convert. It acts like an Anti-Corruption Layer (or a gateway 1) that can ensure our change to the PaymentMethod structure is limited to a single file. The benefit of this split is that, again, the class can be used whenever needed, even in the backend service, just like the Strategy objects we saw above.

And for the usePaymentMethods hook, the code is pretty simple now:

src/hooks/usePaymentMethods.ts…

  export const usePaymentMethods = () => {
    const [paymentMethods, setPaymentMethods] = useState<PaymentMethod[]>(
      []
    );
  
    useEffect(() => {
      fetchPaymentMethods().then(methods => setPaymentMethods(methods))
    }, []);
  
    return {
      paymentMethods,
    };
  };

And our class diagram is changed into something like the one below. We have most of the code moved into non-view related files that can be used in other places.

Figure 12: More granular split makes the responsibility of each part cleaner