Typing API Responses With Zod

Have you ever needed to synchronize types in your frontend app with the backend API?

If you ever had an API action defined like that in your controller:

public record UserViewModel(Guid Id, string Name, string LastName, string Login,
bool IsActive, int LoyaltyPoints, AddressViewModel? Address = null);
//
public List<UserViewModel> AllUsers()
{
var usersViewModels = TestDataGenerator.GetTestUsers();
return usersViewModels.OrderBy(uvm => uvm.Name).ToList();
}

and fetched this data using TypeScript in the following way:

getAllUsers = async (): Promise<UserViewModel[]> => {
const url = `${this.apiEndpoint}/AllUsers`;
const response = await fetch(url);
const users = (await response.json()) as UserViewModel[];
return users;
};

at some point, you probably also experienced the desynchronization of backend (C#, in our example) and frontend (TypeScript) types definitions. What if someone has changed the C# version of UserViewModel, but no one corrected its TypeScript’s equivalent?

Your TypeScript fetching code will tell nothing about that. There will be no error, even though the fetched data doesn’t match the expected UserViewModel type.

I’ll try to address this issue in this article 🙂 Let’s see how typing API responses with zod can help us here.

Synchronization of Backend and Frontend API Typings

First, why would we want to keep the backend and frontend models in sync?

For me, that’s the purpose of using TypeScript. We want our code to be as well-typed as possible. For instance, we normally want the data displayed to the user to be fully typed. TypeScript enhances our programming experience by providing us with typing information. Thanks to that, we know what is what and what contains what. We also express what types of data we expect in particular cases.

The APIs mostly return JSON data, which can be anything. Because of that, it is much easier to have the data returned from the API fully typed in TypeScript. Thanks to that, we know what properties are available on the data models received from the API and whether we can use and display them to the users.

The sample code used within this article is available on GitHub. We will use ASP.NET Core (C#) and React (TypeScript) apps as examples.

Models synchronization example

As we saw in the beginning, a classic example is an API controller that returns a strongly-typed data:

public List<UserViewModel> AllUsers()
{
var usersViewModels = TestDataGenerator.GetTestUsers();
return usersViewModels.OrderBy(uvm => uvm.Name).ToList();
}

The returned data type is a collection of UserViewModel objects. Here’s the C# definition of this type:

public record UserViewModel(Guid Id, string Name, string LastName,
string Login, bool IsActive, int LoyaltyPoints, AddressViewModel? Address = null);

Its equivalent is also defined on TypeScript side:

export interface UserViewModel {
id: Guid;
name: string;
lastName: string;
login: string;
isActive: boolean;
loyaltyPoints: number;
address: AddressViewModel | null;
}

Usage in TypeScript

Cool. With this simple code, we can create a usersService.ts file and fetch our users’ data from the API. Notice how we make this call strongly typed:

export class UsersService {
private apiEndpoint: string;
constructor() {
this.apiEndpoint = "https://localhost:7131/Users";
}
getAllUsers = async (): Promise<UserViewModel[]> => {
const url = `${this.apiEndpoint}/AllUsers`;
const response = await fetch(url);
const users = (await response.json()) as UserViewModel[];
return users;
};
}
view raw usersService.ts hosted with ❤ by GitHub

Everything looks legit. We can use the data retrieved from the API in UsersList component and everything is nicely typed:

UsersList React component

The data is even perfectly displayed:

Table with users rendered on the web page

So, what can go wrong here? 🤔

The Problem – Typings’ Desynchronization

Let’s say that a backend developer implements a requirement to rename “loyalty points” into “fidelity points”. Easy. (S)he renames LoyaltyPoints property in the C#’s UserViewModel to FidelityPoints.

The new C# model looks as follows:

public record UserViewModel(Guid Id, string Name, string LastName,
string Login, bool IsActive, int FidelityPoints, AddressViewModel? Address = null);

Nice! The backend dev is a very good programmer, so (s)he even launches the React web application to make sure that everything still works correctly and there are no errors in the dev console:

Table of users rendered on the web page with empty "Loyalty points" column and no errors on Chrome DevTools console

After a quick look, everything looks awesome. The users list is displayed, there are no errors in the console. Apparently, these test users don’t have any loyalty points assigned – that’s why the empty values in “Loyalty points” column. What’s more, translators will update the column’s translation later. We are good! Let’s go on prod! 😎

I guess you already know what went wrong here. API definition changed, but TypeScript didn’t inform us about that 😔 Our UserViewModel still uses the old property name:

UserViewModel typescript type definition

However, it still works. When rendering the UsersList, we simply get undefined in place of loyaltyPoints:

UsersList rendering at runtime. user.loyaltyPoints property is always undefined

In the end, this is all JavaScript there. What’s interesting, the renamed fidelityPoints property is already there at runtime:

At runtime we can see the new fidelityPoints property already present

but no one cared about it 😔

With the current solution, we will never be informed soon enough about API models changes in our React application. In the best case, we’ll get an undefiend or null error when clicking through the app. However, it’s usually an end user who finds such problems on production. This is definitely not what we want 😶

We can solve this problem by typing API responses with zod. Let’s now see how to do that.

The Solution – zod

Our remedy – zod – is quite a decent npm package with with ~600k weekly downloads. Its GitHub page advertises the library as TypeScript-first schema validation with static type inference.

You can definitely do many things with zod. It can be used together with libraries like react-hook-form to perform complex forms validation. However, in our case, we’ll treat zod as a solution for better typings in TypeScript.

Adding zod to React app

First, let’s install zod into our React application:

npm i zod

First schema definition with zod

With zod, we define our types in a slightly different way. Instead of creating a type or interface directly, we first create a schema. In our case, we can define a UserViewModelSchema using z.object creator function:

export const UserViewModelSchema = z.object({
id: z.string().uuid(),
name: z.string(),
lastName: z.string(),
login: z.string(),
isActive: z.boolean(),
loyaltyPoints: z.number(),
address: AddressViewModelSchema.nullable(),
});

Few interesting parts here:

  • Line 2: notice how zod helps us to define types like Guid with built-in schemas like uuid()
  • Line 8: first, I used AddressViewModelSchema here. This is a custom schema of an AddressViewModel object, which is another type used internally in UserViewModel. You can use such custom schemas in other schemas. Also notice the nullable() call here, which makes the address property nullable

First step done – we have our UserViewModelSchema. But can we use it instead of UserViewModel type? Not really. Schema is used for validation purposes only. We still need the UserViewModel TypeScript’s type.

Inferring type from zod’s schema

Fortunately, zod comes with a handy z.infer function that allows us to infer the type from the schema.

Finally, the userViewModel.ts file looks as follows:

import { z } from "zod";
import { AddressViewModelSchema } from "./addressViewModel";
export const UserViewModelSchema = z.object({
id: z.string().uuid(),
name: z.string(),
lastName: z.string(),
login: z.string(),
isActive: z.boolean(),
loyaltyPoints: z.number(),
address: AddressViewModelSchema.nullable(),
});
export type UserViewModel = z.infer<typeof UserViewModelSchema>;

We can use the exported UserViewModel type as previously used type. It’s an equivalent, inferred from UserViewModelSchema.

Validating API responses with zod schema

One last step is to make use of UserViewModelSchema. Let’s modify the getAllUsers function from usersService to validate the data received from the API against our schema:

getAllUsers = async (): Promise<UserViewModel[]> => {
const url = `${this.apiEndpoint}/AllUsers`;
const response = await fetch(url);
const usersJson = await response.json();
const users = z.array(UserViewModelSchema).parse(usersJson);
return users;
};

Notice the usage of z.array. This function call tells zod to validate an array of objects meeting the rules defined by UserViewModelSchema, not a single object.

Now, let’s run our React app and see what happens when we click the “Fetch users” button:

zod throws schema validation error on Chrome dev tools console

This is awesome! Exactly what we wanted – a schema validation error for API response. Notice how the error message precisely points to the missing (or wrong, in other cases) property. It tells us we expected a number called loyaltyPoints, but instead we received undefined. The reason for this error message is that the loyaltyPoints field is Required in our schema.

After renaming loyaltyPoints to fidelityPoints in UserViewModelSchema and updating the UsersList component accordingly, everything works fine again.

We are now fully typed and prepared for the future, in case an issue with desynchronization of frontend and backend typings happens again 🚀

Summary

Today, we’ve seen how typing API responses with zod can help us detect frontend and backend models desynchronization. Schema validation throws errors when the data doesn’t match its expected shape.

Remember that zod is an extended library with many options. I recommend exploring them on your own. An interesting feature we didn’t cover in this article is strict mode, which doesn’t allow additional fields not present in the schema definition when validating the data object.

The open question remains whether to use schema validation on production. One could think that it’s better to not throw any validation errors, because JavaScript may just work. However, I think that throwing an error is always better than silently letting things through. An error lets programmers, automated tests or manual testers to detect the issue before the end user does 😉

You can explore the complete code presented in this article here.

.NET/Full Stack Developer and digital nomad
5 1 vote
Article Rating
Subscribe
Notify of
guest
2 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
Kampus Swasta Terbaik
28 days ago

thanks for the information