Notes on Thoughtbot’s Stop Using any’”

A couple tweaks and improvements to a good post!

Assumed Audience: Developers working with (or just curious about) TypeScript.

Scrolling through my RSS feed this evening, I came across this excellent post on the Thoughtbot blog. I strongly agree with nearly everything in it: when we land TypeScript more broadly at LinkedIn, we will probably lint against ever using any.

However, a couple things caught my attention as points for improvement in the code samples shared there, so I figured I’d share them more broadly so everyone can benefit from them.

Under the heading I really don’t know what it is, the authors offer this example of using unknown instead of any:

type ParsedType = {
  id: number
}

const parseApiResponse(
  response: Record<string, unknown>
): ParsedType => {
  const convertedResponse = (response as ParsedType)

  // without doing the type cast we would
  // get a type error here
  if (convertedResponse.id >= 0) {
    return convertedResponse
  } else {
    throw Error.new("Invalid response"
  }
}

This definitely is preferable to using any… but what if we could eliminate the (also totally unsafe!) cast as well? Turns out we can! TypeScript’s notion of type narrowing gives us the ability to check for this safely:

type ParsedType = {
  id: number
}

const parseApiResponse = (
  response: Record<string, unknown>
): ParsedType => {
  if (typeof response.id === 'number' && response.id >= 0) {
    let parsed = { id: response.id };
    return parsed;
  } else {
    throw new Error("Invalid response")
  }
}

If the response were more detailed than this, and therefore required more validation, we could extend the checks in that if statement to cover more ground. What’s more, if we want to avoid the extra object allocation, we can do that by reaching for some utility types which let us be as robust as we like. And we could go further in that case and parse even a response typed as unknown:

type ParsedType = {
  id: number
}

function isValid(response: unknown): response is ParsedType {
  return (
    isObject(response) &&
    has('id', response) &&
    typeof response.id === 'number' &&
    response.id >= 0
  );
}

const parseApiResponse = (response: unknown): ParsedType => {
  if (isValid(response)) {
    return response;
  } else {
    throw new Error("Invalid response")
  }
}

Now we have no type-casts, because our runtime code — which is required if this is actually going to be safe in either JavaScript or TypeScript! — proves to the compiler that this is actually a ParsedType. This means that we can use parseApiResponse with a call like fetch and its data.json(), which returns a Promise<any>.1

Second, in their section I have to write a lot of code when I add types, any is less work, they offer this code sample as justification of the (entirely correct, in my view!) claim that if we are writing code without types, we will likely add defensive code to make sure arguments and variables have the correct shape for the program to perform as intended:”

const fullName = (user: any) => {
  if (user?.firstName && user?.lastName) {
    return `${user.lastName}, ${user.firstName}`
  }

  return user?.firstName || ""
}

As a replacement, they suggest this:

interface User {
  firstName: string
  lastName?: string
}

const fullName = ({ firstName, lastName }: User) => {
  if (lastName === undefined) {
    return firstName
  }

  return `${lastName}, ${firstName}`;
}

This is a great improvement!2 However, I feel obliged to note that instead of the or || operator in their any example, it’s worth reaching for the nullish coalescing ?? operator, which treats 0, "", etc. differently than undefined and null! Now, the second example is still better code, and we don’t even need it in that case, but nullish coalescing is a great tool to have in your toolbox. You can imagine that in the case where the User argument here was optional — perhaps as part of some other set of arguments, or because you were dealing with untrusted data, that it would still be useful:

interface User {
  firstName: string
  lastName?: string
}

const fullName = (user?: User) => {
  const first = user?.firstName ?? "";
  const last = user?.lastName ?? "";
  return `${first}, ${last}`;
}

This is contrived, to be sure, but it shows how useful optional chaining and nullish coalescing can be even when you do have well-typed code.

In sum: that post from Thoughtbot had great recommendations, but with a couple tweaks we can make it even stronger!


  1. Shameless plug: I would not throw an error here. Instead, I’d reach for a Result type, like the one in the library a friend and I wrote a couple years ago. That way I would be able to have type-safe error handling, as well! That would end up looking something like this:

    import { Result } from "true-myth";
    
    class ApiError extends Error {
      constructor(readonly response: unknown) {
        super("Invalid response");
      }
      
      static from(response: unknown): ApiError {
        return new ApiError(response);
      }
    }
    
    const parseApiResponse =
      (response: unknown): Result<ParsedType, ApiError> =>
        isValid(response)
          ? Result.ok(response)
          : Result.err(ApiError.from(response));
    

    Now we have a well-typed error, which we can deal with as a value — no need for another try/catch block, and in fact we know some details about the kind of error we have! ↩︎

  2. Now, I would absolutely write this as a ternary and a single-expression function body instead:

    interface User {
      firstName: string
      lastName?: string
    }
    
    const fullName = ({ firstName, lastName }: User) =>
      lastName ? `${lastName}, ${firstName}` : firstName;
    

    …but for our purposes in this post that doesn’t much matter. 😅 ↩︎