Union types do not understand function overloads and cause an error

[maludwig]

Bug Report

🔎 Search Terms

• Overloading
• TS2769: No overload matches this call.

🕗 Version & Regression Information

• This is the behavior in every version I tried, and I reviewed the FAQ for entries about overloads not matching this

⏯ Playground Link

Playground link with relevant code

💻 Code

export function stringChecker(a: string, b: string): boolean
export function stringChecker(a: string, b: RegExp): boolean

export function stringChecker(a: string, b: string | RegExp): boolean {
  if (typeof b === 'string') {
    return a === b
  } else {
    // If b instanceof RegExp
    return b.test(a)
  }
}

const tests: { [key: string]: string | RegExp } = {
  is_it_an_a: 'a',
  is_it_a_b: 'b',
  is_it_a_letter: /[a-z]/,
}

for (const test_name in tests) {
  const test_value = tests[test_name]
  if (stringChecker('b', test_value)) {
    console.log(`'b' passes the ${test_name} test!`)
  } else {
    console.log(`'b' fails the ${test_name} test!`)
  }
}

🙁 Actual behavior

For clarity, I'm not sure if this should be classified as a bug or a feature, but it feels like a bug to me, so I made a bug ticket. I'm sorry if that's wrong.

I get an error here:

TS2769: No overload matches this call.   
Overload 1 of 2, '(a: string, b: string): boolean', gave the following error.     
Argument of type 'string | RegExp' is not assignable to parameter of type 'string'.       Type 'RegExp' is not assignable to type 'string'.   
Overload 2 of 2, '(a: string, b: RegExp): boolean', gave the following error.    
Argument of type 'string | RegExp' is not assignable to parameter of type 'RegExp'.       Type 'string' is not assignable to type 'RegExp'.

The function overloads allow for a string or a RegExp, but passing in a variable that is of type string | RegExp does not cause TypeScript to understand that this function has an overload for both cases. TypeScript should ask, "if it is a string, would this all work?" and then ask "if it is a RegExp, would this all work?"

This is obviously a trivial example, since I could omit the overloads and it would all work just fine, but it is frustrating for a library like supertest which a non-trival case of wanting to list all of the available overloads:

From: https://github.com/DefinitelyTyped/DefinitelyTyped/blob/master/types/supertest/index.d.ts

    interface Test extends superagent.SuperAgentRequest {
        app?: any;
        url: string;
        serverAddress(app: any, path: string): string;
        expect(status: number, callback?: CallbackHandler): this;
        expect(status: number, body: any, callback?: CallbackHandler): this;
        expect(checker: (res: Response) => any, callback?: CallbackHandler): this;
        expect(body: string, callback?: CallbackHandler): this;
        expect(body: RegExp, callback?: CallbackHandler): this;
        expect(body: Object, callback?: CallbackHandler): this;
        expect(field: string, val: string, callback?: CallbackHandler): this;
        expect(field: string, val: RegExp, callback?: CallbackHandler): this;
        end(callback?: CallbackHandler): this;
    }

I could avoid this bug in my code with this pattern, but it looks extremely silly to have the same code in both branches of a conditional:

for (const [test_name, test_value] of Object.entries(tests)) {
  if (test_value instanceof RegExp) {
    if (stringChecker('b', test_value)) {
      console.log(`'b' passes the ${test_name} test!`)
    } else {
      console.log(`'b' fails the ${test_name} test!`)
    }
  } else {
    if (stringChecker('b', test_value)) {
      console.log(`'b' passes the ${test_name} test!`)
    } else {
      console.log(`'b' fails the ${test_name} test!`)
    }
  }
}

🙂 Expected behavior

The code should just compile without complaints. Whoever makes the supertest npm package shouldn't need to manually specify every permutation of overload, in addition to specifying a particular overload. TypeScript should just accept that the code is fine.

[jcalz]

Duplicate of #14107

[maludwig]

Should I close this one? It is defs a duplicate.