[5.3] Type narrowing regression with property comparison to boolean false

[meyfa]

🔎 Search Terms

narrowing, strict equals, property access, boolean false, undefined

🕗 Version & Regression Information

• This changed between versions 5.2 and 5.3
• This also happens in the nightly

⏯ Playground Link

https://www.typescriptlang.org/play?ts=5.3.1-rc#code/JYOwLgpgTgZghgYwgAgGIHt3IN4ChnIBGcUAXEZgDYRwi4C+uuC6IAzmMgOYRgZYBeZAAoAlOX7IAPsgCuIACYQYoCAuQCAfDnzIovWVBDIAsnDAALAHRRaC9AFsxybQAYrAVmQB+HERLk8JRsKPTI5PJKKiBqDEws7JwwmBrcvPxiuMAwIsno3lbEUBoCQkEhojoECWzo1FaU6FzCeaJxuAD0HcgAAmBsALQQAB4ADhAIYENQUOjFA8h5yMBsyKPobGzAhJQAnnKKyqoKzKy19Y3NeYUkbUA

💻 Code

interface Foo {
  bar: boolean
}

const getFoo = (): Foo | undefined => {
  return Math.random() > 0.5 ? { bar: false } : undefined
}

const foo = getFoo()
if (foo?.bar === false) {
  console.log(foo)
}

// @ts-expect-error - foo is possibly undefined
console.log(foo.bar)

🙁 Actual behavior

In line 13 and following, foo is incorrectly narrowed to just Foo (whereas it was previously Foo | undefined), although we learned nothing about it from the preceding if as that did not influence control flow.

Unused '@ts-expect-error' directive.(2578)

🙂 Expected behavior

TS should report an error when accessing foo.bar as foo can potentially be undefined.

Additional information about the issue

I originally discovered this issue in JSX with a snippet similar to:

function Component () {
  const foo = getFoo()
  return (
    <div>
      {foo?.bar === false && 'foo'}
      {foo.bar ? 'true' : 'false'}
    </div>
  )
}

There as well, the access foo.bar is allowed due to the preceding check, although they are completely separate.

Note that changing it to === true or removing the checks entirely fixes the problem.

[Andarist]

It looks very much related to my change: #53714 . I'll investigate this soon. Thanks for the issue report!

[akwodkiewicz]

Here's a different reproduction resembling the one we have in our project, maybe you find it helpful:

export interface Options {
    a?: boolean;
    b?: boolean;
}

function foobar(options?: Options) {
    // @ts-expect-error 'options' is possibly 'undefined'
    if (options?.a === false || options.b) {
    }
}

https://www.typescriptlang.org/play?ts=5.3.2#code/KYDwDg9gTgLgBASwHY2FAZgQwMbDgeTBgQiQGc4BvAKDjrkwH4AuOAIwggBthMkBuWvTYt2nHn0EBfatXQBXJNmKk46Tm0xQAFBCIlyowivIBKKkLoB6K3AACMMgFpQYYMpdQo0OAHI9JmS+iBSQZGQIbFwAnn6KACbA6MjA8b6WiOhwuvqkZIwAdJhwALxlaphcZHgAPjVwAQZkBWzmNPRwMlJAA

EDIT: I tested it with the nightly version of TS from your PR and yes, it's fixed there.

[jakebailey]

This issue has a lot of thumbs up; out of curiosity, are people finding this naturally? Or is there some Twitter thread that's directing people here?

The original PR didn't fail any of the extended test suites, so it's somewhat surprising to me to have a regression like this.

[Andarist]

My guess is that it was posted on a company’s Slack channel or smth and a couple of devs from that channel +1ed this 😉