wrong type inferred in reduce statement

[ivanhofer]

Bug Report

🕗 Version & Regression Information

Tried in 4.4.4 and Nightly version

⏯ Playground Link

https://www.typescriptlang.org/play?#code/KYDwDg9gTgLgBAYwgOwM7wJaoEIQgG2AENk4BeOACgDcj8BXYALjnuQGtkIB3ZASha0GwOFjgAjPIRLkAfHBgBPMMAgAzOEMbkyFAOSSCxZHoDcAKHNI08eqmABBACZEwMYFHJVUi5AgAiwCrITsB+GMCoLOhQGMgA5gDaALoANHBEPn6BwaHhkdEwsQkpcAA+ElLGfHJUoblhCBFRcDFxSckCcImG0sjpvcbJtQDe5nATiCjorPYOWQheWLhGJJSZvgFBYXlNkXzjk0cA-BkLOTuNzYdHEywb2dshV5EAdFDATvQIwJSUYB9qOl6jUyPIAcBqOUKvVLvlUK84ggGKFUJQQek1HR7AcblMbLMRBQ7I4FtDWpsLs94e9Pt9fv9AcCgqDwYDybDqXsEUiUZF0SzMdjgLijh8YPQoKREiT0iT5ptkuYAL6WKzTWxzFxuDwAJi8lHOT12zUKxQ66QeWwa8LN7VKFUGJFZdWNLxabRKnRYPSqJAGfuQwzBcDGR2sM3lZIoy0D6yNNu5B1uk1OVqpJsieKO9wTcO5tK+Pz+EKBcBBtVLHLdNN59FRArAfCF+BxlnDGsJXjzXOahfpJaZ5ZZlfZZRhNYLdYbGLgWNbIvJUc27cm4sl0tlhIVfiVqqAA

💻 Code

export const isBoolean = (value: unknown): value is boolean => typeof value === 'boolean';

const useAdapter = (syncDependencies: string[], asyncDependencies: string[] | boolean) => (dependencies: string[]): [boolean, boolean] => {
    const useAsync = isBoolean(asyncDependencies)
        ? asyncDependencies
        : asyncDependencies.reduce((prev, dep) => prev || dependencies.includes(dep), false)

    const use = useAsync || syncDependencies.reduce((prev, dep) => prev || dependencies.includes(dep), false)

    return [use, useAsync]
}

🙁 Actual behavior

TypeScript thinks that syncDependencies.reduce() returns a string. If i remove useAsync || in front of the reduce statement or place it after the reduce statement || useAsync it is detected as a boolean

🙂 Expected behavior

The reduce statement should return a value of type boolean because the value actually is a boolean

[ahejlsberg]

This is a known design limitation in our type inference algorithm. See here. Basically the issue in the failing repros is that the false argument is contextually typed by boolean, which causes us to preserve the literal type false. We then use that type as the inferred type for prev, which locks the inference, at which point we can't infer the wider type boolean returned by the arrow function. It's unfortunate, but at this point we don't have a better solution. The workaround is to write false as boolean instead just false.

[andrewbranch]

Duplicate of #46438, see also #36554

[ivanhofer]

Thanks @ahejlsberg for the good explanation and the workaround!

[jcalz]

Fixed by #48380, it seems