Regression to #35004

[valler]

🔎 Search Terms

Assertion functions don't work with wildcard imports

🕗 Version & Regression Information

This changed between versions 6 or earlier and 7

⏯ Playground Link

No response

💻 Code

// foo.ts
export function f(x: any): asserts x is boolean {
  if (typeof x === "boolean") return;
  throw new Error("...");
}

// bar.ts
import * as m from "./foo.js";
const { f } = m;
f(!0);

🙁 Actual behavior

TS2775

🙂 Expected behavior

no error

Additional information about the issue

See #35004

[MartinJohns]

Your issue is the deconstruction, not the wildcard import.

[valler]

Thx.
And is it then expected to get TS2775?
Should I close or rather rename the issue?

[jcalz]

As mentioned above this isn't about #35004 (if you just call m.f(!0) it will work).

I would definitely expect TS2775, although it's painful. From #32695, a function call cannot be an assertion unless

each identifier in the function name references an entity with an explicit type

meaning the function's type needs to be annotated. So you can write

const f: ((x: any) => asserts x is boolean) = m.f

and it will work because f has an explicit type annotation. But it's not even possible to explicitly annotate the type of a variable declared via destructuring assignment so you can't do that. Something like #29526 would be necessary (but maybe not sufficient) to get this working. If you want to rename this, you'd probably say something like "allow destructuring assignment of assertion functions" and frame it as a feature request.

[RyanCavanaugh]

Smaller repro would be

function f(x: any): asserts x is boolean {
  if (typeof x === "boolean") return;
  throw new Error("...");
}
const m = { f };

const { f: g } = m;
g(!0);

TS2775 means exactly what it says. You can log a suggestion for adding another special case on this but this isn't a bug and I don't think it's a regression either.