Unable to invoke an arrow function with an assertion predicate

[ssalbdivad]

🔎 Search Terms

arrow function, assertion, predicate, 2775

🕗 Version & Regression Information

• This is the behavior in every version I tried

⏯ Playground Link

https://tsplay.dev/w8QLpN

💻 Code

type A = { x: number };

type B = A & { y: number };

const foo = (a: A): asserts a is B => {
  if ((a as B).y !== 0) throw TypeError();
  return undefined;
};

export const main = () => {
  const a: A = { x: 1 };
  // TypeError: Assertions require every name in the call target to be declared with an explicit type annotation.(2775)
  foo(a);
};

🙁 Actual behavior

It looks like they goal of #45952 was for assertion predicates to work for arrow functions, but this only handles the JSDoc case, which I assume TS treats as equivalent to directly annotating the entire function at once, i.e.:

 const foo: (a: A) => asserts a is B = (a) => {};

For functions with more named parameters, this annotation style quickly becomes unwieldy. There is no semantic difference between this annotation or the style from the repro:

const foo = (a: A): asserts a is B => {};

Is it possible to identify a top-level arrow function like this assigned to a const variable with annotated parameters and return as having an "explicit type," which was defined as the standard for being able to use an assertion predicate like this in @ahejlsberg's initial PR (#32695)?

This style is increasingly common, and I think many developers (my self included) simply write a function like this, see it doesn't work, then choose not to use the feature, not knowing something like moving the annotation to the variable itself could have an effect.

🙂 Expected behavior

(see above)

Additional information about the issue

No response

[fatcerberus]

Unfortunately, it is indeed a design limitation that foo must be explicitly annotated. Annotating the arrow function itself isn't enough - the compiler doesn't know the type of foo until type inference kicks in, which happens after the control flow graph is computed.

[ssalbdivad]

@fatcerberus Do you know if there's a mechanism for this to be special-cased? It seems like it would be relatively simple to detect a top-level arrow function declaration.