Object Literal Assignment Does Not Participate in Type Narrowing

[ericdrobinson]

Bug Report

🔎 Search Terms

object literal definition type narrowing type guard

Possibly related: #16896 (especially this comment).

🕗 Version & Regression Information

This is the behavior in every version I tried, and I reviewed the FAQ for entries about 'object' and 'literal'

⏯ Playground Link

Playground link with relevant code

💻 Code

interface Test
{
    a?: number;
}

let x: Test = { a: 23 };

// Error: Type 'number | undefined' is not assignable to type 'number'. (2322)
const val: number = x.a;

🙁 Actual behavior

The compiler does not recognize that the type of property a should be narrowed to number due to the literal definition. This is always a surprise.

🙂 Expected behavior

I expect the type to be narrowed such that the property a is treated as "narrowed to number" until otherwise modified in the scope that follows the assignment.

This feels entirely natural as it's how most other assignments work. This comment puts it nicely:

Initializations are just treated as assignments. So this is basically const value: number | null; value = 123;, and no one would expect value to be possibly null right after assigning a number to it.

In this case, I do not expect x.a to be missing right after assigning to it an object literal with a defined.

[ericdrobinson]

Frustratingly, this code works around the issue:

interface Test
{
    a?: number;
}

let x: Test = {};
x.a = 23;

// OK.
const val: number = x.a;

But having to break up object literal declarations feels like an unnecessary hoop for something the compiler could presumably detect.

We could also throw in a standard type guard, but that's just extraneous (both at runtime and in the code).

[andrewbranch]

You’re right that the comment you linked to in #16896 is exactly this, but I think #16896 on the whole is not. The deal is that only union types narrow on assignment, which is why let x: number | string = "hello" becomes string right after being assigned. But Test is not a union type, so when you assign a value to it, nothing happens except the initializer is checked for conformance to Test. In your workaround, you have an assignment to x.a, which gets matched as a narrowing reference later. This is just a general limitation of the design of our control flow narrowing system.

[ericdrobinson]

The deal is that only union types narrow on assignment, which is why let x: number | string = "hello" becomes string right after being assigned. But Test is not a union type, so when you assign a value to it, nothing happens except the initializer is checked for conformance to Test.

That is an extremely helpful explanation of what's going on behind the scenes! Thanks! 🍻

In your workaround, you have an assignment to x.a, which gets matched as a narrowing reference later. This is just a general limitation of the design of our control flow narrowing system.

It would be nice for the control flow narrowing system to one day handle this case! 😄

[jtbandes]

I encountered this too: playground link

class C { x?: {a:number} }
class D extends C { x?: {a:number,b:string} }

let c = new C();
// ^?
   c = new D();
// ^?

c.x?.b; // error :(

[ericdrobinson]

@RyanCavanaugh I see that this has been closed as "completed". Does that mean that an upcoming version of the compiler will support this as expected? If so, any details on what version or otherwise what changes fixed this?

[RyanCavanaugh]

The bulk "Close" action I applied to Design Limitation issues doesn't let you pick what kind of "Close" you're doing, so don't read too much into any issue's particular bit on that. I wish GitHub didn't make this distinction without a way to specify it in many UI actions!

The correct state for Design Limitation issues is closed since we don't have any plausible way of fixing them, and "Open" issues are for representing "work left to be done".

[Qrokqt]

This is really frustrating when working with sql query builders. They all seem mostly built on reassignment chaining.

[d07RiV]

What is a suggested alternative syntax? Property assignment outside of initializer works, but if a property is not optional, we have to assign it twice, which doesn't make much sense.