5.3.0 regression - as const is being ignored in assignment, leading to false negatives

[MichaelMitchell-at]

🔎 Search Terms

"as const", 5.3.0, assignment

🕗 Version & Regression Information

• This changed between versions 5.2.2 and 5.3.0 (nightly)

⏯ Playground Link

https://www.typescriptlang.org/play?ts=5.3.0-dev.20231016#code/PTAEAEBcGcFoFMAeAHeBjSCBOWD2XQAVACwEtpRpjcBXAGwBNQAjeUAQ1EgE9VR4c+FunY1obAAZZ47BrgB2dbqADaAFgBMAXQmhyoebkgdo0UgHN57ZnTaRcoCQEEc7bgB55NALassAPgkAKFtjRAAuUBcsN08fP39QAF5VTS0TUDQFaEgAbiCgA

💻 Code

// This should be a type error because `readonly [42]` is not assignable to `Array<number>`
let x: Array<number> = [42] as const;

🙁 Actual behavior

No type error

🙂 Expected behavior

Type error

Additional information about the issue

No response

[Andarist]

I'm almost certain that this was changed by #55229 . I think that this is the expected change in a similar vein that this is expected now: #55522

[MichaelMitchell-at]

I'm almost certain that this was changed by #55229 . I think that this is the expected change in a similar vein that this is expected now: #55522

Ah, no biggie. We were just using this behavior for a "type test", but it's easy enough to fix by assigning the argument to an intermediate variable.

[RyanCavanaugh]

@Andarist finding the PR faster than the bisect script I had going 🐇🐢!

While not identical to the issue in the linked PR, the case for this one being an error also seems a bit weak to me. You still get the desired error in the aliased case (good):

const arr = [42] as const;
let x: Array<number> = arr;

whereas erroring that the provably-only reference of a mutable array is an array you annotated with as const doesn't really seem justified. Effectively, you wrote two type annotations, and we picked the one that made the code work - it's not really a problem.

Aaaaaand I just saw you closed this. Cool 🥰

[Andarist]

@Andarist finding the PR faster than the bisect script I had going 🐇🐢!

I have all Anders' PRs on my speed dial 🤣