False positive: Right operand of ?? is unreachable

[steelbrain]

🔎 Search Terms

Ever since Typescript 5.6, we've been getting false positives of the nullish operand lint rule. Our codebase is littered with it so it blocks us from upgrading to the latest Typescript

🕗 Version & Regression Information

• This changed between versions 5.5 and 5.6 (also there on latest 5.7.2)

⏯ Playground Link

https://www.typescriptlang.org/play/?#code/GYVwdgxgLglg9mABADwDwBUB8AKAzlAJxjAHNcAuRdAUwFsAHAGwEMpqBlQ4sgQQIOYBPADSIAdBIBuzRiGoVEYELQBG1AgG0AugEpK6RAB9FIRo0QBvAFCJbiAtSggCSJWcTNci6pPU27APQBiAB6iIJwIIgQzEgyuHD2dHC+iFAAFtQeKilZjMTUouAOLGwAJmmJGVkqICSiCIyCaZkOlYi4IPT0Dri4-rZBHkjF1KXUFbUkYlYAvlZWQwAicPKKcFCIAO5wBADWlAAy1MCbuDBlWTBeYD7qJmbX6VYQCPiI0rLUlEqq9wC8KFQvzUBEwAAN0jBwYgAPywxAABgWQwA6rs9rhYS83ptPnIAEyIQFoEHqCFQ8E4sDvfHUADMP2UoOJHxkhLhCORQA

💻 Code

function x<T>(strings: TemplateStringsArray, ...values: number[]): T | null {
    return null as never
    // ^ you can also remove the above line, unrelated to the bug, only there to suppress
    // an unrelated error.
}

// Does not work: Left side is never nullish
const value: number = x<number>`hi` ?? 0
// ^ this line causes an error in Typescript

// Works?
const value2 = x<number>`hi`
const value3: number = value2 ?? 0

🙁 Actual behavior

Right operand of ?? is unreachable because the left operand is never nullish

🙂 Expected behavior

No error

Additional information about the issue

I tried looking for duplicate issues but wasn't able to find any. If the bug has already been filed, a link to it would be appreciated. Thanks!

[jakebailey]

This check is syntactic, so it seems like we just need to entirely ignore template expressions like this since we have no idea without type checking that a template could evaluate to a nullable thing :(

[steelbrain]

I'd prefer it being disabled on template expressions to the status quo, which is false positives. Maybe in the future we can infer more information and make it an informed check instead

[steelbrain]

Impressed by the speed of execution here! Good work Team! 👏

[steelbrain]

@jakebailey Is there going to be a 5.7.3 with this change? We're stuck on 5.5 until this fix gets released

[RyanCavanaugh]

In general we wouldn't backport something like this since it's non-crashing, has a workaround, and isn't commonly reported. You can switch to nightly or ts-ignore the affected lines.

[steelbrain]

Neither nightly nor ts-ignore for the hundreds of instances is a good option for us. We'll just wait it out. Thanks for the response!