Question marks (JSDoc types) allowable in instantiation expression type arguments

[jcalz]

Bug Report

🔎 Search Terms

instantiation expressions, question mark, jsdoc nullable, any, Java wildcard

🕗 Version & Regression Information

• This is the behavior in every version since Instantiation expressions #47607 introduced instantiation expressions (TypeScript 4.7 and up)

⏯ Playground Link

Playground link with relevant code

💻 Code

function foo<T>(x: T): T { return x }
const WhatFoo = foo<?> // (x: any) => any 😕
const HuhFoo = foo<string?> // (x: string | null) => string | null 😲
const NopeFoo = foo<?string> // (x: string | null) => string | null 😵
const ComeOnFoo = foo<?string?> //  (x: string | null) => string | null ¿🤪?

class Bar<T> { constructor(public x: T) { } }
const WhatBar = Bar<?>; // Bar<any> 😕
const HuhBar = Bar<string?>; // Bar<string | null> 😲
const NopeBar = Bar<?string> // Bar<string | null> 😵
const ComeOnBar = Bar<?string?> // Bar<string | null> ¿🤪?

🙁 Actual behavior

No compiler errors. The question marks seem to be treated like JSDoc nullable types and a question mark by itself seems to be interpreted as any (not sure if that's true for JSDoc either).

🙂 Expected behavior

Compiler errors on every question mark, saying that it's not valid, perhaps with a "JSDoc types are not allowed here" message.

---

This Stack Overflow question is asking about how to use the Java wildcard syntax in TypeScript, and I'm like 🤯. Apparently this was suggested by VSCode type hints. I don't know much about JSDoc but I assume some support for it has accidentally leaked into instantiation expression checking.

[mike-lischke]

Note that without the question marks my code does not compile.

[jcalz]

Nor should it. You can change ? to any to get it to compile (with all the caveats that come from using any).

[mike-lischke]

any is forbidden in my code base and unknown doesn't compile.

[fatcerberus]

If your codebase forbids use of any (and there are definitely good reasons to do so!) then using an undocumented TS bug to instantiate a ModuleBase<any, any> feels like cheating and probably shouldn't pass code review.

[jcalz]

@mike-lischke That's not really relevant to this issue, unfortunately. By using ? you are essentially using any. I understand that your workflow depends on spacebar heating but this issue is about the (presumable) bug.

[MartinJohns]

any is forbidden in my code base and unknown doesn't compile.

You control the linter. Don't let the linter control you.

[mike-lischke]

If your codebase forbids use of any (and there are definitely good reasons to do so!) then using an undocumented TS bug to instantiate a ModuleBase<any, any> feels like cheating and probably shouldn't pass code review.

I just detected the possibility of using question marks and asked about it on SO, which in turn let to this issue here. Additionally, it solved my problem at hand and it's not yet clear if it is a bug ¯\(ツ)/¯

[fatcerberus]

and it's not yet clear if it is a bug ¯(ツ)/¯

That's fair, though it definitely feels like a bug since ? doesn't work as an alias for any in any other context and TS doesn't have existential types (what Foo<?> would be most likely to denote if that syntax were officially supported).

We'll see what the maintainers say.

[ahejlsberg]

This is definitely a bug. We're forgetting to visit child nodes in checkExpressionWithTypeArguments.