inference problem with constraints involving conditionals and generics

[zpdDG4gta8XKpMCd]

the following is seen in [email protected]:

declare function onlyNullablePlease<T extends (null extends T ? any : never)>(value: T): void;

declare var z: string | null;
onlyNullablePlease(z); // works as expected

declare var y: string;
onlyNullablePlease(y); // error as expected

function f<T>(t: T) {
    var x: T | null = Math.random() > 0.5 ? null : t;
    onlyNullablePlease(x); // should work, unexpected error:
    /*
Argument of type 'T | null' is not assignable to parameter of type 'null extends T | null ? any : never'.
  Type 'null' is not assignable to type 'null extends T | null ? any : never'.ts(2345)
    */
}

[jack-williams]

I suspect this PR #29338 introduced the delta, not the one you link. I also think this is related to #29505. Wrapping the check and extends types in a 1-tuple is a short-term fix.

declare function onlyNullablePlease<T extends ([null] extends [T] ? unknown : never)>(value: T): void;

[zpdDG4gta8XKpMCd]

fyi, the problem is a major breaking change in our project, it prevents us from upgrading to the latest version of typescript, the workaround by @jack-williams helped though

[weswigham]

Personal request: Use unknown instead of any for something like the example in the OP - type parameters extending any are currently.... strange (they're kinda checked and kinda not checked). See #29571 where we plan to fix the strangeness.

[zpdDG4gta8XKpMCd]

unknown doesn't help

[weswigham]

Oh, yeah, it doesn't help here, but I just wanted to inform that extends any behaves poorly and might want to be avoided, since extends unknown behaves far better.