Repository navigation
Generic lookup type wrongly narrowed to never with strictNullChecks #26130
Description
I ran across this while migrating my project to strictNullChecks and it's positively bizarre.
TypeScript Version: master (50f442f)
Search Terms: mapped type narrow narrowing null strictNullChecks never
Code (with strictNullChecks)
let mappedObject: {[K in "foo"]: null | {x: string}} = {foo: {x: "hello"}};
declare function foo<T>(x: T): null | T;
function bar<K extends "foo">(key: K) {
const element = foo(mappedObject[key]);
if (element == null)
return;
const x = element.x;
}Expected behavior: Compile without error.
Actual behavior: Error: Property 'x' does not exist on type 'never'.
Playground Link: link (remember to enable strictNullChecks)
Related Issues: Didn't find any.
Activity
RyanCavanaugh commentedon Aug 1, 2018
This is just weird. Writing
eliminates the error, but there shouldn't be anything for
!to take awayRyanCavanaugh commentedon Aug 1, 2018
Not a regression FWIW
weswigham commentedon Aug 2, 2018
FWIW
!, as a cast, also resets a control flow-narrowed type to its declared type. We have an open issue to make it respect narrowing: #16945 (I was looking at it recently)mattmccutchen commentedon Aug 2, 2018
I found the problem: in
getAssignmentReducedType, we have a declared type of{x: string} | nulland an assigned type of{foo: {x: string} | null}[K] | null. When we ask if the assigned type is maybe assignable to{x: string}, we decompose the first level of union and check if{foo: {x: string} | null}[K]is assignable to{x: string}, but it isn't because its constraint is{x: string} | null. So we throw{x: string}out of the declared type and are left withnull, and after the null check, we are left withnever.Using the comparable relation in place of
typeMaybeAssignableTofixes this issue. I'm going to see what else breaks if I do that.Have getAssignmentReducedType use the comparable relation instead of
mattmccutchen commentedon Aug 2, 2018
Proposed fix is in #26143.
Have getAssignmentReducedType use a new relation that checks for weak
Go back to the old narrowing algorithm (pre microsoft#26143) and avoid …
Merge pull request #26425 from mattmccutchen/issue-26405