Fixed an issue with apparent mapped type keys #57838
Conversation
typescript-bot
added
the
For Uncommitted Bug
src/compiler/checker.ts
Outdated
| @@ -22486,7 +22486,7 @@ export function createTypeChecker(host: TypeCheckerHost): TypeChecker { | |||
| /*stringsOnly*/ false, | |||
| t => void mappedKeys.push(instantiateType(nameType, appendTypeMapping(targetType.mapper, getTypeParameterFromMappedType(targetType), t))), | |||
| ); | |||
| return getUnionType(mappedKeys); | |||
| return mappedKeys.length ? getUnionType(mappedKeys) : stringNumberSymbolType; | |||
There was a problem hiding this comment.
The problem here is that when modifiersType is unknown (or {}) then keyof modifiersType is never. Since never is assignable to everything, the checker assumed that keyof { [P in keyof T as T[P] extends string ? P: never]: any } is assignable to string.
Then based on that information it managed to simplify the instantiated substitution type of Extract to just keyof { [P in keyof T as T[P] extends string ? P: never]: any } (its newBaseType). Then in turn getNarrowableTypeForReference concluded that it should substitute constraints. Those were substituted with stringNumberSymbolType (for the keyof ...) and that's not assignable to string.
Before #56742 , it didn't mistakenly assume that keyof { [P in keyof T as T[P] extends string ? P: never]: any } is assignable to string so the substitution type was instantiated as (keyof { [P in keyof T as T[P] extends string ? P: never]: any }) & string and that is assignable to a string.
…ype-keys-with-unknown-modifier
|
Hi! I appreciate that reviewers can be very busy but I discovered this PR when bisecting my own issue with a bit of an absurd reproduction that was erroring with a much more opaque error "Type instantiation is excessively deep and possibly infinite." so I was unable to find any workarounds before the process of bisecting. I'd really like to see this PR merged because when I built it locally it makes my code work! I wouldn't want to ask this of anyone else without being willing to do it myself so I did also review it and I can't find anything objectionable, though of course my opinion doesn't weigh as much given I'm not a maintainer. |
…ype-keys-with-unknown-modifier
|
@typescript-bot pack this |
|
Hey @gabritto, I've packed this into an installable tgz. You can install it for testing by referencing it in your and then running There is also a playground for this build and an npm module you can use via |
|
@typescript-bot test it |
|
@gabritto Here are the results of running the user tests with tsc comparing Everything looks good! |
|
Hey @gabritto, the results of running the DT tests are ready. Everything looks the same! |
|
@gabritto Here they are:
tscComparison Report - baseline..pr
System info unknown
Hosts
Scenarios
Developer Information: |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
||
| function h<T>(z: StringKeys2<T>) { | ||
| z = "foo"; // error |
There was a problem hiding this comment.
Maybe add a const f: string = z; here too to show that it will still error?
There was a problem hiding this comment.
Shouldn't this be OK though? Only string keys are extracted by StringKeys2 so z has a string constraint~ and is assignable to string. This is how it works today (TS playground) and this PR doesn't change that
| let sourceMappedKeys: Type; | ||
| if (nameType && isMappedTypeWithKeyofConstraintDeclaration(mappedType)) { | ||
| sourceMappedKeys = getApparentMappedTypeKeys(nameType, mappedType); | ||
| if (sourceMappedKeys.flags & TypeFlags.Never) { |
There was a problem hiding this comment.
I find it a bit weird that:
keyof unknownisundefined.- the fix is looking at the result of
getApparentMappedTypeKeys. I'd expect it to be insidegetApparentMappedTypeKeys. Isn't there a case wheregetApparentMappedTypeKeysisneverandsourceMappedKeysshould stay asnever?
This doesn't look wrong to me, but I can't yet form a good mental model of what's happening, so I don't know what to think. 😅
I'll revisit in a few days and maybe I'll have a better idea of what this all means.
There was a problem hiding this comment.
keyof unknown is undefined.
If we think about keyof T as a set of types that can be safely used to access properties of T in T[keyof T] then never makes sense here. never is the only type that can be safely used as keyof T after we instantiate T as unknown.
the fix is looking at the result of getApparentMappedTypeKeys. I'd expect it to be inside getApparentMappedTypeKeys. Isn't there a case where getApparentMappedTypeKeys is never and sourceMappedKeys should stay as never?
That's what I started with but mapping never to PropertyKey would break the other call site where getApparentMappedTypeKeys is used. In there, it's used to get the target keys. So this would remove the error below:
type StringKeys2<T> = keyof {
[P in keyof T as T[P] extends string ? P : never]: any;
};
function h<T>(z: StringKeys2<T>) {
z = "foo"; // error
}
|
@gabritto Here are the results of running the top 400 repos with tsc comparing Everything looks good! |
Andarist
force-pushed
the
fix/apparent-mapped-type-keys-with-unknown-modifier
branch
from
2f2b500
to
cad532a
Compare
fixes #57827