-
Notifications
You must be signed in to change notification settings - Fork 12.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Narrow type of variable when declared as literal (also tuples) #16896
Comments
Agreed - it would be really nice to make type narrowing work for object literals. Also consider this case:
It would be nice if the above code compiled (with strictNullChecks enabled) because |
The more I think about this the more I think it would be highly desirable. I cannot think of any case that would be negatively impacted by making literals be a more narrow type since you can always widen the type. But I can see a lot of positive changes coming from this since you cannot automatically narrow a wide type. Consider these examples: declare interface Animal { type: 'cat' | 'dog' }
declare function interactWith (animal: Animal): void
const charlie = { type: 'dog' }
// Argument of { type: string } is not assignable to Animal
interactWith(charlie) A fix in this simple case would be to add e.g. type Schema = NullSchema | BooleanSchema | NumberSchema | StringSchema | ArraySchema | ObjectSchema
interface NullSchema { type: 'null' }
interface BooleanSchema { type: 'boolean' }
interface NumberSchema { type: 'number' }
interface StringSchema { type: 'string' }
interface ArraySchema { type: 'array', items: Schema }
interface ObjectSchema { type: 'object', properties: Record<string, Schema> }
declare function validate (schema: Schema, input: unknown): boolean
const schema = {
type: 'object'
properties: {
name: { type: 'string' }
items: {
type: 'array',
items: {
type: 'object',
properties: {
type: { type: 'string' }
}
}
}
}
}
// Argument of ... is not assignable to Schema
validate(schema, null) Here the fix is quite annoying, you need to go in and declare the type of every nested schema by making all strings string literals. const schema = {
type: 'object' as 'object'
properties: {
name: { type: 'string' as 'string' }
items: {
type: 'array' as 'array',
items: {
type: 'object' as 'object',
properties: {
type: { type: 'string' as 'string' }
}
}
}
}
} I would love it if we could move forward on this. Is there something that I could contribute with? Do we need to do some kind of research on how this would affect current projects using TypeScript? |
Surely the workaround for this is to simply add a type annotation? const schema : Schema = {
type: 'object',
properties: {
name: { type: 'string' },
items: {
type: 'array',
items: {
type: 'object',
properties: {
type: { type: 'string' }
}
}
}
}
} |
@Richiban, unfortunately, that workaround will throw away any information on what type of scheme it is, so you would no longer be able to do The logical next step is to declare it as |
ProposalI think that it is important to give full control on type inference. Just a part of the puzzle is missing, because most of the work was done in #29510. So I propose introducing a new keyword This would also give full power to the user over generics (not having to explicitly pass them): const f = <T>(t: T) => t
// default, widening behavior
const test0 = f({a: {b: 'hello'}})
// => {
// a: {
// b: string;
// };
// }
// non-widening, immutable behavior
const test1 = f({a: {b: 'hello'}} as const)
// => {
// readonly a: {
// readonly b: 'hello';
// };
// }
// non-widening, mutable behavior
const test2 = f({a: {b: 'hello'}} as self)
// => {
// a: {
// b: 'hello';
// };
// } This could give full control on type inference, in general. So we could either use
|
For Tuples the way I'm doing it currently is by using
|
TypeScript Version: 2.4.1
Code
The following doesn't compile, because
x
has inferred typestring
. I think it would be helpful if it did, but I still wantx
to have inferred typestring
:Desired behavior:
This would compile.
Actual behavior:
demo.ts(4,6): error TS2345: Argument of type 'string' is not assignable to parameter of type '"foo" | "bar"'.
Suggestion
The type of
x
should be inferred to bestring
, but it should be narrowed to"foo"
via flow-sensitive typing, much like the below program:Note that this program currently works as desired:
x
is still inferred to have typestring
, but the guard immediately narrows it to type"foo"
. The assignment after the function call widensx
's type back tostring
.My suggestion is for TS to treat declarations that assign variables to literals (as in the first example) to automatically perform this flow-sensitive type narrowing, without the need for an unnecessary guard.
I'm suggesting it only for declarations, not general assignments (so only with
let
,var
, andconst
) or other operations. In addition, I'm only suggesting it for assignments to literal values (literal strings or literal numbers) without function calls, operations, or casts.Syntax Changes
Nothing changes in syntax.
Code Generation
Nothing changes in code generation.
Semantic Changes
Additional flow-sensitive type narrowing occurs when variables are declared as literals, essentially making the existing
behave the same as (in checking, but not in codegen)
Reverse Compatibility
Due to
x
being assigned a more-precise type than it was in the past, some code is now marked as unreachable/invalid that previously passed:The error is correct (in that the comparison could never succeed) but it's still possibly undesirable.
I don't know how commonly this occurs in production code for it to be a problem. If it is common, this change could be put behind a flag, or there could be adjustments made to these error cases so that the narrowing doesn't occur (or still narrows, but won't cause complaints from the compiler in these error cases).
If the existing behavior is strongly needed, then the change can be circumvented by using an
as
cast or an indirection through a function, although these are a little awkward:Tuples and Object Literals
It would be helpful if this extended also to literal objects or literal arrays (as tuples).
For example, the following could compile:
with
x
again having inferred typenumber[]
but being flow-typed to[number, number]
. The same basic considerations apply here as above.Similarly, it could be useful for objects to have similar flow-typing, although I'm not sure if this introduces new soundness holes:
See #16276 and #16360 and probably others for related but different approaches to take here.
The text was updated successfully, but these errors were encountered: