Type {Model} apparently doesn't satisfy type {Instance}

[RealLostFormer (Prism)]

Type Model satisfies type Instance, however, when typed as arrays, they are considered to not be similar by the typesolver.

--!strict
local model: Model = Instance.new("Model")
local ins: Instance = model -- No Type Errors

local ModelArray: {Model} = {}
local InstanceArray: {Instance} = ModelArray -- Type '{Model}' could not be converted into '{Instance}' caused by: Property '[indexer value]' is not compatible. Type 'Model' could not be converted into 'Instance' in an invariant context

local typecastTest: {Instance} = ModelArray :: {Instance} -- Cannot cast '{Model}' into '{Instance}' because the types are unrelated

I also tried using custom types to see if they would also reproduce the bug, but no type errors occured.

--!strict

type Vec2 = {x: number, y: number}
type Vec3 = Vec2 & {z: number}

local Vec3Array: {Vec3} = {}
local Vec2Array: {Vec2} = Vec3Array

local v3: Vec3 = {x = 5, y = 2, z = 8}
local v2: Vec2 = v3

Since types Model and Instance are similar, their array types should also be similar, since every value in {Model} is an Instance.

[Judgy_Oreo (DisplayName)]

You intentionally cannot widen types through arrays.

Luau typechecker won't widen an array of a given type to a more generic type

Hi! This is intentional behavior. Widening an array like this is actually unsound, and can lead to type safety bugs. In technical terms, even though BasePart is a subtype of Instance, {BasePart} is not a subtype of {Instance}. Arrays (tables with an indexer, to use Luau terminology) are invariant, meaning they do not have a subtyping relationship with other arrays. This is because if they had subtyping relationships (they were covariant), you could write code like this: local x: {BasePart} = {}…

[RealLostFormer (Prism)]

This doesn’t really explain why Instance arrays can’t be narrowed, but the custom types can.

[Judgy_Oreo (DisplayName)]

Do you have the New Luau Type Solver beta enabled in your Beta Features?

[RealLostFormer (Prism)]

Just checked, I don’t have this feature on.

[Judgy_Oreo (DisplayName)]

I get a warning for local Vec2Array: {Vec2} = Vec3Array with it enabled. The part after that one, where a Vec3 is assigned to a Vec2, is intentional behavior.

You need to be using the New Solver to have all the latest features, such as Type Functions and numberous typechecking improvements. I am under the impression that the team will not be updating the old Solver in order to focus their efforts on the new version.

[RealLostFormer (Prism)]

I don’t use the new solver often because its pretty buggy, so I guess ill just anycast when I can.

[MercuryBromine0]

Hello, thank you for the report! The behavior for Instance and Model is expected, the linked post explains why this is. As for why the table types can be widened under the old solver, that is probably a bug, but we are not prioritizing work on the old solver.

I don’t use the new solver often because its pretty buggy, so I guess ill just anycast when I can.

We’d highly appreciate any bug reports for the new solver!