wasmparser(CM+GC): Assert that SubtypeCxs are always using the same typing context
#2170
Conversation
… typing context A `SubtypeCx`, used for subtype-checking in the component model, is constructed with two `TypesRef`s. These `TypesRef`s must be associated with the same validator, so that we know that core types are canonicalized the same across the two of them and we can compare `CoreTypeId`s for equality. I had originally hoped to move the `SubtypeCx` to wrapping a single `TypesRef`, but this cannot work because (unlike core types) new component model types can be pushed into each different `TypesRef` for different components.
| pub fn new_with_refs(a: TypesRef<'a>, b: TypesRef<'a>) -> SubtypeCx<'a> { | ||
| assert_eq!(a.id(), b.id()); | ||
| Self::new(a.list, b.list) | ||
| } |
There was a problem hiding this comment.
Could this go further and take a single TypesRef as a constructor argument?
There was a problem hiding this comment.
I tried that, but it doesn't work because while the TypesRefs are always associated with the same validator, they can still be distinct and have different sets of component types. See my last paragraph in the OP:
I had originally hoped to move the
SubtypeCxto wrapping a singleTypesRef, but this cannot work because (unlike core types) new component model types can be pushed into each differentTypesReffor different components.
There was a problem hiding this comment.
Wait hang on, while TypesRef can be different aren't theTypeList values the same? Otherwise how are core type canonicalizations shared?
There was a problem hiding this comment.
The underlying TypeLists are not necessarily the same, but one will always be the extension of another, so any types they both share are canonicalized the same.
There was a problem hiding this comment.
Ah ok right, those are snapshots of a list at a point in time, ok makes sense!
A
SubtypeCx, used for subtype-checking in the component model, is constructed with twoTypesRefs. TheseTypesRefs must be associated with the same validator, so that we know that core types are canonicalized the same across the two of them and we can (for example) compareCoreTypeIds for equality.I had originally hoped to move the
SubtypeCxto wrapping a singleTypesRef, but this cannot work because (unlike core types) new component model types can be pushed into each differentTypesReffor different components.