[CFE] No error in CFE if an extension type is not well-bounded

[sgrekhov]

The following program produces an expected error in analyzer but not in CFE

// SharedOptions=--enable-experiment=inline-class

extension type V<T extends V<T>>(T id) {}
//                         ^
// [analyzer] unspecified
// [cfe] unspecified

main() {
  print(V<Never>);
}

After extension type erasure we'll have here extension type V<T extends T>(T id) {} which is an error. So, analyzer correctly expects COMPILE_TIME_ERROR.TYPE_PARAMETER_SUPERTYPE_OF_ITS_BOUND here

Tested on Dart SDK version: 3.3.0-edge.5918e012cc875df2b96ec25e3fee0de263feed80 (main) (Mon Nov 20 08:03:39 2023 +0000) on "linux_x64"

[chloestefantsova]

Thanks for reporting that, @sgrekhov! As far as I understand, the related passage in the spec is the following:

Let X1 extends B1, .. Xs extends Bs be a declaration of the type parameters of a generic entity (it could be a generic class, extension type, mixin, typedef, or function). Let BBj be the extension type erasure of Bj, for j in 1 .. s. It is a compile-time error if X1 extends BB1, .. Xs extends BBs has any compile-time errors.

Also, the compile-time message emitted by the Analyzer could be more helpful. Currently it says 'T' can't be a supertype of its upper bound., even though T is not a supertype of its bound, V<T>.

[eernstg]

has any compile-time errors.

I don't know if this is obvious, but the transformation that replaces each bound by its extension type erasure could of course create compile-time errors in the body whenever a receiver has a type which is a type variable, and it used to invoke an extension type member (but when the bound is no longer that extension type, the member does not exist, or it has a different member signature). So we're only requiring that it would not be an error of a type parameter to declare the type parameters as bounded by the extension type erasures, we don't care that there could be errors where the type parameter is used.