Fixes #2059. Don't expect an error in a weak mode

[sgrekhov]

This test used to test a strong mode. Now in a weak mode it fails. As far as I understand here we have class

class C<T> {
  final T x;
  const C.t(Object? o, T t) : x = o as T;
}

When we call

const c1 = C.t(null, cLegacyInt);

T is inferred as int*. int* became int and x = o as int; failed. But now in a weak mode is doesn't throw, so we should expect no error

[eernstg]

LGTM, after a bit of thinking.

[eernstg]

Github won't let me comment on earlier lines, but I think CONST_CANONICAL_TYPE(int*) is int*. (CONST_CANONICAL_TYPE adds the * on almost every type (the only exception I can easily spot is that CONST_CANONICAL_TYPE(T?) is T2? for some T2).

However, I don't think this matters for the source code below.

[eernstg]

OK, tricky as usual. ;-)

The inferred type argument for C needs to be a supertype of Null and a supertype of int* (the type of cLegacyInt), but there are several types satisfying this. So we shouldn't have a compile-time error.

Next, no-sound-null-safety execution will not throw when o as T is tested in a situation where o evaluates to null. So we should indeed expect that there is no run-time error.

Very good, should work!