Analyzer erroneously performs .call tearoff on expressions whose type is a nullable type variable

[stereotype441]

I'm currently trying to write up a specification for type inference within function bodies and initializer expressions, and this includes specifying when implicit .call tearoff should occur. I've discovered a difference in behavior between the analyzer and CFE.

The following code is accepted by the analyzer but rejected by the CFE:

class C {
  void call() {}
}

void Function() test<T extends C>(T? t) => t;

The analyzer accepts the code because, when considering whether to apply implicit .call tearoff to the expression t, it sees that the static type of t is T?, which is a type parameter type (with a ? suffix). It ignores the suffix and replaces the type parameter type with its bound, C. Then it sees that C contains a .call method with an appropriate type, so it inserts an implicit tearoff of .call. This is not sound, however, because t might be null.

The CFE correctly gives an error for this code.

I'll work on a fix.