Throwing a null should raise TypeError, not NullThrownError

[chloestefantsova]

In the following text it is expected that throwing the null value results in NullThrownError. In NullSafe mode an attempt to throw null should raise TypeError instead.

• co19/Language/Expressions/Throw/evaluation_t04
• co19/LibTest/async/Future/Future.sync_A01_t03
• co19/Utils/tests/Expect/throws_A01_t03

cc @lrhn

[lrhn]

The null safety specification says:

It is an error if the static type of e in the expression throw e is not assignable to Object.

It doesn't explicitly say what happens for null typed as dynamic, but I believe it should be treated the same way as other places where we have a required type:

• The context type schema for e is Object.
• It's a run-time type error if the value of e is not an instance of a subtype of Object (aka. "implicit downcast from dynamic").

@leafpetersen Do you agree with this?

Example showing that we have bool as context type in conditions, but not Object in throw:

void main() async {
  if (ct(false)) print("ok"); // "CT: bool"
  throw ct("a"); // "CT: dynamic"
}

// Captures static context type.
T ct<T>(Object? value) {
  print("CT: $T");
  return value as T;
}

[leafpetersen]

The preference would be to use Object as the context type, as discussed in the original issue here. It looks like we neither specified it as such, nor implemented it as such though. Technically breaking to change it now.