Type inference should exploit "is" and "is!" semantics

[chalin]

Consider this implementation of equals excerpt from the code shown in this section of the Library Tour:

  @override
  bool operator ==(dynamic other) {
    if (other is! Person) return false;
    Person person = other; // (*)
    return ...
  }

As of Dart 2-dev.26, the type of other at (*) is dynamic.
Couldn't it be inferred as Person?

Similar arguments can be given for an if condition that makes use of is.

cc @kwalrath @stereotype441 @leafpetersen @eernstg

[eernstg]

The language specification gives some hints about the rationale for making exceptions about dynamic during type promotion (where we are considering v is T and S is the static type of v):

We do not want to refine the type of a variable of type \DYNAMIC{}, as this could lead to more warnings rather than fewer.
The opposite requirement, that $T \ne \DYNAMIC{}$ is a safeguard lest $S$ ever be $\bot$.

So the main point was, arguably, to let dynamic code be dynamic code.

We probably still want to maintain that no variable can be promoted to dynamic (which is unlikely anyway, because dynamic is now a top type), but support for promoting away from dynamic seems to be in line with the general development of the language from Dart 1 to Dart 2. I just marked this issue with area and type for a language enhancement request.

[leafpetersen]

Promotion from dynamic is part of Dart 2, but promotion in the continuation of a conditional is not. I believe that the extended type promotion proposal that @bwilkerson wrote up handles this (and more), and that proposal had broad support on the language team. While technically breaking, I'm still hoping that we will be able to land this post Dart 2 either as a minimally breaking change, or by using some sort of opt-in mechanism.

[eernstg]

Sorry, I overlooked that the example relies on return and is!. However, there is also #30906 which is about promotion and dynamic in Dart 2.

[lrhn]

Done.