Type propagation/inference not symmetric for ?:

[munificent]

This program generates an error on line 2, but not line 1:

class A {}
class B extends A {
  b() {}
}

main() {
  A a = new B();
  a is B ? a.b() : null; // 1
  a is! B ? null : a.b(); // 2
}

I would expect them to both be valid.

[lrhn]

You say "error", but it should only be a warning on line 2, and it's required by the specification.

The type promotion is only defined for positive tests by the current specification. That means that the static type of a is A in all of line 2.

[munificent]

You say "error", but it should only be a warning on line 2, and it's required by the specification.

In strong mode, where I noticed this, type warnings are promoted to errors. But, yes, this is reported as a warning when strong mode is not enabled.

The type promotion is only defined for positive tests by the current specification.

Should I reassign to area-language?

[lrhn]

By all means request a better type promotion specification. I can't find an open feature request for it, but there might be one already that I couldn't find.

Or request that the analyzer can be allowed to not show warnings that it knows are spurious.

[leafpetersen]

@munificent This is covered by the proposal that you and @bwilkerson are working, correct?

[munificent]

Yes. Duplicate of #29624.