[cfe] Adjust computation of flatten

[eernstg]

This is the CFE specific issue for #47053, which has the details.

[eernstg]

@johnniwinther, am I right in guessing that you have been working on this?

[johnniwinther]

Dima and I have been working various parts of it. As I understand it we have update our flatten computation, right? Is it a spec change or just an implementation bug?

[eernstg]

We did not change the spec, but it looks like flatten(T & Future<int>) returns T or T & Future<int>, but it should return int.

(I just asked Dmitry about the fact that the type of x is reported as T rather than T & Future<int>, but assigning x to a variable of type Future<int> is allowed even though the type prints as T, so I assume that it must be represented as T & Future<int> internally.)

[johnniwinther]

The runtimes only look at the T part of T & X.

[eernstg]

(I'm always confused when a type named X isn't a type variable, and a type variable isn't named X, Y or something thereabouts. But I do know that T is a type variable in a lot of concrete cases. ;-)

The runtimes shouldn't need to worry about this, because it's concerned with a static error (and even in the case where the program is allowed to run, this type isn't reified).

[chloestefantsova]

The issue is being worked on at https://dart-review.googlesource.com/c/sdk/+/212467.