Reduce away intersections that aren't assignable to constituents in CFA

[Anders Hejlsberg (ahejlsberg)]

Fixes #37659.

[Wesley Wigham (weswigham)]

Anders Hejlsberg (@ahejlsberg) shouldn't A & B be assignable to A in this case? A & B by definition has all of A's private fields, no?

[Anders Hejlsberg (ahejlsberg)]

Wesley Wigham (@weswigham) Well, yes, but A and B also have a conflicting private member x and we consider it an error to access conflicting members (introduced years ago in #13990). The access happens as we're checking the properties from the whole intersection against the properties of each of A and B, a check that was introduced recently in #37195.

[Anders Hejlsberg (ahejlsberg)]

Wesley Wigham (@weswigham) We're basically trying to have it both ways. It would seem reasonable for A & B to be assignable to A and B even with the conflicting private properties, yet we rely on an error being issued in the link below as an early warning that a mixin produced an invalid type.

https://github.com/microsoft/TypeScript/pull/13990/files#diff-4b0cabe62740b5899b9a7850e502007aR95

[Anders Hejlsberg (ahejlsberg)]

Superceded by #37762.