Visibility checking was originally introduced to reduce the number of path resolution inconsistencies. However, the current implementation is too restrictive, and even my attempt to make it less restrictive is still not good enough.

So, instead I have come to the conclusion that we are better off without visibility checking. This also aligns with our general assumption that programs are valid, for instance our type inference logic does not actually check that the program is well-typed.

DCA shows that we gain an additional 5 % true-positive call edges (up 385,386 from 367,484).