Issue 14211 - Compiler should devirtualize calls to members of final class

[9rnsr]

https://issues.dlang.org/show_bug.cgi?id=14211

Add CallExp::directcall to handle devirtualizing of virtual function call in front-end.
It can also handle a direct call with DotTypeExp by the same logic.

[yebblies]

You could test this by corrupting the vtable of a class and checking the call succeeds anyway.

eg

void main()
{
    auto c = new Class();
    *cast(void**)c = null;
    c.func();
}

[9rnsr]

@yebblies Unfortunately the test does not work for non-release build, because a class method asserts invariant and it requires valid this instance reference.

class B
{
    void func()
    {
        // _d_invariant() call will cause Access Violaton
    }
}
final class C : B
{
}
void main()
{
    auto c = new C();
    *cast(void**)c = null;
    c.func();
}

[yebblies]

Urrgh. Why are we calling invariants on classes that don't have invariants?

[9rnsr]

Class B is not final, so its derived class may define invariants.

[yebblies]

Ah, that makes sense. extern(C++) classes don't/can't call invariants, so this should work.

extern(C++)
class B
{
    void func()
    {
    }
}
extern(C++)
final class C : B
{
}
void main()
{
    auto c = new C();
    *cast(void**)c = null;
    c.func();
}

[9rnsr]

Good idea! Added test.

[yebblies]

I'm not 100% sure this should be in the frontend, but it doesn't look like it will do any harm.

[yebblies]

Auto-merge toggled on

[yebblies]

win32 just went red on the autotester. Any idea how this could have caused it?