[Issue 20687] Allow member function address as const initializer

[d-bugmail at puremagic.com]

https://issues.dlang.org/show_bug.cgi?id=20687

Steven Schveighoffer <schveiguy at yahoo.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |schveiguy at yahoo.com

--- Comment #5 from Steven Schveighoffer <schveiguy at yahoo.com> ---
It's very clear to me it should be available at compile time and runtime.
Whether it is callable on its own is a matter for debate, but that would be a
breaking change.

If we made it unavailable at runtime, it is still available with
&instance.fun.funcptr. What does funcptr return? If we started disallowing
funcptr manipulation, then maybe this becomes invalid, but that is a whole new
ball of code breakage that I don't think we want to get into right now.

Even if we disallowed creating a variable that accepts such a member-function
pointer, a delegate has no type information for it's context pointer, so
something like this would still be possible:

struct S
{
  int x;
  void foo() {++x;}
}

struct T
{
  int *x;
  void bar() {*x = 5;}
}

auto s = S(100000);
T t;
auto dg = &s.foo;
dg.funcptr = (&t.bar).funcptr;

dg(); // uses s.x as if it were a pointer

If we wanted to talk about type safety in this area, the status quo does not
cut it. While not type-safe, at least it can be made consistent.

--