Remove extra indirection from statics with linkage?

[RalfJung]

I just learned from @bjorn3 that the way we represent statics with linkage, such as #[linkage = "extern_weak"], is kind of strange:

unsafe extern "C" {
    #[linkage = "extern_weak"]
    static test_symbol_that_does_not_exist: Option<unsafe extern "C" fn()>;
}

Here, test_symbol_that_does_not_exist will actually be a global variable local to this module that points to the weak symbol.

Now that we have the &raw address-of operator, we could avoid this extra indirection. Whether we should, I do not know.
Cc @bjorn3 @nikic @chorman0773

[chorman0773]

Note that this replacement isn't possible for functions, only statics (since &raw const fn doesn't work). Otherwise, yeah, I agree that the added level of indirection is unneeded (and personally, not something I'd want to use).

[RalfJung]

As @bjorn3 mentioned on Zulip, it's possible, just awkward: you'd do something like

unsafe extern "C" {
    #[linkage = "extern_weak"]
    static weak_fn: ();
}

let fn_ptr: Option<fn(i32) -> i32> = transmute(&raw const weak_fn);
// ...

[bjorn3]

Targets where function pointers are larger than data pointers would be an issue, but the (unstable) std::marker::FnPtr::addr method would truncate on those targets too.

[chorman0773]

That also loses the ability to use the definition safely.