llvm dev - Dec 2006 - [LLVMdev] weak linkage

I am implementing weak linkage support on the ARM backend and I
noticed this code on the X86 and PPC backends:
------------------------------------------------
// If the initializer is a extern weak symbol, remember to emit the weak
// reference!
if (const GlobalValue *GV = dyn_cast<GlobalValue>(C))
  if (GV->hasExternalWeakLinkage())
    ExtWeakSymbols.insert(Mang->getValueName(GV));
------------------------------------------------

Can a initializer be a weak reference? For example,
------------------------
int a __attribute__ ((weak));
int b = a;
------------------------
fails to compile with the error:
test.c:2: error: initializer element is not constant

Best Regards,
Rafael

On Wed, 6 Dec 2006, [UTF-8] Rafael Esp?ndola wrote:
> I am implementing weak linkage support on the ARM backend and I
> noticed this code on the X86 and PPC backends:
> ------------------------------------------------
> // If the initializer is a extern weak symbol, remember to emit the weak
> // reference!
> if (const GlobalValue *GV = dyn_cast<GlobalValue>(C))
>  if (GV->hasExternalWeakLinkage())
>    ExtWeakSymbols.insert(Mang->getValueName(GV));
> ------------------------------------------------
>
> Can a initializer be a weak reference? For example,
> ------------------------
> int a __attribute__ ((weak));
> int b = a;
Try: int *B = &a;

-Chris

-- 
http://nondot.org/sabre/
http://llvm.org/

> Try: int *B = &a;
In LLVM IL this is
-------------------------------
%b = global int* %a
%a = weak global int 0
-------------------------------

compiling with llc -march=x86-64 produces
-----------------------------
        .globl b
.data
        .align  8
b:                              # b
        .size b, 8
        .quad   a

        .comm   a,4,4           # a
-----------------------------
Note that no ".weak" directive is printed. gcc prints a ".weak
a"
after the ".quad a".

I believe that this is a similar (but more uncommon) problem to one I
am having in ARM when a weak symbol is placed is a global table. I
will send my current hack to the ML in a minute.
> -Chris
Thanks,
Rafael