[Patch mach-o/gas] fix thinko in indirect symbol handling.

[Patch mach-o/gas] fix thinko in indirect symbol handling.

[Iain Sandoe]

indirect symbols are stored away as we encounter them, and then  
processed/validated fairly late in the output.

this causes a problem if one uses a "L" local symbol as an indirect  
(something which compiler-output doesn't seem to, I only found this  
with one of my artificial test-cases).

the problem is that (the output-time) processing of the symbol as an  
indirect causes it to be 'promoted' to a real one (with a bfd  
counterpart).  This ICEs GAS because the symtab is frozen by then.

Anyway, the simplest solution is to force 'promotion' of "L"ocal  
symbols at the point they are known to be indirect refs.

OK?
Iain

gas:

	* config/obj-macho.c (obj_mach_o_indirect_symbol): Force promotion of  
any local
	symbol used as an indirect.

iff --git a/gas/config/obj-macho.c b/gas/config/obj-macho.c
index f4706ab..925fe06 100644
--- a/gas/config/obj-macho.c
+++ b/gas/config/obj-macho.c
@@ -1182,6 +1182,11 @@ obj_mach_o_indirect_symbol (int arg  
ATTRIBUTE_UNUSED)
  	  }
  	  *input_line_pointer = c;

+	  /* The indirect symbols are validated after the symbol table is
+	     frozen, we must make sure that if a local symbol is used as an
+	     indirect, it is promoted to a 'real' one.  Fetching the bfd sym
+	     achieves this.  */
+	  symbol_get_bfdsym (sym);
  	  isym = (obj_mach_o_indirect_sym *)
  			xmalloc (sizeof (obj_mach_o_indirect_sym));

Re: [Patch mach-o/gas] fix thinko in indirect symbol handling.

[Tristan Gingold]

On Feb 10, 2012, at 1:37 PM, Iain Sandoe wrote:

> indirect symbols are stored away as we encounter them, and then processed/validated fairly late in the output.
> 
> this causes a problem if one uses a "L" local symbol as an indirect (something which compiler-output doesn't seem to, I only found this with one of my artificial test-cases).
> 
> the problem is that (the output-time) processing of the symbol as an indirect causes it to be 'promoted' to a real one (with a bfd counterpart).  This ICEs GAS because the symtab is frozen by then.
> 
> Anyway, the simplest solution is to force 'promotion' of "L"ocal symbols at the point they are known to be indirect refs.
> 
> OK?

Yes.  For sure this is a particular case, but no reasons not to handle it when it is simple.

Thanks,
Tristan.

> Iain
> 
> gas:
> 
> 	* config/obj-macho.c (obj_mach_o_indirect_symbol): Force promotion of any local
> 	symbol used as an indirect.
> 
> iff --git a/gas/config/obj-macho.c b/gas/config/obj-macho.c
> index f4706ab..925fe06 100644
> --- a/gas/config/obj-macho.c
> +++ b/gas/config/obj-macho.c
> @@ -1182,6 +1182,11 @@ obj_mach_o_indirect_symbol (int arg ATTRIBUTE_UNUSED)
> 	  }
> 	  *input_line_pointer = c;
> 
> +	  /* The indirect symbols are validated after the symbol table is
> +	     frozen, we must make sure that if a local symbol is used as an
> +	     indirect, it is promoted to a 'real' one.  Fetching the bfd sym
> +	     achieves this.  */
> +	  symbol_get_bfdsym (sym);
> 	  isym = (obj_mach_o_indirect_sym *)
> 			xmalloc (sizeof (obj_mach_o_indirect_sym));
> 
>