[PATCH] bug 5543: do not allow register symbols to be global

[Nick Clifton]

Hi H.J.,

> We get into trouble only when we try to use bar3. This patch catches
> all cases. But it doesn't report the correct line number. 

I think that we should make an attempt to report the correct line 
number.  If there was code in read.c:s_global() for instance you could 
catch the 'bar2' example in your testcase.


> +void
> +elf_obj_validate_symbol_value (symbolS *symbolP)
> +{
> +  if (!IS_ELF)
> +    return;
> +
> +  if (symbolP->bsym->section == reg_section)
> +    {
> +      const char *name = S_GET_NAME (symbolP);
> +
> +      /* Report error only if we know the symbol name.  */
> +      if (name != reg_section->name)
> +	as_bad (_("Invalid value for symbol `%s'"), name);
> +    }
> +}

I think that it would be better if this function did not emit an error 
message but just instead returned a boolean saying whether the symbol 
was valid or not.  Then the caller the choose how to report the problem 
to the user, if it needs reporting at all.

Cheers
   Nick