Re: elf_gc_sections and relocatable sections

[Nick Clifton <nickc at redhat dot com>]

Hi Earl,


> The code fragment at elf_gc_sections() is key to resurrecting my
> patch with 2.15.91.0.2:
>
> >   if (!get_elf_backend_data (abfd)->can_gc_sections
> >       || info->relocatable
> >       || info->emitrelocations
> >       || info->shared
> >       || !is_elf_hash_table (info->hash))
> >     {
> >       (*_bfd_error_handler)(_("Warning: gc-sections option ignored"));
> >       return TRUE;
> >     }
>
>
> Can anyone shed any light as to why elf_gc_sections() cannot
> proceed if info->relocatable or info->emitrelocations is set?

I think that for the info->emitrelocations case the theory is that if 
the user wants the relocations preserved then the linker does not know 
the exact purpose of the binary that it is creating.  Therefore it is 
better to be paranoid and assume that the user might want the 
relocations in sections that would otherwise be removed via garbage 
collection.

As for info->relocatable I believe that you already covered that in your 
emails with Richard Henderson.  The output of a relocatable link can be 
used as the input for another link, so unless the linker can definitely 
establish any and all entry points into the files being linked, it 
cannot determine which sections are unreferenced and hence can be 
discarded via garbage collection.

Possibly the simplest way around these problems is to add a new command 
line switch, eg --force-gc-sections which causes the garbage collection 
operation to happen despite the settings of other switches.

Cheers
  Nick