Re: Objcopy fix for relocation sections

[Daniel Jacobowitz <drow at false dot org>]

On Thu, Aug 05, 2004 at 06:57:04PM +0930, Alan Modra wrote:
> On Tue, Aug 03, 2004 at 06:56:01PM -0400, Daniel Jacobowitz wrote:
> > 	* elf.c (assign_file_positions_except_relocs): Revert unintended
> > 	change from 2004-04-08.
> 
> Oops, please fix.
> 
> > 	(_bfd_elf_set_section_contents): Call
> > 	_bfd_elf_assign_file_positions_for_relocs when starting output
> > 	for executables and shared libraries.
> 
> Doesn't this break mips?  For example, I see an assignment to
> rel_hdr->sh_size in mips_elf64_write_rel.  That size should determine
> section placement, but you're doing the placement before
> mips_elf64_write_rel is called.

Sparc64 does the same thing.  The problem looks plausible.  But I can't
reproduce it; I can't get the linker to pass MIPS combined relocations
through to a dynamic or executable module.  If I try to fake it using
-q, some code that specifically takes care of the MIPS ugliness falls
over in elflink.h (now in elflink.c, I'm looking at an older
toolchain): in elf_bfd_final_link, esdi->rel_hdr.sh_entsize == 0.

So I have no way to check ;-)

If you're right, then I would like to go back to my first patch, the
one which places each reloc section as _bfd_elf_set_section_contents
is called on it.  That will change the behavior of objdump without
changing the behavior of the assembler or linker.

-- 
Daniel Jacobowitz