Your elf-bfd.h and elflink.c changes caused ld regressions for cris-axis-elf (likely others)

Your elf-bfd.h and elflink.c changes caused ld regressions for cris-axis-elf (likely others)

[Hans-Peter Nilsson]

(Forgot to CC the list.  I'm not *that* polite. :-)

Apparently (because it's the only change) this change:

2005-03-18  Alan Modra  <amodra@bigpond.net.au>

	* elf-bfd.h (_bfd_elf_link_renumber_dynsyms): Delete.
	* elflink.c (_bfd_elf_link_renumber_dynsyms): Make static, add
	section_sym_count param, and return number of section symbols.
	(bfd_elf_size_dynamic_sections): Clear section symbol area of
	.dynsym contents.  Don't bother calling swap_symbol_out on the
	first all-zero dynsym.
	(elf_mark_used_section): Formatting.  Avoid twiddling flags in
	special sections like bfd_abs_section.
	(bfd_elf_gc_sections): Spelling fix.

caused these failures for cris-axis-elf:

Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-checks/checks.exp ...
FAIL: check sections 1
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-cris/cris.exp ...
FAIL: ld-cris/def3
FAIL: ld-cris/ldsym1
FAIL: ld-cris/noglob1
FAIL: ld-cris/obj1
FAIL: ld-cris/obj2
FAIL: ld-cris/undef3
FAIL: ld-cris/v10-va
FAIL: ld-cris/v32-ba-1
FAIL: ld-cris/v32-va
FAIL: ld-cris/va-v10
FAIL: ld-cris/va-v32
FAIL: ld-cris/warn2
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-crx/crx.exp ...
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-cygwin/exe-export.exp ...
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-d10v/d10v.exp ...
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-discard/discard.exp ...
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-elf/elf.exp ...
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-elf/exclude.exp ...
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-elf/frame.exp ...
Running /h/hp/binutils/cvs_latest/src/ld/testsuite/ld-elf/sec64k.exp ...
FAIL: ld-elf/64ksec

All errors look like:
/n/asic_slask/hp/autobinutest/ld/ld-new: tmpdir/asm.x: could not find output section .ctors for input section .ctors
/n/asic_slask/hp/autobinutest/ld/ld-new: final link failed: Nonrepresentable section on output
FAIL: check sections 1

brgds, H-P

Re: Your elf-bfd.h and elflink.c changes caused ld regressions for cris-axis-elf (likely others)

[H. J. Lu]

On Fri, Mar 18, 2005 at 04:08:28PM +0100, Hans-Peter Nilsson wrote:
> (Forgot to CC the list.  I'm not *that* polite. :-)
> 
> Apparently (because it's the only change) this change:
> 
> 2005-03-18  Alan Modra  <amodra@bigpond.net.au>
> 
> 	* elf-bfd.h (_bfd_elf_link_renumber_dynsyms): Delete.
> 	* elflink.c (_bfd_elf_link_renumber_dynsyms): Make static, add
> 	section_sym_count param, and return number of section symbols.
> 	(bfd_elf_size_dynamic_sections): Clear section symbol area of
> 	.dynsym contents.  Don't bother calling swap_symbol_out on the
> 	first all-zero dynsym.
> 	(elf_mark_used_section): Formatting.  Avoid twiddling flags in
> 	special sections like bfd_abs_section.
> 	(bfd_elf_gc_sections): Spelling fix.
> 
> caused these failures for cris-axis-elf:
> 

I will check it in shortly.


H.J.
----
--- bfd/elflink.c.zero	2005-03-18 06:23:37.000000000 -0800
+++ bfd/elflink.c	2005-03-18 06:55:10.322391518 -0800
@@ -9012,7 +9012,9 @@ elf_mark_used_section (struct elf_link_h
       || h->root.type == bfd_link_hash_defweak)
     {
       asection *s = h->root.u.def.section;
-      if (s != NULL && s->output_section != NULL && s->output_section != s)
+      if (s != NULL
+	  && s->output_section != NULL
+	  && !bfd_is_const_section (s->output_section))
 	s->output_section->flags |= SEC_KEEP;
     }
 

Re: Your elf-bfd.h and elflink.c changes caused ld regressions for cris-axis-elf (likely others)

[Alan Modra]

On Fri, Mar 18, 2005 at 07:37:23AM -0800, H. J. Lu wrote:
> On Fri, Mar 18, 2005 at 04:08:28PM +0100, Hans-Peter Nilsson wrote:
> > (Forgot to CC the list.  I'm not *that* polite. :-)
> > 
> > Apparently (because it's the only change) this change:
> > 
> > 2005-03-18  Alan Modra  <amodra@bigpond.net.au>
> > 
> > 	* elf-bfd.h (_bfd_elf_link_renumber_dynsyms): Delete.
> > 	* elflink.c (_bfd_elf_link_renumber_dynsyms): Make static, add
> > 	section_sym_count param, and return number of section symbols.
> > 	(bfd_elf_size_dynamic_sections): Clear section symbol area of
> > 	.dynsym contents.  Don't bother calling swap_symbol_out on the
> > 	first all-zero dynsym.
> > 	(elf_mark_used_section): Formatting.  Avoid twiddling flags in
> > 	special sections like bfd_abs_section.
> > 	(bfd_elf_gc_sections): Spelling fix.
> > 
> > caused these failures for cris-axis-elf:

Huh.  What cris weirdness is this exposing?  Symbols defined in output
sections?

Looks..  Oh, OK, the symbols defined in output sections are some
provided by the linker script.  I guess it's not cris weirdness after
all.  ;-)

> I will check it in shortly.
> 
> 
> H.J.
> ----
> --- bfd/elflink.c.zero	2005-03-18 06:23:37.000000000 -0800
> +++ bfd/elflink.c	2005-03-18 06:55:10.322391518 -0800
> @@ -9012,7 +9012,9 @@ elf_mark_used_section (struct elf_link_h
>        || h->root.type == bfd_link_hash_defweak)
>      {
>        asection *s = h->root.u.def.section;
> -      if (s != NULL && s->output_section != NULL && s->output_section != s)
> +      if (s != NULL
> +	  && s->output_section != NULL
> +	  && !bfd_is_const_section (s->output_section))
>  	s->output_section->flags |= SEC_KEEP;
>      }

You may as well leave off the !bfd_is_const_section test, or revert to
as you had it.  At one stage bdf_abs_section was const, so attempting to
change flags would cause segfaults.  I overreacted to think it incorrect
to twiddle bfd_abs_section.  It doesn't really matter to just set
SEC_KEEP.

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre

Re: Your elf-bfd.h and elflink.c changes caused ld regressions for cris-axis-elf (likely others)

[H. J. Lu]

On Sat, Mar 19, 2005 at 10:06:59AM +1030, Alan Modra wrote:
> > --- bfd/elflink.c.zero	2005-03-18 06:23:37.000000000 -0800
> > +++ bfd/elflink.c	2005-03-18 06:55:10.322391518 -0800
> > @@ -9012,7 +9012,9 @@ elf_mark_used_section (struct elf_link_h
> >        || h->root.type == bfd_link_hash_defweak)
> >      {
> >        asection *s = h->root.u.def.section;
> > -      if (s != NULL && s->output_section != NULL && s->output_section != s)
> > +      if (s != NULL
> > +	  && s->output_section != NULL
> > +	  && !bfd_is_const_section (s->output_section))
> >  	s->output_section->flags |= SEC_KEEP;
> >      }
> 
> You may as well leave off the !bfd_is_const_section test, or revert to
> as you had it.  At one stage bdf_abs_section was const, so attempting to
> change flags would cause segfaults.  I overreacted to think it incorrect
> to twiddle bfd_abs_section.  It doesn't really matter to just set
> SEC_KEEP.
> 

I reverted it to the old version.


H.J.