Re: [parisc-linux] Another possible linker problem

Re: [parisc-linux] Another possible linker problem

[Alan Modra]

[cc to list in case others have the same problem.]

On Thu, 6 Jul 2000, Paul Bame wrote:

> I tried to x-compile ncurses-5.0 and the executables (tic, clear, tput,
> and so forth) will not link.  They seem to be created from a couple of
> large .o files, produced from 'ld -r' I think.  Here're the messages:
> 
> hppa1.1-linux-gcc ../objects/tic.o ../objects/dump_entry.o -L../lib -L/usr/lib -lncurses      -o tic
> /home/bame/xc/lib/gcc-lib/hppa1.1-linux/2.96.0614/../../../../hppa1.1-linux/bin/ld: can not size stub section: Bad value

I reckon this error is probably fixed by the patch I checked in last
night.  elf32_hppa_size_stubs had this code in it

		      if (hash->root.type == bfd_link_hash_undefined
			  || hash->root.type == bfd_link_hash_undefweak)
			continue;

		      while (hash->root.type == bfd_link_hash_indirect
			     || hash->root.type == bfd_link_hash_warning)
			hash = (struct elf_link_hash_entry *)
			  hash->root.u.i.link;

and the problem is that we may have an indirect symbol (results from
symbol versioning) pointing to an undefined symbol.  The simple fix is to
traverse the indirect links before checking for undefined syms.

Regards, Alan Modra
-- 
Linuxcare.  Support for the Revolution.

Re: [parisc-linux] Another possible linker problem

[Jeffrey A Law]

  In message <Pine.LNX.4.21.0007070848540.12175-100000@front.linuxcare.com.au>y
ou write:
  > [cc to list in case others have the same problem.]
  > I reckon this error is probably fixed by the patch I checked in last
  > night.  elf32_hppa_size_stubs had this code in it
  > 
  > 		      if (hash->root.type == bfd_link_hash_undefined
  > 			  || hash->root.type == bfd_link_hash_undefweak)
  > 			continue;
  > 
  > 		      while (hash->root.type == bfd_link_hash_indirect
  > 			     || hash->root.type == bfd_link_hash_warning)
  > 			hash = (struct elf_link_hash_entry *)
  > 			  hash->root.u.i.link;
  > 
  > and the problem is that we may have an indirect symbol (results from
  > symbol versioning) pointing to an undefined symbol.  The simple fix is to
  > traverse the indirect links before checking for undefined syms.
Yea, I don't think I ever implemented indirects in the PA tools.  I suspect
there's other PA code that doesn't handle them correctly.

I'm glad to see someone making the right things happen with that old crusty
code.


jeff