Re: binutils 2.11 sh-rtems/sh-elf dies building gcc

[amodra at one dot net dot au]

On Thu, Apr 12, 2001 at 07:17:47AM -0500, Joel Sherrill wrote:
> 
> 
> amodra@one.net.au wrote:
> > 
> > On Wed, Apr 11, 2001 at 07:37:42AM -0500, Joel Sherrill wrote:
> > > In fact line 3335 is in the routine tc_gen_reloc which is
> > > clearly inside a BFD_ASSEMBLER ifdef so I am pretty sure
> > > bfd_gas=yes. :)
> > 
> > OK, this time I actually looked at the source instead of jumping
> > to wrong conclusions, but I can't see how you can arrive at the
> > result you get. :-(  Try running under gdb with a breakpoint at
> > the offending as_bad_where, then look at the stdoutput bfd info.
> > xvec->reloc_type_lookup should be pointing at
> > sh_elf_reloc_type_lookup.  If it isn't you probably have some sort
> > of build problem.  If it is, you might have a compiler problem,
> > wild pointers, or any of a range of nasty problems.
> 
> Why elf?  This is supposed to be coff.

Thinko on my part.

> To follow up a bit more ...
> 
> We have duplicated this problem on Linux and FreeBSD hosts.
> I have now duplicated this bug using the sh-coff target and
> am using gdb on that as-new with the j.s file I posted earlier.
> 
> Breaking at line 3328:
> 
> 3327      rel->howto = bfd_reloc_type_lookup (stdoutput, r_type);
> 3328      if (rel->howto == NULL)
> 3329        {
> 3330          as_bad_where (fixp->fx_file, fixp->fx_line,
> 
> (gdb) p *stdoutput
> $3 = {filename = 0x8082d84 "a.out", xvec = 0x808fce0, iostream =
                                      ^^^^
                                        |-- this one
> Which field now?