PC-relative Processing

[Tracy dot Kuhrt at microchip dot com]

In the fixup_segment function, there exists the following code.  Could 
someone please tell me why pcrel is set to 0, but MD_PCREL_FROM_SECTION is 
not subtracted from add_number?  I am working on a new port of binutils, 
and this is causing me to calculate an incorrect offset when I do one of 
the following bra foo - . or bra . - foo.

      if (sub_symbolP)
        {
...
          else if (S_GET_SEGMENT (sub_symbolP) == add_symbol_segment
                   && SEG_NORMAL (add_symbol_segment))
            {
              /* Difference of 2 symbols from same segment.
                 Can't make difference of 2 undefineds: 'value' means
                 something different for N_UNDF.  */
#ifdef TC_I960
              /* Makes no sense to use the difference of 2 arbitrary 
symbols
                 as the target of a call instruction.  */
              if (fixP->fx_tcbit)
                as_bad_where (fixP->fx_file, fixP->fx_line,
                              _("callj to difference of 2 symbols"));
#endif /* TC_I960  */
              add_number += S_GET_VALUE (add_symbolP) -
                S_GET_VALUE (sub_symbolP);

              add_symbolP = NULL;
              pcrel = 0;        /* No further pcrel processing.  */

              /* Let the target machine make the final determination
                 as to whether or not a relocation will be needed to
                 handle this fixup.  */
              if (!TC_FORCE_RELOCATION_SECTION (fixP, this_segment_type))
                {
                  fixP->fx_pcrel = 0;
                  fixP->fx_addsy = NULL;
                  fixP->fx_subsy = NULL;
                }
            }

Any thoughts on what the appropriate solution is?  I do not wish to break 
any other ports by suggesting adding the following line before turning off 
pc-relative processing:
        add_number -= MD_PCREL_FROM_SECTION (fixP, this_segment_type);


Tracy