ICE in dwarfout.c during bootstrap

[Robert Lipe]

I think the last time I tried a build on i686-pc-sco3.2v5.0.5 was around
April 4.  I did have some dwarf-related problems that showed up then.
They just got worse. :-)

During a plain bootstrap, cc1 now aborts.


    /play/negcs/gcc/xgcc -B/play/negcs/gcc/ -O2   -DIN_GCC    -g -O2 -I./include   -g1  -DIN_LIBGCC2 -D__GCC_FLOAT_NOT_NEEDED    -I. -I../../egcs/gcc -I../../egcs/gcc/config -c ${file}; \
    if [ $? -eq 0 ] ; then true; else exit 1; fi; \
    ar rc tmplibgcc2.a ${oname}.o; \
    rm -f ${name}.s ${oname}.o; \
  fi; \
done
../../egcs/gcc/frame
/play/negcs/gcc/xgcc -B/play/negcs/gcc/ -O2   -DIN_GCC    -g -O2 -I./include   -g1  -DIN_LIBGCC2 -D__GCC_FLOAT_NOT_NEEDED    -I. -I../../egcs/gcc -I../../egcs/gcc/config \
  -c ../../egcs/gcc/cp/tinfo.cc
../../egcs/gcc/dwarfout.c:5063: Internal compiler error in function dwarfout_file_scope_decl


This is this code in dwarfout.c:

  /* If this ..._DECL node is marked to be ignored, then ignore it.  We
     gotta hope that the node in question doesn't represent a function
     definition.  If it does, then totally ignoring it is bound to screw
     up our count of blocks, and that it turn will completely screw up the
     labels we will reference in subsequent AT_low_pc and AT_high_pc
     attributes (for subsequent blocks).  (It's too bad that BLOCK nodes
     don't carry their own sequence numbers with them!)  */

  if (DECL_IGNORED_P (decl))
    {
      if (TREE_CODE (decl) == FUNCTION_DECL && DECL_INITIAL (decl) != NULL)
        abort ();
      return;
    }



Let me know Friday if there's anything I can test or debug to help 
find this.   I'll be out of town next week.

Thanx,
RJL