- Makefile.am: add cxxdummy to EXTRA_DIST.
 - Makefile.am: add cpptest contents explicitly to EXTRA_DIST (instead of
                entire subdir which would include CVS/

 - Makefile.am: export host_alias, host_os, and host_cpu.

 - cexp.c: set terminal to VMIN/VTIME=1/0 in native readline (no readline, no tecla).

 - Makefile.am: export rtems_bsp, RTEMS_BSP, enable_rtemsbsp, package_subdir
   variables so third-party makefiles 'inherit' them (in case any is embedded
   in an installation path/prefix).

 - configure.ac, bfd.c: Added test for <strings.h>. rtems-4.10/newlib-1.18
   declare strcasecmp() in <strings.h> (POSIX). Hence we include that
   if present.

 - configure.ac: check for multilibbed cpukit only if host_os is RTEMS.

 - Makefile.am:
    o enforce linking undefined symbols needed by test modules (demo program)
    o link demo with CXX
 - bfdstuff.c: BUGFIX -- Ignore relocations that refer to a symbol in a
               discarded linkonce section.
               This should also fix a problem reported by Luciano Piccoli where
               a module refused to load because the .eh_frame section
               had a relocation referring to the section name of a discarded
               linkonce. Since that section had been folded into .text in
               the main executable it could not be found.
               During testing I also found that just relocating to the
               previous instance of a linkonce section is incorrect because
               to compiler would sometimes generate slightly different code
               resulting in different section sizes (original vs. discarded
               linkonce). Unless the relocation referring the discarded linkonce
               is ignored (-> value 0 which is in turn ignored by the libgcc
               run-time code) a bad/incorrect exception frame could be registered.
 - cexp.c:    Added 'sigaction' handler which prints offending SIGSEGV address.
 - cexpmod.c:
    o BUGFIX -- 'addrInModule()' algorithm skipped system module which
      is never associated with any segment. Thus 'lkaddr()' never found
      anything located in the main executable :-(
    o BUGFIX -- when a module is unloaded (linux demo) then we cannot
      just revoke 'executable' memory permission; the module memory is
      not page aligned and therefore 'x' could be removed from adjacent
      'live' module. Just always leave executable permission on...
 - configure.ac, Makefile.am, cpptest/*.c, cpptest/*.cc, cpptest/Makefile.am,
   cpptest/st.test: Added test code. So far, loading of common symbols,
   linkonce sections, c++ exceptions and constructor/destructor ordering
   are tested.
 - configure.ac: Check for availability of HAVE_SIGNALS, HAVE_SIGINFO
   on non-rtems systems.
 - m4/rtems-tools.m4: Look for 'nm' utility, too.

   cpptest/st.test: Added test code. So far, loading of common symbols,
   linkonce sections, c++ exceptions and constructor/destructor ordering
   are tested.

   cpptest/st.test: Added test code. So far, loading of common symbols,
   linkonce sections, c++ exceptions and constructor/destructor ordering
   are tested.
 - configure.ac: Check for availability of HAVE_SIGNALS, HAVE_SIGINFO
   on non-rtems systems.

    o BUGFIX -- 'addrInModule()' algorithm skipped system module which
      is never associated with any segment. Thus 'lkaddr()' never found
      anything located in the main executable :-(
    o BUGFIX -- when a module is unloaded (linux demo) then we cannot
      just revoke 'executable' memory permission; the module memory is
      not page aligned and therefore 'x' could be removed from adjacent
      'live' module. Just always leave executable permission on...

               discarded linkonce section.
               This should also fix a problem reported by Luciano Piccoli where
               a module refused to load because the .eh_frame section
               had a relocation referring to the section name of a discarded
               linkonce. Since that section had been folded into .text in
               the main executable it could not be found.
               During testing I also found that just relocating to the
               previous instance of a linkonce section is incorrect because
               to compiler would sometimes generate slightly different code
               resulting in different section sizes (original vs. discarded
               linkonce). Unless the relocation referring the discarded linkonce
               is ignored (-> value 0 which is in turn ignored by the libgcc
               run-time code) a bad/incorrect exception frame could be registered.

    o enforce linking undefined symbols needed by test modules (demo program)
    o link demo with CXX
 - configure.ac, Makefile.am, cpptest/*.c, cpptest/*.cc, cpptest/Makefile.am,
   cpptest/st.test: Added test code. So far, loading of common symbols,
   linkonce sections, c++ exceptions and constructor/destructor ordering
   are tested.

 - bfdstuff.c: Completely ignore zero-sized sections marked SEC_ALLOC
   (by removing the SEC_ALLOC flag) so that no section symbol is
   recorded. Since no space is allocated for such a section the
   value of a section symbol would be invalid and this results in
   gdb confusion...

 - libtecla/configure.in: added --enable-ansiterm configure option (enabled
   by default for RTEMS if not explicitly disabled and disabled by default
   on other OSes) to avoid dependency on curses/termcap (assuming ANSI
   terminal).

 - BUGFIX: cexpsh("script") would not terminate because of missing 'else'
   in nested if (which should have been 'else if'). This was a new bug
   introduced with the '-c' option.

   the grammar.
 - added stdio redirection.
 - deprecated '<' for sourceing scripts; should now use '.'
 - now have a dedicated FILE descriptor for parser error messages
   (can be passed to cexpCreateParserCtx())
 - added new callback argument to cexpCreateParserCtx() which
   allows the user to be notified when stdio redirection occurs.
 - added new '-c' option instructing cexpsh() to parse the option
   string argument.

 - bfdstuff.c: added undocumented global variable
   '_cexpForceIgnoreObjAttrMismatches' which can be set to nonzero
   in order to force the BFD loader to ignore object attribute
   mismatches.

 - att-gnu-powerpc.c: BUGFIX: scanning for ABI tags would access
   bad memory if tag was not present in object file.
   Added Tag_GNU_Power_ABI_Struct_Return.
   Added GNU_Power_ABI_FP_SP_HARD (new value for FP tag).
 - attrP.h, attpbfasdestroy.c, att-gnu-powerpc.c, attpbfasmatch.c,
   attbpfasprint.c, attpbfasread.c, attset.c:
   Introduced new union combining 'public' and 'opaque' values.
   Previous implementation was wrong since it would not correctly
   store 'opaque' values (arrays assumed 'public' layout).
   The new union is now used in both, the array and list parts
   of an attribute table.
 - attrP.h, attset.c, att-gnu-powerpc.c: Added new function
   'pmelf_attribute_get_tag_val()' for retrieving the value associated
   with a particular tag.

 - bfd.c: Fixed conditional compilation in elf2bfdsec():
   If 64-bit support is enabled then the 32-bit code
   must still be compiled (there is a run-time test for
   64/32-bit).

 - bfd.c, dmpehdr.c, dmprels.c, getehdr.c, pmbfd.h,
   pmelf.h, bfd-reloc-sparc.c (stub only): Added basic
   support for SPARC (no relocations yet).

   o dmprels.c: Fixed wrong member name
       (p_p_s64 -> p_s64) in dmprels.c

 - Added ChangeLog
 - minor 64-bit fixes:
   o getehdr.c: added missing 'break;' to switch
       statement in pmelf_getehdr().

   after all.

   there is no proper TLS run-time support.

 - added dependency so that demo program is relinked after a builtin-symbtab object
   file is generated.

   support them. This is just enough to get the cexp demo program
   going.

   fail if no vendor attribute parser is found because none might
   be available (nor needed) for the run-time architecture.
   Still, pmbfd will warn if it encounters a .gnu.attributes section
   but has no parser registered (but the 'normal' case where there
   is no parser but no .gnu.attributes sections either works OK).

   can test what happens if no vendor is registered but a .gnu.attributes
   section is found.
 - don't fail but ignore if same vendor is already registered (so that
   we can pass two objects for attribute comparison).
 - silenced a few compiler warnings.