Re: [PATCH] Remove -ffunction-sections debugging warning (PR libstdc++/35597)

[David Edelsohn <dje at watson dot ibm dot com>]

>>>>> Daniel Jacobowitz writes:

Dan> I think most optimizations affect debugging less seriously, but I
Dan> don't know exactly what the effects are for this.  Does anyone?
Dan> I suspect it is of the general nature "debugging information is
Dan> completely useless", and I suspect that it's primarily talking about
Dan> COFF/PE.  If it's really bad somewhere, maybe we should replace the
Dan> whitelist with a blacklist.

	If we want to change the code so that it only warns for COFF/PE,
that's fine.  Note that -ffunctoin-sections only is enabled for targets
and object file formats that support it.  Earlier we have:

  if (! targetm.have_named_sections)
    {
      if (flag_function_sections)
        {
          warning (0, "-ffunction-sections not supported for this target");
          flag_function_sections = 0;
        }

Presumably if the target maintainer addss the support for
-ffunction-sections, he or she knows the trade-off.  Also, GCC consider
XCOFF a superset of COFF, so this will become convoluted.

	I suggest that it would be better for targets to warn about this
limitation themselves instead of GCC assuming generalizing limitations to
whole classes of object file formats.

	Again, I ask that either this patch be approved or the offending
change from libstdc++ be removed from mainline and 4.3 branch ASAP.

Thanks, David