> i consider our ELF shlib implementation broken without a way
    > to globally define places to look for shlibs, *if i want*.
   
   ...and I consider our ELF implementation broken if we supply
   ld.so.conf functionality for it.

can you expand on "broken" further?  what, specifically, is
broken about it?  and don't restate the reason below -- i
don't think it *matters*.

   
   ELF JUST WASN'T DESIGNED TO WORK THAT WAY.  Plenty of other systems can
   deal with the lack of it, so I don't see why we must adopt Linux/FreeBSD
   braindamage for our ELF.


i don't really care what linux or freebsd do.  i care what we do.
currently, "we" support ld.so.conf.  this is win to me.  it has
saved my ass a number of times!  and now you want to make me
switch execution formats and lose this feature.


so what if ELF wasn't designed that way?  it does not break
anything to have to this feature -- all it does is retain
something NetBSD has supported for over 5 years, that i've
used and appriecated number of times (some of these would
not have happened with rpaths in a.out, but not all).


"other systems" don't so much "deal with a lack of it" as they
just don't have this feature.  hell, we don't even have a
tool to alter the in-binary rpath, which some "other systems"
at least have.  i'd like to have this tool inaddition to
ld.so.conf -- sometimes you just can't rewrite a binary not
matter what.