Re: changing the default i386 compilation target? sse2?

[Greg Troxel <gdt%lexort.com@localhost>]

Given the comments, it seems like more than a tiny number of people are
using pre-sse2 hardware (even if some are running old netbsd or not
using recent binary packages and thus wouldn't be affected).

I see options:

  A) just set defaults to -march=i686 -msse2, and let people change that
  if they want to build their own packages

  B) continue as we are, so that packages get marked BROKEN_ON for i386

     B1) Also hope someone tilts at upstream windmills and fixes the
         needs-sse2 bugs.

  C) adopt a mk/cpu/i386-sse2.mk (perhaps mk/i386-sse2.mk if we think an
     extra dir hurts more than the risk of N files) that will add the
     flags if i386, *just like mk/atomic.mk does*, and include that in
     packages that insist on sse2 instead of BROKEN

A has been argued to be unreasonable.
B isn't really reasonable either.

C seems ok to me, with no real downside as anything that needs sse2 and
hence won't be usable on old i386 wouldn't have been built anyway.  It
seems no worse than atomic.mk in any case.

Please reply privately with preference between

  mk/cpu/i386-sse2.mk
  mk/i386-sse2.mk

if it's just preference without some illuminating rationale.  (Clearly
one is an extra dir, and the other risks many files).  With illuminating
rationale or an alterative with-rationale proposal, please followup on
list.

Are there any objections to option C?   If not, I would head to
mk/cpu/i386-sse2.mk.