[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Mozilla shouldn't force -O2 optimisation on user

On Mon, Apr 14, 2003 at 12:24:44PM -0700, Adam Weinberger wrote:
> Hash: SHA1
> >> (04.14.2003 @ 1209 PST): Maxim Sobolev said, in 0.6K: <<
> > I think that every concept of "WITHOUT_OPTIMIZE" added in
> > rev.1.136 of ports/www/mozilla/Makefile is wrong. No port
> > should force optimisation level on user - we have CFLAGS
> > for this and each user should be able to select any
> > performance/stability ratio he wants.
> >> end of "Mozilla shouldn't force -O2 optimisation on user" from Maxim Sobolev <<
> I, as a user, have always appreciated knowing when developers thought
> certain optimization levels were useful enough to suggest that people
> use them. I never go above -O2 unless I have specific reason to believe
> that -O2 won't cause any problems. And for ports, WITH_OPTIMIZATION
> flags are what I often look for.
> I have no problem with making it default of off, and making WITH_ an
> option, but I don't think the option should be removed altogether.

In general I agree with you. The problem is that we have two
classes of users: ones that want best-of-breed performance
right out of the box and ready to rely on whatever mantainer
considers as safe, and ones that would like to stay on safe
side or/and want an easy way to have a full control on code
generation options. From my developer's perspective I fall info
the second category - often for debugging reasons I need to
recompile one port or another with "-g3 -O0", just to make sure
that optimisation bugs don't stay on the way. Moreover, I don't
really trust anything above -O - the performance gain is
unnoticeable, while the risk of hitting one or another gcc
optimisation bug is significantly higher, not even to mention that
with -O2 and above compilation performance is much worse than
with -O.

What we really need, is some global option honoured by all
performance critical ports, which will select one behaviour
or another.


Visit your host, monkey.org