Re: Call for port reports

Bruce Momjian wrote:
> Peter Eisentraut wrote:
> > Bruce Momjian writes:
> >
> > > > BUT: The default CFLAGS are set by configure to -O2, although the template
> > > > wants -O. I manually modified the CFLAGS to -O after configure.
> > >
> > > template/alpha has:
> > >
> > > case $host_cpu in
> > > alpha*) CFLAGS="$CFLAGS -O";; # alpha has problems with -O2
> > > esac
> > >
> > > Is this not getting invoked?
> >
> > After further consideration, I think that the recent patch series that
> > tried to centralize the CFLAGS handling in configure should be reverted to
> > configure.in revision 1.293. Otherwise, it's much to complicated to
> > handle all the special cases. There is, after all, a reason we have been
> > forced to keep it this way all these years.
>
> Remember the old code had CFLAGS="" in lots of platforms, meaning they
> got no optimization.
>
> It seems right now Alpha is our only problem, and it is really just a
> message problem because the later flags override the earlier ones. Why
> can't get just remove -O2 from the alpha CFLAGS line via makefile magic?
> Frankly, we could just do CFLAGS="-O" and be done with it because we
> would not be bringing in the -O2, but I would rather keep it clean and
> remove just -O2.
>
> I don't think going backwards is a good solution because it spreads the
> problem down to the templates again.

In fact, another question is why this alpha test is only done in
freebsd? Certainly other alpha gcc platforms must have problems with
-O2?

I am inclined to add something to configure.in for all alpha compiles
that changes -O2 to -O.

--
Bruce Momjian | http://candle.pha.pa.us
pgman(at)candle(dot)pha(dot)pa(dot)us | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup. | Newtown Square, Pennsylvania 19073