kern/77355: Detect i*86 subarches for uname
Bruce Evans
bde at zeta.org.au
Fri Feb 11 06:10:30 PST 2005
The following reply was made to PR kern/77355; it has been noted by GNATS.
From: Bruce Evans <bde at zeta.org.au>
To: "Devon H. O'Dell" <dodell at sitetronics.com>
Cc: Robert Millan <rmh at debian.org>, freebsd-gnats-submit at freebsd.org
Subject: Re: kern/77355: Detect i*86 subarches for uname
Date: Sat, 12 Feb 2005 01:05:50 +1100 (EST)
[gnats restored in Cc:]
On Fri, 11 Feb 2005, Devon H. O'Dell wrote:
> On Fri, 2005-02-11 at 13:25 +0100, Robert Millan wrote:
> > On Fri, Feb 11, 2005 at 11:18:25AM +0100, Devon H. O'Dell wrote:
> > > >
> > > > With this change, the config.guess triplet becomes i686-unknown-freebsd5.3
> > > > (or whatever suitable). Some programs detect this and use it for optimisation.
> > >
> > > The proper way to specify optimizations in FreeBSD is with relevant
> > > entries in /etc/make.conf.
> >
> > But this only affects the port system when passing --host and --build to
> > configure scripts, right?
> >
> > I'm more concerned about programs that run config.guess on their own.
>
> Actually, make.conf is used to modify various flags used in the files
> in /usr/share/mk and /usr/ports/Mk. Nothing would be passed to --host
> and --build; the Makefile would use the desired CFLAGS when building,
> which might include -march=whatever.
>
> Since FreeBSD encourages the use of its ports collection (and other
> package systems such as pkgsrc use similar tricks to do this), I really
> don't see this as an issue. Perhaps the only thing I see this useful for
> is if you don't particularly care to use any package system, which isn't
> really good for the project since we don't get to benefit from the
> software you're using. It's farily simple to wrap any software into a
> port file, regardless of what utilities you are using.
It is needed for correctness. From "man uname | col -bx":
%%%
-m Write the type of the current hardware platform to standard out-
put.
...
-p Write the type of the machine processor architecture to standard
output.
%%%
-p is supposed to give the arch (e.g., -i386) and -m is supposed to
give the platfrom (is that the sub-arch?) (e.g., i686). It is useless
for these to return the same string. (i686 is also useless, since it
is the same for all i386's newer than about 8 years old, but that is
another bug. The hw.model sysctl gives more useful info (e.g.,
"AMD Athlon(tm)"), but uname(1) only uses uname(2) which doesn't go
near this sysctl.)
However, I don't like changing the -a output (-a gets -m but not -p).
Bruce
More information about the freebsd-bugs
mailing list