Re: [PATCH] arch: new program

[Jim Meyering]

Paul Eggert <address@hidden> wrote:
> Jim Meyering <address@hidden> writes:
>
>> I suppose arch should not be used in new scripts, but that's just
>> a gut feeling.  If so, documentation should make it clear.
>
> I agree that the documentation should advise people to use uname and
> to avoid arch.  'uname' is standardized, but 'arch' is not.  The Sun
> documentation already advises people not to use 'arch'.
>
> One other thing.  This implementation of 'arch' is incompatible with
> Solaris.  Solaris 'arch' uses the following algorithm:
>
>    If "uname -p" outputs "sun4", output "sun4"; otherwise output
>    whatever "uname -m" returns.
>
> Also, Solaris 'arch' accepts a "-k" option that causes "arch" to
> behave just like "uname -m".
>
> Also, Solaris 'arch' accepts an optional operand that causes it to
> silently exit with status 0 if "arch" would have printed that operand,
> 1 otherwise.

Ouch.  This makes it look like coreutils should not install arch
by default.  What do you think of a new configure-time option that
would list extra programs like arch that you'd like to install?
I would probably add "su" to the list, since most installers don't
want the version from coreutils.

I'm not too interested in making arch compatible with the Solaris one,
so there should probably be some fail-safe to force installation of
"arch" on a Solaris system.  The goal was simply to pull in the one
from the util-linux package.

I hate to say it, after Karel has done most of the work, but I suppose
simply not adding it to coreutils should be considered an option, too.

Opinions?