Re: procfs problems

H. Peter Anvin (hpa@transmeta.com)
Tue, 15 Apr 1997 15:19:44 -0700 (PDT)


> How? Naively, I would assume that it is trivially harder to look for
> i?86 instead of i386 if you don't care about the extra information,
> and non-Linux Unices are going to return something different anyway.

Surprise! Programs *don't*, unless they are written explicitly for
Linux.

> In fact, on most Unices the information is *less* useful than it is on
> Linux. A quick survey of UNIX variants I have access to do vary
> widely with respect to what uname -m reports, so I'm not sure what
> 'common' behavior those programs could rely on. [IRIX reports CPU
> board type; AIX reports the machine ID number; Solaris returns the
> hardware configuration type].
>
> FYI:
> AIX: 006027974C00
> Solaris: sun4m
> IRIX: IP20
> Linux: i586
>
> what exactly does 'i586' break that wouldn't be horridly broken on any
> of the above architectures (especially AIX!) anyway?

It seems that for every processor generation, programs break this
way. However, what you point out here might be a misunderstanding on
my part: maybe the right thing to do is to implement "uname -p" and
have *it* return "i386".

I just checked my Solaris box, and you're right: uname -m returns
"sun4m" and uname -p returns the processor architecture (sparc).

-hpa