On 20/07/15 22:16, Assaf Gordon wrote:
> tag 21098 notabug
> stop
> 
> Hello,
> 
> On 07/20/2015 03:26 PM, Norbert de Jonge wrote:
>> Maybe someone has time and energy to make some minor improvements to
>> uname's man page. The problem lies in the vagueness and similarity of
>> the options -m, -p and -i, combined with the program's unpredictable
>> output.
> 
> The man-page could be re-worded, but,
> the values that are reported by uname are very system specific.
> 
> The Coreutil FAQ contains an entry for uname:
>   http://www.gnu.org/software/coreutils/faq/coreutils-faq.html#uname-is-system-specific
> 
> And some past discussions contain more information:
>    http://debbugs.gnu.org/cgi/bugreport.cgi?bug=14388
>    http://lists.gnu.org/archive/html/bug-coreutils/2003-10/msg00098.html
>    http://lists.gnu.org/archive/html/bug-coreutils/2004-11/msg00032.html
> 
>> To make people's lives easier, the description for -m should be
>> changed. I also think it would be useful to add to the description, in
>> parentheses, "e.g. x86_64, i686".
> [...]
>> "If you want to determine whether a system is 32-bit or 64-bit, use this option."
> 
> mentioning "x86_64/i686" in the documentation is Linux-kernel specific, and coreutils is not limited to Linux-kernel, and would be incorrect/misleading on other systems (e.g. on BSD systems the returned value is "amd64").
> 
> In practice,
> The values of 'uname' are not standardized over all OSes/hardwares.
> The recommended way is to first detect the system/kernel type (e.g. just 'uname'),
> and then decode the hardware type, using the values that are common to that kernel.
> An example to such code is given here:
>    http://debbugs.gnu.org/cgi/bugreport.cgi?bug=14388#14
> 
>> Now, that works on Ubuntu, but not on, for example, Arch. Ubuntu, Mint
>> and probably also Debian, return the same thing for -m, -p and -i.
> 
> I'm not familiar with Arch;
> some GNU/Linux distros return the values reported by the Linux Kernel, and GNU coreutils does not change that value.
> Other distros patch GNU coreutils (or the kernel?) to return other values.
> See related discussion here:
>     http://lists.gnu.org/archive/html/bug-coreutils/2012-11/msg00173.html
>     http://lists.gnu.org/archive/html/bug-coreutils/2012-11/msg00177.html
> 
> I believe that if the kernel is detected as "Linux", then the developer can assume that "uname -m" would suffice (based on known Linux kernel values) [other participants - please correct me if that's wrong]:
> 
> e.g:
>      test "$(uname)" = Linux \
>         && test "$(uname -m)" = x86_64 \
>         && echo 64bit-linux || echo other

BTW for a more direct/portable method, one might consider:

  test "$(getconf LONG_BIT)" = '64'

Also that would work better in a 32 chroot on 64 bit kernel.

> 
> For some 'uname -m' values of common OSes (not just Linux-based), see here:
>     http://pretest.nongnu.org/versions/

Very useful info thanks!

I see this was also discussed at:
http://bugs.gnu.org/13001
http://bugs.gnu.org/15757

Some quick testing here gives:

System                  uname -i        uname -m        uname -p
----------------------------------------------------------------
OS X - coreutils        Macmini7,1      x86_64          i386
OS X - native           illegal option  x86_64          i386
Solaris 10              SUNW,SPARC...   sun4v           sparc
Fedora 22               x86_64          x86_64          x86_64
GNU/Linux               unknown         x86_64          unknown

I find it incorrect that the Linux distros just copied -m to -i and -p.

BTW for a script using -m and -p (but not -i) values see:
http://git.sv.gnu.org/gitweb/?p=gnulib.git;f=build-aux/config.guess;hb=HEAD

I wonder should be just be more explicit about discouraging
use of the -i and -p options, and leave it at that.
That's done in the attached.

thanks,
Pádraig.