[klibc] Register parameters are unsafe with gcc 3.3.2

[H. Peter Anvin]

Olaf Hering wrote:
> 
> Its indeed much larger:
> 
>     dirent.o          634
>       show.o     810  918
>   mystring.o    1232  1372
>       mail.o    1988  
>     syntax.o    2125  1839
>       init.o    2992  2664
>       echo.o    3004  
>   builtins.o    3416  2916
>  miscbltin.o    3816  3752
>       kill.o    3900  
>      nodes.o    3904  3760
>      alias.o    4216  
>   memalloc.o    4828  4632
>      redir.o    5180  4704
>      error.o    5572  3652
>       trap.o    6388  4344
>       main.o    6528  6124
>      arith.o    6532  
>     output.o    6584  6188
>       test.o    7044  
>      input.o    7096  5688
>         cd.o    7988  7240
>    options.o    8764  6304
>        var.o   11520  8700
>     printf.o   11684  
>  arith_lex.o   12464  
>       exec.o   13028  11060
>       eval.o   19336  15976
>     expand.o   23572  14952
>       jobs.o   23932  12480
>     parser.o   25784  20836
>           sh  119640  71624
> 
> I will look at it, but not before July.
> And the question is, what should klibc sh do for us? Just a dumb /init helper, or more?
> mail, kill and alias is not really needed, but test and possible printf is.
> 

test is obviously essential; printf is really nice to have in certain 
circumstances, as is arithmetic (the other thing which seems to be bloating it 
up), but it's certainly not essential.  jobs.o, expand.o and var.o have really 
grown - might be something there.

My option of klibc sh is that it should be a tool to make early userspace 
scriptable.  Realistically speaking it's the only way we're going to be able 
to tell people to customize things themselves, and the ability to customize is 
one of the things which makes early userspace a major leap forward; right now 
it's all but impossible to tell people to do anything sane with nfsroot, for 
example, if the kernel doesn't do what they want it to, but it's easy to ask 
people to put in a shell script to do what they need it to.

	-hpa