[BusyBox] Re: [patch] Why busybox xargs is broken.

[Rob Landley]

On Friday 03 October 2003 05:23, Erik Andersen wrote:
> On Fri Oct 03, 2003 at 01:39:12PM +0400, Vladimir N. Oleynik wrote:
> > Erik,
> >
> > >>+			waitpid(p,&retval,0);
> > >
> > >I recommend using just wait(&retval)
> > >
> > >>+			return retval>>8;
> >
> > I do not understand, what for to change this my piece of a code.
> > I for a long time have noticed, that many do not understand
> > as vfork() function works and try to get rid of it. :-(
>
> The problem is on some Linux architectures (such as powerpc)
> vfork does not work or is not implemented at all (for example
> with all 2.0.x kernels).  In glibc (and in uClibc to try and
> match glibc behavior) for these architectures, vfork() is
> implemented with just a call to fork() and there is not even a
> call to wait() in the fake vfork implementation...
>
> So unfortunately, we can not really rely on the vfork semantics
> -- sometimes the parent process will not actually be blocked
> until the child exits.  So adding the extra call to wait() is a
> good idea to avoid wierd sequencing problems.
>
>  -Erik

Alright, would somebody please tell me what the difference between "vfork" and 
"fork" is?  I've wondered for years. The man page on vfork doesn't give you 
any useful information, it just repeats "use fork instead, infidel!" about 
five hundred times and then goes on to insult vfork's ancestry.  (Really!  Go 
read it!)

It goes out of its way to avoid giving any information about why linux 
bothered to implement it, or why one might actually want to use it...

Rob