Re: CVS commit: src

[David Young <dyoung%pobox.com@localhost>]

On Thu, May 26, 2011 at 05:35:20PM +0100, David Laight wrote:
> On Thu, May 26, 2011 at 11:32:51AM -0500, David Young wrote:
> > On Thu, May 26, 2011 at 06:10:42PM +0200, Lars Heidieker wrote:
> > > On 05/26/11 06:25, Masao Uebayashi wrote:
> ...
> > > with those changes I can't compile a kernel without USERCONF option set.
> > > The changes in x86_machdep.c should be with in if defs on that options
> > > 
> > > Ok to commit the attached patch?
> > 
> > Conditional compilation clutters the code up.  What if instead you add
> > to kern_stub.c,
> > 
> > #include <sys/userconf.h>
> > __weak_alias(userconf_bootinfo, voidop);
> 
> I'd either:
>     leave userconf_bootinfo() defined but with no body
> or:
>     #define userconf_bootinfo()
> (or maybe both!)

If you do both, ISTM that the preprocessor will turn this:

void
userconf_bootinfo(void)
{
#ifdef USERCONF
#endif /* USERCONF */
}

to this:

void
{
#ifdef USERCONF
#endif /* USERCONF */
}

which won't compile.

Leaving userconf_bootinfo() undefined but with no body seems equivalent
to aliasing to voidop() *except* that there's an extra 'ret' instruction
in the kernel somewhere. :-) Also, it precludes linking with an object
containing a real userconf_bootinfo(), later.  There's still the #ifdef
clutter of userconf_bootinfo(), albeit greatly reduced.

Dave

-- 
David Young             OJC Technologies
dyoung%ojctech.com@localhost      Urbana, IL * (217) 344-0444 x24