Re: [RFC: 2.6 patch] kernel/sys.c: possible cleanups

[Andrew Morton <akpm@osdl.org>]

Arjan van de Ven <arjan@infradead.org> wrote:
>
> On Mon, 2006-05-01 at 00:49 -0700, Andrew Morton wrote:
> > Arjan van de Ven <arjan@infradead.org> wrote:
> > >
> > > > If removing exports requires a process, adding exports requires a 
> > >  > similar process.
> > > 
> > >  alternatively we should bite the bullet, and just stick those 900 on the
> > >  "we'll kill all these in 3 months" list, have a thing to disable them
> > >  now via a config option (so that people actually notice rather than just
> > >  having them in the depreciation file) and fix the 5 or 10 or so that
> > >  actually will be used soon in those 3 months.
> > > 
> > 
> > I'd instead suggest that we implement a new EXPORT_SYMBOL_UNEXPORT_SCHEDULED
> 
> EXPORT_UNUSED_SYMBOL() ? (with comment of date)

OK.

> (well you didn't apply the patch for that I sent you so I suppose you
> hate it ;)

err, memory fails me.  Wanna dig it out?

> 
> > (?) and use that.  Suitable compile-time and modprobe-time warnings would
> 
> compiletime warning is hard, because it would require changing
> prototype, which is a lot more churn, and I'll bet a lot of exports
> don't even have a prototype at all. modprobe time is easy. (middle
> ground would be depmod time; that's almost compile time I suppose but a
> lot easier, but might need modutils help)

modprobe-time printk is good.

> > be needed.  Put the unexport date in a comment at the
> > EXPORT_SYMBOL_UNEXPORT_SCHEDULED site or even in the modprobe-time warning
> > message, if that's convenient:
> > 
> > EXPORT_SYMBOL_UNEXPORT_SCHEDULED(foo, "Dec 2006");
> 
> comment is easy, putting a date in is really sucky since it grows the size of exports even more..
> (which means people pay even more export tax than they do today)

ok, whatever.