Re: powerpc allmodconfig

[Takashi Iwai <tiwai@suse.de>]

At Thu, 16 Oct 2008 11:21:57 +0300,
Adrian Bunk wrote:
> 
> On Thu, Oct 16, 2008 at 09:57:11AM +0200, Takashi Iwai wrote:
> > At Thu, 16 Oct 2008 10:38:36 +0300,
> > Adrian Bunk wrote:
> > > 
> > > On Thu, Oct 16, 2008 at 07:57:29AM +0200, Takashi Iwai wrote:
> > > > At Wed, 15 Oct 2008 21:33:37 -0700,
> > > > Andrew Morton wrote:
> > > > > 
> > > > > sound/soc/soc-dapm.c:1029: warning: 'snd_soc_dapm_connect_input' is deprecated (declared at sound/soc/soc-dapm.c:1026)
> > > > > sound/soc/soc-dapm.c:1029: warning: 'snd_soc_dapm_connect_input' is deprecated (declared at sound/soc/soc-dapm.c:1026)
> > > > 
> > > > These are definitions of deprecated interfaces.
> > > > We can remove it in 2.6.29.  If we don't want to be conservative, it
> > > > can be removed in 2.6.28, too.
> > > >...
> > > 
> > > Since it's an in-kernel API there's no reason to keep it once there are 
> > > no users left.
> > 
> > Right.  But, IMO, now is no suitable time.
> > A thing like API removal should have been tested in linux-next, and we
> > had plenty of time indeed for 2.6.28.
> >...
> 
> A grep through the tree and one test compile that covers 
> sound/soc/soc-dapm.c should be enough testing.
> 
> And having it then in -next once should be enough to discover if someone 
> wrongly added a new user.

My point is the time for removal.  The API changes should have been
done in the merge window, and it should have been tested *before* the
merge window.

> I have removed many functions in the kernel, and there isn't much that 
> can go wrong - even adding a PCI ID to a driver has a bigger risk of 
> introducing a regression.

Yeah, IMHO, adding PCI IDs blindly at the late stage should be
avoided, too, although many people love that.


thanks,

Takashi