Re: [Patch] eliminate CLONE_* duplications

From: Herbert Poetzl <herbert@13thfloor.at>
To: Roman Zippel <zippel@linux-m68k.org>
Cc: Andrew Morton <akpm@osdl.org>,
	Linux Kernel ML <linux-kernel@vger.kernel.org>
Subject: Re: [Patch] eliminate CLONE_* duplications
Date: Tue, 27 Sep 2005 18:22:42 +0200	[thread overview]
Message-ID: <20050927162242.GC21927@MAIL.13thfloor.at> (raw)
In-Reply-To: <Pine.LNX.4.61.0509271705380.3728@scrub.home>

On Tue, Sep 27, 2005 at 05:25:41PM +0200, Roman Zippel wrote:
> Hi,
> 
> (Sorry for the delay.)

no big deal ...

> On Thu, 22 Sep 2005, Herbert Poetzl wrote:
> 
> > _what_ do you consider 'logically organized' because
> > putting all the CLONE_* stuff into a separate file is
> > pretty logical for me ... but obviously not for you.
> 
> "logically organized" mainly means reducing dependencies by organizing
> them by their logical dependencies. 

did you consider that separating out the clone
stuff might be that basis for reducing dependencies?

> If a large header file is included by a lot of other files, some parts
> maybe separated to reduce header dependencies. The same can be done
> for config dependencies, so that a config change doesn't necessarily
> recompiles the whole kernel.

> Your change doesn't reduce any dependecies and it's not such a big 
> cleanup:

aha, so we want a big all-in-one patch now which
does many changes at once, instead of a small
(and obvious) cleanup first, then later maybe
a restructuring ... yes?

>  arch/alpha/kernel/asm-offsets.c         |    2 --
>  arch/alpha/kernel/entry.S               |    1 +
>  arch/cris/arch-v10/kernel/asm-offsets.c |    3 ---
>  arch/cris/arch-v10/kernel/entry.S       |    1 +
>  arch/cris/arch-v32/kernel/asm-offsets.c |    3 ---
>  arch/frv/kernel/kernel_thread.S         |    2 +-
>  arch/ia64/ia32/ia32_entry.S             |    3 ++-
>  arch/ia64/kernel/asm-offsets.c          |    4 ----
>  arch/parisc/kernel/entry.S              |    4 +---
>  arch/ppc/kernel/asm-offsets.c           |    2 --
>  arch/ppc/kernel/misc.S                  |    1 +
>  arch/ppc64/kernel/asm-offsets.c         |    3 ---
>  arch/ppc64/kernel/misc.S                |    1 +
>  arch/v850/kernel/asm-offsets.c          |    4 ----
>  arch/v850/kernel/entry.S                |    1 +
>  include/asm-cris/arch-v10/offset.h      |    3 ---
>  include/asm-cris/arch-v32/offset.h      |    3 ---
>  include/linux/clone.h                   |   32 ++++++++++++++++++++++++++++++++
>  include/linux/sched.h                   |   30 +-----------------------------
>  19 files changed, 42 insertions(+), 61 deletions(-)

> The noise generated by the separation is larger than the avoided 
> duplication.

hmm, interesting argument ...

> The hardcoded defines actually do need fixing, frv is especially bad,
> as it even has hardcoded structure offsets.

so instead of fixing the issue properly, we 
'mend' it by adding new code to */asm-offsets.c

> > I have absolutely no problem with different, more
> > logical splitups, and I'm willing to break down the
> > entire sched.h if that will help the cause ... so
> > please enlighten me here ...

> sched.h is especially challenging due to dependencies between headers
> under asm and linux. It's not just splitting sched.h, it also requires
> analyzing its dependencies.

which you obviously think is nothing I can do
'properly' ...

thanks for the info,
Herbert

> bye, Roman