[2493] in SIPB_Linux_Development
Re: Kernel interface changes (was Re: cdrecord problems on
daemon@ATHENA.MIT.EDU (Khimenko Victor)
Thu Feb 4 19:52:58 1999
Date: Fri, 5 Feb 1999 03:50:38 +0300 (EET)
From: Khimenko Victor <khim@sch57.msk.ru>
To: Arvind Sankar <arvinds@MIT.EDU>
Cc: xiphmont@MIT.EDU, alan@lxorguk.ukuu.org.uk, warlord@MIT.EDU,
linux-kernel@vger.rutgers.edu, linux-dev@MIT.EDU, jered@MIT.EDU,
nemo@MIT.EDU, cox@idecnet.com
In-Reply-To: <19990204164832.B15769@anjala.mit.edu>
On Thu, 4 Feb 1999, Arvind Sankar wrote:
> On Thu, Feb 04, 1999 at 10:17:54PM +0300, Khimenko Victor wrote:
> > In <199902041856.NAA00105@cutter-john.MIT.EDU> Monty (xiphmont@mit.edu) wrote:
> >
> > M> No more cc:s to cdwrite or linux-scsi, please.
> >
> > >>> 2.2 is supposed to _be_ stable, not gradually stabilize. That's what
> > >>> 2.1/2.3 are for.
> > >>
> > >>Unfortunatelly it's not possible. Joe average will not even try "development"
> > >>kernel and some errors could not be found without A LOT OF users.
> >
> > M> So fix the bugs. Don't try to roll in new features or optimizations.
> > M> One specific trouble I was talking about happened when 2.1 lasted long
> > M> enough that developers decided to roll 2.1 features back into 2.0. Of
> > M> *course* you're going to miss something!
> >
> > You could not get 100% compatibility with ANY change (even if you'll just fix
> > bug this will probably broke someone program! that's why quite a few well-known
> > bugs in Windows could not be fixed)...
>
> I assume this is a joke...
>
Unfortunatelly not. I'm could kill ANY WindowsNT system without Admin or
even Power User access just by doing some nasty things with DDE. Why it's
not fixed ? "Oh, this will break so many programs..."
> > As fast as this does not affect userspace API it's not a problem to me. Source
> > compatibility is good but binary-compatibility... Even if binary-only modules
> > was mistakenly allowed bu Linus in first place it's not excuse to make
> > developers of them happy :-)
>
> Look, it's not just a question of binary-only modules. Even if you have source in
> a distributed environment, you still have the problem of rebuilding 1000's of client
> machines (and, at least in one case, just because someone decided to shave off a few
> microseconds with better cache alignment). This doesn't make sense.
Why you must rebuild 1000's of client machines ? You should compile kernel
anyway to make upgrade -- why not compile all needed modules as well ????
