[linux-elitists] New Kernel Dev model...

Greg KH greg@kroah.com
Thu Jul 22 19:43:59 PDT 2004


On Fri, Jul 23, 2004 at 10:49:34AM +1000, Mike MacCana wrote:
> On Thu, 2004-07-22 at 13:36 -0400, Greg Folkert wrote:
> > I have mixed feelings about this... leaving the stabilizations to the
> > Vendors/Distros... It was already happening... but why make it defacto.
> 
> As you say, this just codifies existing practice. Mainline 2.4 wasn't
> stable (think VM change), 2.6 won't be either. 

Late 2.5, and all of 2.6 development is done radically different than
2.4 was.  Basically, we are going to continue to do nothing differently,
other than pointing out to the world, that we are operating quite
differently than people really think we are.

I'm going to try to hijack my OLS talk tomorrow (krefs, bah, how boring)
to talk about this in more detail, given the apparent lack of
understanding.  I'll post a link here to my presentation when I write it
up (I still have a few hours...)

> In both cases, distros were responsible for handing stability. This
> keeps that, but without kernel.org confusing people by telling them X
> means stable when it doesn't.

Stable is in the eye of the user.  2.5 was incredibly stable for me,
depending on what task I wanted to do at the time :)  It's all relative,
and perception.  Bah, I just described marketing, someone shoot me...

> 
> > I like development models that are Dev->Test->QA->RC->Release for
> > Production builds.
> 
> Me too - but distros are in a better position to do that testing than
> kernel.org. Think tens of thousands of people running 2.x.y with the
> same build options.

A certain 3 letter company just put the next 2.6 enterprise Linux kernel
distro release through more testing the past few months than it ever has
had in the past.  I know this based on the bugs that were found, which
had been there since at least 2.4.0.  Hopefully we can run that test on
every release, and a smaller subset on every nightly snapshot.

I know another big company which has recently acquired a few Linux
companies, that also has an incredible test department due to their
prior product offering, that is now starting to be revamped over to test
Linux.

Combine that with the small thousands of people who build their own
kernels from kernel.org when every new release happens, with their odder
build options, and we have the start of a really giant regression test
suite that will ensure that we don't break anything major, and not
realize it quite quickly.

> > This WILL slow the adoption of Linux more than SCO did... those weenie CXOs are the
> > ones trembling when a development model changes like this. 
> 
> CIOs (who, like developers, may or may not be weenies) never used
> kernel.org kernels. Most people don't - AFAIK, Debian is the only
> mainstream distro that ships anything like a vanilla kernel.

Hahahaha.  That Debian kernel is full of patches.  Slackware is the
closest I've seen, and Gentoo now only has 4 things[1] added to their
kernel (not including bug fixes that are already in the current tree,
that will be thrown away with the next release.)  And as I'm the
maintainer of that package for them, I'm trying to get that down to 0.

thanks,

greg k-h

[1] For those interested they are:
	- bootsplash
	- speakup
	- squashfs
	- supermount
    and I think I'll drop supermount next to see who screams...



More information about the linux-elitists mailing list