Re: too much untested code in new kernels (fwd)

Andrew E. Mileski (
Sat, 4 Jan 1997 00:11:30 -0500 (EST)

> Let me share some opinions. Some of these are how I think things should
> be and some are how I think things are despite how they should be.
> 1) The primary purpose of kernel releases is to get the kernel distributed
> to a large number of people for testing.

Yes, I'd have to agree.

> 2) The purpose of releasing even numbered kernels is to test them for
> stability and robustsness.

AFAIK, their purpose is strictly to fix bugs. Ideally, there should
only be a single *.0.0 kernel release for example - of course the "rules"
do not apply to Linus :-) and he is free to change them at any time
(which I think he did for v2.0 - several thought it was started too soon,
but he stated he wanted a wider test audience).

> 3) The purpose of releasing odd numbered kernels is to test new features
> and keep everyone synchronized.

Yes, but I'd word it differently. AFAIK, their purpose is to introduce
new code (throw it against the wall and see if it sticks) and fix bugs.

> 4) Instead of saying that even numbered kernels are "production" kernels,
> we should encourage people who need stable production systems to use
> a distribution. The distribution maintainers generally do a good job
> of deciding which kernels are stable.

I have a fundamental bias against promoting the idea Linux = distribution,
as I see distributions as only a starting point. A distribution is
someone elses idea of a perfect Linux system, and this puts some people
in the mindset that "this is all Linux can do - Linux is crap."
There is a mind boggling amount of amazing software for Linux that is
not in any distribution - it couldn't be - which a distribution centric
view of Linux causes to be overlooked.

I'd have given up Linux long ago if I hadn't found it so darn useful,
and wasn't constantly suprised at the amazing things it can do using
FREE software (for the same functionality under Win95, I'd easily be
able to spend >$10k in software alone).

I started with Slackware 1.0. I customized and upgraded it right up
to v2.0.* time - it was fully modern. The system crashed quite badly
(my fault - experiment), and though I mostly recovered, it was never
100% cured. I grabbed Red Hat v3.0.3, and started over. I don't expect
to ever upgrade the distribution again, barring acts of God and bad luck,
or just plain stupidity.

Andrew E. Mileski
Linux Plug-and-Play Kernel Project
XFree86 Matrox Team