kernel numbering scheme (stable or not)

Jacques Gelinas (jack@solucorp.qc.ca)
Fri, 25 Apr 1997 11:03:33 -0400 (EDT)

The problem with 2.0 (or any "stable" release) is that until you declare
something to be stable, not enough people will jump on it. If you tell
them that this is 1.3.100, or even pre-2.0 as was tried by Linus, you get
a limited impact and a limited amount of testers (Which probably outnumber
any beta test program of any OS producer, by far).

I have followed linux kernel almost release per release since 1.3.x.
Generally, I installed these releases on all my computers. I would say
that all release have worked correctly for me. This is the problem. 2.0
was known to be stable and was used by numerous developpers with success.
But it was not perfect and this was shown early when a larger audience
jumped on it. So 2.0.1 was done.

Now what is the difference between 2.0.1 vs 2.0.0 and 1.3.90 vs 1.3.89.
While one will say that the change should be minimal (no new features
allowed), we knows that fixing a single bug may change quite a few things,
even potentially making the machine non bootable for some. It has been
seen. Some people were not even able to boot some 2.0.x release.

While you can always release test version (as done with 2.0.29-ISS), you
never get the impact of a normal 2.0.x release, so many bugs are not
discovered. And this is only when you will release your test version to
the public (official 2.0.x release) that some will complain that the
"supposed to be stable" kernel does not even boot on their (broken ? :-) )
system.

Now what is a stable 2.0.x release. There are peoples who monitor the net
and create distribution. They will generally ship what they believe is the
stablest one. This is the purpose of a distribution. Think about them as
brokers.

You can invent any numbering scheme you want, but many bugs will be seen
only when you will release it to a wider audience with the "stable"
qualifier on it.

The current strategy on 2.0.x is the correct one. 2.2 will be much later
this time. 2.0 must have a life. Now, you can renumber it to 2.0.30beta
(which is not appropriate, NT 4.0 IS a beta) and then loose much of the
audience (They will wait for official 2.0.30).

Those who track the latest and greatest can monitor the net for a week or
two before jumping on a new kernel. Further it is so easy to move back and
forth with many kernel on the same machine. Those who can't afford this
monitoring can always stick with distribution and their upgrade packages.
They will have a working system.

--------------------------------------------------------
Jacques Gelinas (jacques@solucorp.qc.ca)
Linuxconf: The ultimate administration system for Linux.
see http://www.solucorp.qc.ca/linuxconf
new development: linuxconf configure dhcpd since 1.9r7