Re: kernel structures 2.0.29->2.0.30

Chel van Gennip (
Sun, 27 Apr 1997 1:04:04 +0100 (WETDST)

I try to convince management to accept Linux as a platform for our
products. To do so a version that is stable for a long time would be
nice. However I know this is impossible. Hardware vendors do change
product specifications often (adaptec 2940, 3Com 3c905, etc.)
Furthermore new functionality is needed IPV6, fixes for SYN attacs ...
So new versions will appear, not all of them being major releases.
The stability of these new versions will and can not be gradually
increasing. That's the world we live in.

Alan Cox wrote
> major change. From 2.0 to 2.1 a smaller change and from 2.0.29 to 2.0.30
>> just some bug fixes. Otherwise, we are not better than Microsoft which
>> uses versioning for marketing.
>One problem we have is no version space for an "incremental upgrade".
>Solaris for example goes 2.2,2.3,2.4,2.5 and then has the 2.5.1 release
>which is 2.5 patched and improved a bit but not a major change. 2.0.30
>is the equivalent of that

Such a numbering system is too complex for the problem I think. The current
numbering system is fine, if you use it with a sort of stability index.
I use the date and length of the patchfiles as a stability index.

>> more stable and more careful with changes in *stable* versions. Or, since
>> Linux is only supported by people who believe in it, not by managers who
>> dictate its use, we will lost supporters.
>You'll notice vendors don't leap onto new subrevisions already for exactly
>this reason. Red Hat 4.0 shipped a fixed 2.0.18 after the big ping bug
>rather than a 2.0.25.

That's the way we do it to, we jumped from 1.2.13 -> 1.3.59 -> 2.0.14 -> 2.0.28.
The jump to 1.3.59 was for functionality, the jump to 2.0.14 to be in line with
the stable kernel version and to support new hardware, the jump to 2.0.28 again
for new hardware. The strategy is: don't change unless you must.
Fo us this is not a problem at all, we deliver complete boxes (hardware, OS, application)
To convince some software vendors I think specific versions should be certified as
"rock solid".