It was recently asked on the Linux Kernel Mailing List why the 2.6.0 kernel [forum] was planned for release so soon, when there are still a number of known issues [story] still needing to be fixed. Larry McVoy [interview] replied, "The 'stable' series of the kernel is never really stable for a while. A better way to think of it is as 'that place where things become stable by refusing to take any new changes except bug fixes'." He goes on to explain:
"The news media hasn't picked up on this yet, they seem to think that 2.6.0 is something that will be useful. It won't be, there will be a period of months during which things stablize and then you'll see the distros pick up the release."
From: Tim Cambrant [email blocked] To: Linux Kernel Mailing List [email blocked] Subject: Too soon for stable release? Date: Sat, 29 Nov 2003 18:49:16 +0100 I am sorry if this offends someone or if I'm totally on the wrong track here, but it seems odd to actually call the Beaver On Detox "stable", considering the amount of misc. problems people have been having the last week with -test11. Since you are in schedule, a -test12 with the bugs fixed might be a better idea, don't you think? Since I'm not a coder, and havn't had any previous kernel development experince, I might not be the one to suggest such a thing, but I really don't see a reason to be hasty with the release, and end up with a whole bunch of users having problems. The weeks Linus suggested we would wait before releasing the final 2.6.0-kernel could be a perfect time to implement the fixes that would be needed. Test11 is working perfectly for me, but you can clearly see that many people are having problems. -- Tim Cambrant [email blocked] GPG KeyID 0x59518702 Fingerprint: 14FE 03AE C2D1 072A 87D0 BC4D FA9E 02D8 5951 8702
From: Larry McVoy [email blocked] Subject: Re: Too soon for stable release? Date: Sat, 29 Nov 2003 09:01:04 -0800 On Sat, Nov 29, 2003 at 06:49:16PM +0100, Tim Cambrant wrote: > I am sorry if this offends someone or if I'm totally on the wrong track > here, but it seems odd to actually call the Beaver On Detox "stable", > considering the amount of misc. problems people have been having the > last week with -test11. The "stable" series of the kernel is never really stable for a while. A better way to think of it is as "that place where things become stable by refusing to take any new changes except bug fixes". The news media hasn't picked up on this yet, they seem to think that 2.6.0 is something that will be useful. It won't be, there will be a period of months during which things stablize and then you'll see the distros pick up the release. I don't remember where it was exactly (2.4.18?) but Red Hat waited quite a while before switching to 2.4 from 2.2. This is normal and it works out quite well in practice. -- --- Larry McVoy lm at bitmover.com http://www.bitmover.com/lm
From: Russell King [email blocked] Subject: Re: Too soon for stable release? Date: Sat, 29 Nov 2003 17:11:11 +0000 On Sat, Nov 29, 2003 at 09:01:04AM -0800, Larry McVoy wrote: > The news media hasn't picked up on this yet, they seem to think that > 2.6.0 is something that will be useful. It won't be, there will be a > period of months during which things stablize and then you'll see the > distros pick up the release. I don't remember where it was exactly > (2.4.18?) but Red Hat waited quite a while before switching to 2.4 > from 2.2. This is normal and it works out quite well in practice. Red Hat did a 2.4.2 release which was 2.4.2 + a lot of stability changes. IIRC, RH7.x was based on 2.4.7, with updates to 2.4.9, 2.4.18 and finally 2.4.20-based kernels. However, I also seem to remember each of these had a fair number of patches applied. I'm sure Arjan will correct me if I got the above wrong. -- Russell King Linux kernel 2.6 ARM Linux - http://www.arm.linux.org.uk/ maintainer of: 2.6 PCMCIA - http://pcmcia.arm.linux.org.uk/ 2.6 Serial core
From: Arjan van de Ven [email blocked] Subject: Re: Too soon for stable release? Date: Sat, 29 Nov 2003 18:42:48 +0100 On Sat, 2003-11-29 at 18:11, Russell King wrote: > Red Hat did a 2.4.2 release which was 2.4.2 + a lot of stability changes. which was basically a 2.4.4-pre > IIRC, RH7.2 was based on 2.4.7, 2.4.7 lived for half a day but the VM of 2.4.7 was so bad we had to go to 2.4.9 immediately..
i'll not use 2.6 for long time
I will not use 2.6 kernel until the matroxfb will be fixed.
Does java works in the 2.6 kernel? I have debian sid and i cannot see java applet.If i rebot with the stable kernel, everything is ok.
Matrox rulez!
java in 2.6
Yes, java works in 2.6, i'm developing java aplications with eclipse on a 2.6 kernel.
However, i'm using linuxthreads, not the new: nptl (new posix threading library) implementation of posix threads that's in the CVS glibc.
Everything works quite well.. i'm amazed with the improvements in 2.6 :D
is really worth using it.
Miguel
matroxfb
I have to agree on that. Matroxfb doesn't work for me; i'm running 2.6.0. I'm not going to use the 2.6 branch until matroxfb is fixed.
When Will 2.6 Be Truly Stable?
... when it's ready
I am not able to use 2.6.0, because moanoe going to fix this bug
http://bugzilla.kernel.org/show_bug.cgi?id=1502
Really?
My question was that interesting? Perhaps the answers were the really interesting stuff, but nevertheless... I think Larry and Russel cleared up most of the doubts I had. Perhaps the fact that an x.x.0-kernel isn't actually stable should be included in some FAQ or something?
stable is relative
Two reasons, not to put this information in the faq. You can consider them to be invalid, but they have some truth in them:
Typical end users may have different experience, but they also have different machines and different needs, so they catch different bugs.
What Is Stable, Anyway?
One could argue that for the great majority of users even the "not quite stable" .0 release is at least as stable as any release of Windows has *ever* been. I don't bring this up to argue that we should accept parity with Windows as good enough, only to emphasize that, as previously stated, stability is relative. :-)
But seriously, you have to think of stability is a goal, not a condition. 100% stability is impossible. You'll never reach it and the best you can do is approach it asymptotically. Linus' release policy is evidence that he understands this very well.
You will have noticed that in the release notes for the past couple of test releases Linus expressed a desire to see the number of patches coming in trickle down to (almost) nothing. He has said that seeing the number of patches go down is the factor that will determine the timing of the .0 release.
When the number of bugfix patches drops low, stability is no longer approaching perfection at the rate we'd all like to see. The only way to get more bugfixes to come in (i.e., to move again toward perfect stability) is to widen the userbase, thereby widening the testing activities. And making a .0 release is the best way to achieve that.
And don't worry about the lack of a warning: anyone who admins a production box and is able to download a stock kernel and compile it is virtually guaranteed to know exactly what .0 means in a Linux kernel. People who get their kernels from a distribution will not see a 2.6 kernel for awhile.
David Merrill
WorksForMe(tm)
Right now, i'm using a 2.6.0-test9 kernel. It's pretty stable, althought some stuff (like sane) dislikes it... I'm runnning it on the top of a Slackware 9.0. Uptime below:
$ uptime
19:34:03 up 16 days, 21:32, 11 users, load average: 0.58, 0.38, 0.25
The machine runs the usual desktop tasks (mozilla, xmms (w/ALSA drivers), XFree, openbox, irssi) - plus MLdonkey, CD-R writting and NFS.
--
That's it.
True
The current test-kernel is very stable, and I havn't had any problems with it so far. What I meant is that it seems foolish to release a new "stable" version of the kernel without fixing the bugs that we know about. Since you all clarified that for me, I agree, and it seems absolutely resonable to release the kernel to allow (trick?) more people into testing it. Thanks for the explanations though. You're all very helpful :)
ide-scsi
Hi all,
Will 2.6 final eventually have ide-scsi? I'm using -test11 from http://people.redhat.com/arjanv/2.5/ and i'm wondering if arjanv will pack the next rpm with ide-scsi. I've been using /dev/hdc as mentioned in the README but still i've problem getting my cdwriter/external floppy/digi cam to work. 2.6 really does improve my laptop performance and i'm thinking a way to fix all of this so i can stick to my laptop.