Am Samstag, 21. Juni 2008 schrieb Greg KH:
On Fri, Jun 20, 2008 at 03:51:46PM +0200, Marcus Meissner wrote:
I believe it is 'too dangerous'. We tested 2.6.25 in betas and rcs, while 2.6.26 would be just dumped to users.
With some testing first of course :)
No way. Thats what the BETA and RC phase was for, we cannot do this for the released product.
So, you are saying that the kernel developer community doesn't test their kernel releases enough for you? :)
Greg, what Marcus is saying here, is that average kernel developers are bad in tracking userspace changes. They throw the new kernels on their systems and watch what broke, fixing it silently or not not at all, if that subsystem isn't too important for them. They're used to break some functionality from their systems on a regular basis. And you would only note the breakage, if the affected subsystem is essential for them AND no fix/workaround is available somewhere in the deep space of OS. On the contrary, take the average user of a high level distro like openSUSE. If things break on their system out of the deep blue, they usually don't have any idea, what broke - is it the kernel, userspace, some config magic badly chiming in, what ever, they're unable to locate the breakage other then - my wlan doesn't work anymore, bluetooth headset dysfunctional, no UI (aka X) and so on. While I occasionally want newer kernels for certain reasons, the switch is often painful - e.g. my notebook had a iwl4965 based wlan card with openSUSE 10.2, and I've upgraded the kernel to 2.6.24.4. I was unable to get wlan to work again in an acceptable timeframe (was on holiday, didn't tested that subsystem after upgrade). YaST didn't cope with the new infrastructure, but adjusting stuff manually was just too painful - starting from firmware, it would have included to upgrade a lot of userspace - with references back to kernel space (madwifi), that was where I lost interest.. (aka pain was stronger than the outcome). The problem, I see here, is even the _released_ product contains an awful amount of glitches still - and fixing them is unpossible, if you change too many values in the equation at once..
What would you like the kernel developers to do in order for you to feel better about doing exactly what Fedora does here (remember, they have more users and they do this all the time...)
I fear, that the kernel developers aren't able to do anything in this respect, unless changesets from release to release shrink _immensely_. It would take a tremendous effort of something like a test center, with lots of human and hardware resources to test all the permutations that are possible then in loads of different environments (and no, kernel compile testing of random configs is not enough). Note, it is a general difference in using a computer system from day to day in the usual kernel developer way, which exercises mostly the main paths of IO, standard FS, MM, etc compared to those systems running in the wild, used creatively with always changing environments and hardware.
OTOH, having easy way of installing 2.6.26 _in addition_ to 2.6.25 kernel would be nice ...
I guess adding 2.6.26 into boot menu as an non-default option by update would be acceptable compromise?
That sounds fine to me.
But this will be very difficult to do and error prone. :(
Why? It shouldn't be that hard to tell the bootloader to not mark this kernel as the main one to install.
To complete the already too long message, may I remember you, that me as a single user suffer from major problems, a few days after rollout of 11.0 to 3 selected systems here - regressions to things used to work before: - Bug 401119 - ntp doesn't work with serial DCF receiver - Bug 401730 - gphoto2 capture causes usb disconnect and subsequently a kernel crash and apart from Bug 401259 - libzypp duplicates gpg keys in the database, this distribution feels great, so far! But I didn't upgrade my main system, yet. Imagine, what happens, if I would rollout that distribution to say 50 systems - all with different hardware, capabilities, and usage profiles. That's impossible at the moment, since daytime is fixed, and I still have to support my family, too. And now, you try to throw in another kernel - which I do understand form a certain perspective, but I fear, I'm unable to ever get into a stable state here (which I need, since the operation system is the basis of my day to day work, not the work itself, at least most of my time). Pete -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org