| |||||||||||||
The embedded Linux nightmare - an epilogueThe embedded Linux nightmare - an epiloguePosted May 2, 2007 10:14 UTC (Wed) by NAR (subscriber, #1313)Parent article: The embedded Linux nightmare - an epilogue During another discussion about backporting a large feature into an old kernel, I asked why a company would want to do that. I guess they want the new feature (e.g. wireless stack), but don't want the new bugs in e.g. the memory management, scheduler or block layer. And please don't tell me there are no new bugs.
(Log in to post comments)
The embedded Linux nightmare - an epilogue Posted May 2, 2007 13:55 UTC (Wed) by jreiser (subscriber, #11027) [Link] they want the new feature... but don't want the new bugsThis is exactly what matters for an embedded product, and exactly where the Linux development process fails. Last week Adrian Bunk quit in disgust at the regressions http://lwn.net/Articles/231993/ and Andrew Morton said the situaition is likely to get worse http://lwn.net/Articles/232432/. Consumers do not tolerate such shoddy products.
The embedded Linux nightmare - an epilogue Posted May 2, 2007 16:26 UTC (Wed) by khim (subscriber, #9252) [Link] This is exactly what matters for an embedded product, and exactly where the Linux development process fails. Not really. If you'll try to backport new features to ancient kernel you'll get new features, new bugs (in subsystems which must be modified to accommodate new features) and your own unique bugs (because you'll have mostly-untested mix of old and new kernels). Consumers do not tolerate such shoddy products. Sure they do. If you'll take line of mobile phones from any manufacturer and count number of bugs and regressions - you'll probably find more bugs/regressions then 14. And most of them are known. And some are even fixed in unofficial firmware. This is waay worse then what happens with kernel - yet Nokia and Samsung are selling millions of phones...
The embedded Linux nightmare - an epilogue Posted May 2, 2007 14:50 UTC (Wed) by bene42 (subscriber, #32623) [Link] On the other hand they don't like old bugs. So you have to choose. And it is far more easier to get help from nearly everybody to fix a new bug, then backport and maintain a bunch of patches to fix old bugs.
The embedded Linux nightmare - an epilogue Posted May 2, 2007 15:48 UTC (Wed) by NAR (subscriber, #1313) [Link] On the other hand they don't like old bugs.Yes. But they spent half year on testing their stuff with the old kernel with the workloads they expect, so they can be pretty sure that even if there are old bugs, these bugs don't cause problems. Or if they do, they can be fixed (even by backporting from newer versions) or by working around them.
The embedded Linux nightmare - an epilogue Posted May 2, 2007 16:34 UTC (Wed) by bene42 (subscriber, #32623) [Link] Where is the difference?The have to spend a complete QA cycle in any case. Every shortcut here is window-dressing. Or lets say it in other words: Every shortcut is the programmers guaranteed right to fire up some viewy firework... see 6/4/1996
The embedded Linux nightmare - an epilogue Posted May 2, 2007 15:51 UTC (Wed) by tglx (subscriber, #31301) [Link] > I guess they want the new feature (e.g. wireless stack), but don't want the new bugs in e.g. the memory management, scheduler or block layer. And please don't tell me there are no new bugs.
Right, and therefor they backport the wireless stack, which includes the modifications to the network stack on which it relies. Now they need to fixup the memory managment changes in both and some other nasty details all over the place.
This is done by a very small group of people. Do they have the full expertise and overview on all this and the testing and verification capacities at hand to verify this hackery on their own ?
If yes, then they should really go and identify and fix the bugs in mainline. They would be a great asset to the kernel community.
If no, the nightmare is there in it's full glory.
The embedded Linux nightmare - an epilogue Posted May 2, 2007 16:02 UTC (Wed) by NAR (subscriber, #1313) [Link] They are probably a little naive and think that e.g. the wireless network stack and the block layer are independent of each other and there are well defined and stable internal APIs within the kernel, so it doesn't get redesigned in every two months.
|
|||||||||||||