Slashdot Mirror
Video Interview With Linus On Linux 2.7
daria42 writes "ZDNet Australia has put up a video interview of Linux creator Linus Torvalds talking about the kernel development process, explaining why the unexpected resilience of kernel version 2.6 has delayed the move to 2.7." From the interview: "One of the original worries was that we would not be able to make big changes within the confines of the development model... I always said that if there is something so fundamental that everything will break then we will start at 2.7 at that point... We have been able to do fairly invasive things even while not actually destabilizing the kernel... Having stable and unstable in parallel: I think it used to be a great model, and I think we may see that the kernel has actually become more mature and stable and it just doesn't seem to be that great a model, for the kernel."
Visit the download page from a Linux browser and you can download Flash 9 for Linux now. And P.S. the beta was out for months before this was...
STOP . AMERICA . NOW
Flash 8 has been shown to destabilize the 2.6 kernel... Supposedly it will play nicely with the 2.7 kernel, though.
Having stable and unstable in parallel: I think it used to be a great model
It certainly works when dual-booting.
The theory of relativity doesn't work right in Arkansas.
He's basically saying that no one is really developing a 2.7 kernel because 2.6 is extremely stable even with whatever experimentation they've done. He states that theres been times where they've gone over the 2 month release cycle because of the "big changes" they've done on the kernel. He states that unstable next to stable used to be a good model but it isn't good anymore. He states that if there was a 2.7 kernel they'd have to do all sorts of backporting to get whatever fixes on the 2.7 kernel to work on the 2.6 kernel.
In my opinion, the real reason for no 2.7 is:
If we open up an unstable branch, I have less testers. --Linus Torvalds
I'm not saying the 2.6 series is unstable or anything, either. However as I watch Linux's development from the sidelines, I get the impression that most policy decisions Linus makes are designed to make his life easier. See also: Bitkeeper.
The ABI rules haven't changed at all: the user-kernel ABI (system-call interface) is supposed to be backwards compatible indefinitely; the internal ABI (e.g. for drivers) changes without warning whenever it's convenient.
What's changed is the release cycle--we no longer have this odd-numbered fork where the kernel's half-broken for years at a time.... Which is a good thing.
Maybe he uses the user agent switcher to work around broken websites, you insensitive... logged in user
Chernobyl 'not a wildlife haven' - BBC News
explaining why the unexpected resilience of kernel version 2.6 has delayed the move to 2.7.
Uh...resilience?
2.6 releases have "shipped" numerous times with some serious bugs, probably because Linus and company have let lots of people slip major new features into the 2.6 kernel, when it's supposed to be stable. 2.6 kernels regularly make it SEVERAL "point" releases into each point release:
Go and look at the timestamps on 'em on ftp.kernel.org. Some of the sub-versions are just a few days apart. How the hell are end-users supposed to know when the kernel is ACTUALLY useable, if there are THIRTY SEVEN bug-fix releases?
One of the more amazing bugs involved a bug in md that would hose raid partitions, and I assure you, it was not the only serious filesystem bug. I lost a reiserfs partition thanks to a half-baked 2.6 release.
Please help metamoderate.
can anyone post this "fragmented" and unaccessible interview video to youtube or google video as one or two big file(s)?
That scheme ended when 2.6 came out. The new system consists of 3 or 4 numbers formatted as:
a.b.c
or
a.b.c.d
a changes only when there is a massive restructuring of the kernel
b changes when there are large sweeping changes, but not of quite the same order as a. (linus, in the interview, says they'll do a 2.7 when and if they need to make changes large enough that they will be breaking everything.)
c changes when new features and/or drivers are added
d changes for small bug fixes and security patches. after a new c release the d number is ommitted when the c number has just changed.
No single raindrop believes that it is responsible for the storm.
Add this line: FIREFOX_DSP="aoss" (remove FIREFOX_DSP=)
Install the alsa-oss package.
Restart FF, and you are playing sound! Um, I'm not sure what planet you are living on, but that's not "easy". It's tedious and frustrating.
Lesbian Nazi Hookers Abducted by UFOs and Forced Into Weight Loss Programs - -all next week on Town Talk.
VFS probably needs to be addressed. Reiserfs4 sort of exposed some of the issues. There are others though. To my knowledge ext2/3 are the only OSes that actually code strcitly against VFS and the other layers. XFS, JFS, Reiserfs, etc.. are all hacked in to it. If you follow the kernel list, you'll see nobody uses JFS and XFS seems to have regular crash reports. Upon using it myself (for 5 years) it has memory leaks, it routinely has trouble with new kernels. There have been regular performance regressions. Now, I don't really care about the filesystem itself that much but it seems fundamentally broken to me that a non-experimental filesystem has such routine problems. Either the API is uses is broken, the filesystem is broken or both. I'm becoming more inclined to think that it's VFS. This creates a circular sort of problem, you don't need VFS if ext2,3,4 are the only filesystems that are really supported, it's not nearly as important as it is treated. Either that or the process of having something included and non-experimental needs to include some kind of support aspect and maybe be rethought. So far as I can tell, IBM isn't really doing much more with JFS and nobody uses it, let's move to remove it (bummer too because it's a quite clean and elegant FS, much better than reiserfs or xfs in terms of code and design quality and cleanness.) There isn't a clean process for removing stuff from the kernel. Reiserfs is a prime example, Reiserfs3 isn't supported, time to move to remove it; it has known bugs and design flaws which are not being addressed. This particular area is more complex also because selinux depends on filesystem support, LVM behaves differently with different filesystems, different filesystems have different and variable tools support.. System filesystems need some work too, what's debugfs? configfs? How is sysfs different than configfs or procfs?
Filesystems are just an easy to see and expose portion of this problem there are other APIs which have the same issues. We retooled the build system a few years back, it's much better but there are major flaws still. There are drivers which cannot work unless loaded as a module and yet they can be linked in. There are a huge number that depend on other subsystems and you can easily misconfigure them (SATA depends on parts of SCSI. So I can static link some SATA modules in and dynamic link parts of the SCSI system in and the build system won't complain. Worse, if I break it just so, I can actually get it to build cleanly and freak out at runtime) I'm not advocating making it more difficult to hack on the kernel or add new modules to the build but it's fucked if it doens't catch that stuff. Worse, the driver is fucked if it can't be statically linked and if that's an acceptable limitation then it should be an option. (the Fusion series of RAID/SCSI/SAS type drivers is one that suffers from this problem) At the same time the build system is holy, good luck changing that without pissing off half the free world, and I don't even want to think about what would have to happen if it required a change to a .config file to take it to the next level. Part of the beauty of Linux in this regard is that it is remarkably simple to build and get involved with, there really aren't any tricks or anything to building it. This is something else where there needs to be a support component. There are good companies with well supported drivers and there are orphans. I'd rather have modules marked as supported or unsupported than whether or not they are GPL clean or tainted, I'd like to see that