This is really a silly discussion:
2.6.x release is the initial base for each stable release
!with absolutely no intention of stalling non-useland API's!
2.6.x.y releases are the stability updates to the base release
and in kernel API's will usually stay stable
If you want an ultra stable kernel you should start with the last stable
release and start tracking what you consider critical fixes from the
next base kernel (2.6.x) forward (essentially creating your own vendor
branch). Alternatively you should use a vendor branch which already
does this for you with the addition of back porting important new device
drivers (Debian Stable, RHEL, SLES, ...).
The 2.6 development model finally gives developers the ability to dump
cruft, fix broken architecture, and add performance enhancements in a
timely manor. Linux development hasn't worked this well since 1.2 was
small enough to test and release quickly.
OTOH it would be nice if core userland (libc, udev, binutils,
shellutils) were managed as a single project (as with OpenBSD) so that
userland breakage would be better managed. :-)
On Sunday 04 December 2005 10:24, Jonathan A. George wrote:
> OTOH it would be nice if core userland (libc, udev, binutils,
> shellutils) were managed as a single project (as with OpenBSD) so that
> userland breakage would be better managed. :-)
Well, there's always the combination of busybox and uClibc. :)
But we don't do gcc, binutils, and make. (There's tcc, but it doesn't quite
build the unmodified kernel yet, doesn't do make, and its optimizer still
sucks pretty badly...)
Rob
--
Steve Ballmer: Innovation! Inigo Montoya: You keep using that word.
I do not think it means what you think it means.