please unblock kernel-package 11.005
Hi,
Could you please consider unblocking kernel-package, perhaps
after it spends some time in Sid? The changes are quite extensive, but
the fact is that kernel-package was in really bad shape (I was not
kidding about the RC bug), and thus much work was required to bring it
up to release quality. I am sorry it happened so late into the release
cycle.
Why should we want to have this new version in Lenny?
The new version works with the merged x86 code in recent
kernels, while retaining compatibility with older kernel sources. It
correctly generates the right set of headers. It is again
cross-compilation friendly. The postinst no longer runs lilo when it
thinks there is no other bootloader (it used to detect grub, but not
grub2). It correctly installs firmware in a versioned location under
/lib/firmware.
More significantly, the build system has moved to a more
streamlined, make -j friendly build system While I am not sure of this
fixes some of the nagging problems we have been facing in recent
versions of kernel-package, where we used double colon rules, which
were convenient, sure, but played havoc with ordering of the rules, and
had to have various band-aids to help out with the ordering. The system
was rapidly growing complex, with clear indication that it was actually
faster.
The new target mechanism does away with doublecolon rules, and
should play better with parrallel compilation. We try to use upstream
kbuild as far as possible, to reduce churn as the files upstream
installs change. Some added checks of the Makefile are now in place so
we retain backwards compatibility. This should improve things lot wrt
header files. We also now add dependencies to more packages actually
required to build kernel images.
We also try to look for the kbuild created KERNELRELEASE
variable, which is designed to be used by distros to figure out where
modules are to be loaded from, etc. This should help reduce version
mismatches. We also prepare the kernel.release file early, to help
that.
We also refitted to support the new XEN code in mainstream, in
that the same image can be booted normally or be used as a XEN
image. This support probably needs to be improved.
The make target dependencies have been extensively reworked, to
minimize surprises and wasted effort. We also strip modules, based on
DEB_BUILD_OPTIONS (nostrip).
Extra care is now taken so we do not accidentally remove
./debian while cleaning, thanks to upstream helpfully removing ./debian
when cleaning. This should prevent dpkg-buildpackage from accidentally
shooting itself in the foot by removing ./debian as its first action.
Finally, the changes have made it possible to create a
kernel-image straight out of a git working directory, partially because
the upstream script does not think that the changes kernel-package
makes to the source make it dirty, and partially because we run the
kernel.release creation script early, just after patching the
sources, but before generating the ./debian/changelog, and this,
abetted by using KERNELRELEASE, ensures that we correctly capture the
version.
I have also added dependencies to kernel package, the kerel
source package, the kernel header package, with the basic tools
required to build a kernel, so by installing the source package, or the
header package, the user should have most of the things required to
compile their own kernel.
changelog follows.
manoj
kernel-package (11.005) unstable; urgency=critical
* [32c42dc] Fix a cut and paste typo which substituted a XEN postinst
This caused initrds not to be created. I knew there was a downside to
code hacking all-nighters.
Bug fix: "linux-image*.postinst does not create an initrd", thanks to
Sven Joachim (Closes: #501793).
-- Manoj Srivastava <srivasta@debian.org> Fri, 10 Oct 2008 09:18:38 -0500
kernel-package (11.004) unstable; urgency=high
* [c68891f] A new bug fix version )closes serious bug).
This is a pure documentation change, it removes language that might
mislead people into thinking that there is any relation between
kernel-package and official kernels. (Closes: #370651)
* [680cbd4] Upstream fixed the problem of thinking the sources were dirty
* Bug fix: "Bogus version number when building from Git", thanks to Josh
Triplett (Closes: #423721).
-- Manoj Srivastava <srivasta@debian.org> Fri, 10 Oct 2008 02:47:49 -0500
kernel-package (11.003) unstable; urgency=high
* [e12e615] Fix the build system to accommodate XEN changes, use kbuild as far as possible
These are fairly extensive changes. Firstly, we try to use upstream
kbuild as far as possible, to reduce churn as the files upstream
installs change. Some added checks of the Makefile are now in place so
we retain backwards compatibility. This should improve things lot wrt
header files. We also now add dependencies to more packages actually
required to build kernel images.
We also try to look for the kbuild created KERNELRELEASE variable,
which is designed to be used by distros to figure out where modules
are to be loaded from, etc. This should help reduce version
mismatches. We also prepare the kernel.release file early, to help
that.
We also refitted to support the new XEN code in mainstream, in that
the same image can be booted normally or be used as a XEN image. This
support probably needs to be improved.
The make target dependencies have been extensively reworked, to
minimize surprises and wasted effort. We also strip modules, based on
DEB_BUILD_OPTIONS (nostrip).
Extra care is now taken so we do not accidentally remove ./debian
while cleaning, thanks to upstream helpfully removing ./debian when
cleaning. This should prevent dpkg-buildpackage from accidentally
shooting itself in the foot by removing ./debian as its first action.
Bug fix: "No such file or directory", thanks to Marc Haber. I am fairly
sure this is fixed. The code around the cleaning has been
thoroughly revamped, and there have been a large number of test runs.
(Closes: #410035).
Bug fix: "kernel-package is suffering from bit rot, and is severely
broken", thanks to Manoj Srivastava (Closes: #475036).
Bug fix: "make-kpkg kernel_image fails on upstream 2.6.23 when
CONFIG_XEN is defined (new Xen paravirt_ops)", thanks to Ben McKeegan
Not only do we handle the normal image package, we should also provide
a working Xen setup. (Closes: #446879).
Bug fix: "Version check compares wrong vars", thanks to Matthijs
Kooijman. Actually, this is ont the case. We do check the upstream
version against our best guess of what whould be the correct string --
and then we ensure what we think is the right thing is what the URS
header thinks is the corect one. In any case, the KERNELRELEASE
changes above make this moot. (Closes: #433217).
Bug fix: "local version not set when config is out of date", thanks to
Matthijs Kooijman. Now we always run oldconfig before we create the
changelog, so things should work OK. (Closes: #432711).
Bug fix: "modules_clean after build with --append-to-version demands
correct --append-to-version", thanks to Marc Haber. Cleaning out the
dependency targets above cleaned out this bug as well. At least, it
worked for me. (Closes: #460507).
Bug fix: "fails to build 2.6.25-2 kernel source", thanks to Arthur
Marsh. This was extensively tested on 2.6.25.8 and 2.6.26.
(Closes: #480436).
* [18152d7] Document targets must that be run as root
The targets in question are binary, binary-arch, and
binary-indep. This is not out of line with other packages,
really. Fixing this would be too kludgey.
Bug fix: "make-kpkg fails with non-root user and fakeroot", thanks to
C. Dominik Bodi. (Closes: #501699)
* [7c6c88e] Make sure we always patch the sources before generating the changelog
The tricky part was to make sure that the first, minimal invocation
also did the right thing. The solution was to create a ./debian
directory just good enough to bootstrap a correct ./debian dir in the
second phase.
Bug fix: "debian target does not patch the source as
advertised", thanks to Dylan Thurston (Closes: #500897).
-- Manoj Srivastava <srivasta@debian.org> Fri, 10 Oct 2008 00:31:31 -0500
kernel-package (11.002) unstable; urgency=high
* Redo the NMU changes.
* [814f453] use DEB_HOST_ARCH_OS
* [46753e6] Do not quote values in kernel-img.conf examples
* [9360526] add support for building lguest launcher when lguest host
support is built-in or module
Bug fix: "add support for building lguest launcher when lguest host
support is built-in or module", thanks to Paul TBBle Hampson
(Closes: #464147).
* [c7fb8b6] Fix a typo that was preventing CROSS_ARG from being passed
to the modules.
Bug fix: "Option --cross-compile=- dosn't work with target modules*",
thanks to Jose Calhariz. This is an important bug fix.
(Closes: #417948).
* [1717962] fix bashism in an example script
Bug fix: "bashism in /bin/sh script", thanks to Raphael Geissert. This
is a release goal, since using dash as /bin.sh is desirable.
(Closes: #489617).
* [1872151] Fix asm symlink issues that created broken header packages
Bug fix: "not suitable for m-a).", thanks to Valentin QUEQUET. The fix
was entirely due to Paul <bugs@tuxcnc.org>
(Closes: #468839, #475029, #461228).
* [540ff1c] Adapt to the recent combination of i386 and amd64 images
Bug fix: "i386/x86_64 bzImage symlinks scheduled for removal in April
2010", Thanks to Frans Pop. As a result of the integration of i386 and
amd64 in the upstream kernel, kernel package failed to build debs for
a while as it could no longer find the bzImage file, until
compatibility symlinks were added. Add a version specific change for
the location of the image file. (Closes: #473705)
* [162d814] Only run a bootloader (lilo, usually) when explicitly
asked
Bug fix: "postinst doesn't grok grub2?", thanks to Trent Buck. If lilo
is installed, and the user has said run boot loader explicitly, then
run it. Do not test for grub in order to run lilo. (Closes: Bug#432124)
* [6bc983d] Remove deprecated option image_in_boot from man page,
added support and warning to postinst
Bug fix: "created kernel postinst silently ignores image_in_boot in
kernel-img.conf", thanks to pmaydell@chiark.greenend.org.uk</a>; There
is now a postinst warning, and support is added back in, for
image_in_boot (a confusing, deprecated, and now undocumented option.
(Closes: #441317).
* [dca4a03] Set a versioned path to install kernel firmware into
Bug fix: "Firmware fall-out.", thanks to Jason
Riedy. /lib/firmware/<version> is being used as a path. Fix from Brian
Rogers. (Closes: #492516).
* [701f87d] Typo fix in man page.
Bug fix: "Typo in make-kpkg(1)", thanks to Reuben Thomas
(Closes: #442080).
* [bc8d2a9] Typo fix in debconf template.
Bug fix: "typo fixes: a vs an", thanks to Frederik Schwarzer
(Closes: #486750, #487855).
* [22c79c6] Move to a streamlined, make -j friendly build system
While I am not sure of this fixes some of the nagging problems we have
been facing in recent versions of kernel-package, where we used double
colon rules, which were convenient, sure, but played havoc with
ordering of the rules, and had to have various band-aids to help out
with the ordering. The system was rapidly growing complex, with clear
indication that it was actually faster.
The new target mechanism does away with doublecolon rules, and should
play better with parrallel compilation.
-- Manoj Srivastava <srivasta@debian.org> Wed, 08 Oct 2008 23:49:52 -0500
--
Forgive and forget. Cervantes
Manoj Srivastava <srivasta@acm.org> <http://www.golden-gryphon.com/>
1024D/BF24424C print 4966 F272 D093 B493 410B 924B 21BA DABB BF24 424C
Reply to: