[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

linux-kernel-headers package



OK, I've created a separate linux-kernel-headers package, and I've modified
glibc (nptl branch) to build using it.  I've checked linux-kernel-headers
into CVS; it's in the same CVS repository as glibc-package but the module is
named linux-kernel-headers:
  cvs -d :ext:<user>@cvs.debian.org:/cvs/glibc co linux-kernel-headers

You'll need an .orig.tar.gz also; it's in ~dan/public_html/ on ftp-master. 
Unpack it into the same directory, it's just the contents of include/ from a
recent -bk kernel.  Um, ftp-master is down right now but I'm sure it will be
back.

I just checked in glibc support to use it.  Unfortunately - this is
regrettable - it makes glibc a bit harder to build for this first time.  You
need linux-kernel-headers installed.  It conflicts with old versions of
libc6-dev.  Glibc will build without libc6-dev installed but you'll need an
appropriate chroot.  If anyone has a better suggestion I'd love to hear it,
but I really believe we need to get this moved out to a separate package. 
It makes tracking patches against the kernel headers a whole lot easier, and
we've already got one (MIPS msq).  Plus eventually it will be the home for
abi-headers; and it simplifies things for non-Linux ports (what we had now
would not build on hurd-i386, or would include linux headers anyway). 

Making a chroot to build from on your own machine is easy.  Building it on
Debian machines will be a little trickier.  For now, if you unpack
linux-kernel-headers into a directory and use dpkg-buildpackage -d to ignore
the build dep and set LINUX_SOURCE to point at linux-kernel-headers at it,
then everything should work.  So we can build test packages without getting
linux-kernel-headers installed and then let the buildds sort out the
bloodshed later for the official build; James and I both think that this
will work without upsetting the daemons too much.

If you build in a chroot without libc6-dev four extra tests will fail.
annexc, isomac, check-textrel, check-c++-types.sh.  Don't worry about it,
this will go away once removing libc6-dev is no longer necessary (i.e. any
time after linux-kernel-headers and a new libc-dev are in unstable).

I now have no serious blocking issues for x86.  There are failing NPTL
tests.  I am considering testing a newer glibc/nptl CVS snapshot to see if
that fixes them.

PowerPC and x86 build.  ia64 still should also.  I don't remember status on
Alpha but I think it was OK.  ARM has some issues and Phil is on it - at
least two patches are needed.  That leaves mips/mipsel (I believe more CPU
time should be available next week for testing), m68k (<shrug> kernel
headers may be a problem here), sparc (linux-kernel-headers needs some
<asm/*> wrapper magic; Jeff Bailey is doing this now); hppa (Carlos is on
it); and s390 (I have no reason to think it's broken but we'll have to find
out; s390x requires a little voodoo).  I think we're almost as ready as
we're going to get.

What do folks think?  Are we almost ready?  How much do you want done before
we put this into unstable - should we ask debian-ports and debian-devel for
arch builds and testing first?

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer



Reply to: