On Mon, Mar 5, 2012 at 9:32 AM, Ben Pfaff <[email protected]> wrote: > This has a few advantages: > > * It eliminates the need to infer the kernel source directory based on > the build directory, which has broken several times during Open vSwitch > development. > > * It eliminates most of the kernel-related Autoconf code, which should > make it easier to in turn eliminate the openvswitch-datapath-dkms > dependency on libc6-dev. (Currently, openvswitch-datapath-dkms has to > depend on libc6-dev because "configure" fails hard if the C compiler > cannot generate executables, even though the dkms package never does > that.) > > * Related to the previous point, it would make it easier to distribute > a tarball that builds only the kernel module and does not contain any > of the userspace code. (This may not be a worthwhile goal, but it > makes sense to only distribute kernel module code in the Debian > packages that build kernel modules.) > > * If the kernel headers change (e.g. someone points a symlink to a new > kernel tree), then everything works OK without a new "configure" step. > > It has some disadvantages: > > * This is a new technique with which I have no experience, and so it > might have new and unexpected pitfalls. > > * The makefile syntax that replaces the Autoconf syntax might actually > be harder to read. > > Signed-off-by: Ben Pfaff <[email protected]>
I certainly like the concept although I also don't have a good idea of whether there are hidden pitfalls. The one thing that I noticed is the first time that you use this the kcompat directory won't exist, causing it to fail. _______________________________________________ dev mailing list [email protected] http://openvswitch.org/mailman/listinfo/dev
