Sorry for the long reply time, but the vacation time got extended at bit :). On 6/24/07, James <[EMAIL PROTECTED]> wrote: > On 6/23/07, Troels Liebe Bentsen <[EMAIL PROTECTED]> wrote: > > I'm going on vacation next week so I won't have a chance to look at > > changing the PKGBUILD before after(2. July). But if you have some old > > discussing about this topic I need to look at, just send an URL and I > > have a read. > > > > But how does the people currently maintaining the driver packages feel > > about adopting a dkms version of it? I will be happy to port the old > > one. I can also be the maintainer of the ones I have hardware for if > > non of the current developers feel like it. > > > > dkms is: > a) automagical > b) build on install > c) not as simple
Being an old RPM spec developer, I kind of like that fact that PKGBUILD's does not support multiple packages in one file. As I see it the fact that it makes it difficult to split packages up is also what helps keep Arch simple. If you look at a lot of the RPM based distros you see the result of making it easy to split packages. The whole problem of handling out of tree drivers/modules is in it self not simple, but I think dkms goes a long way to make it so. I would compare it to the initramfs scripts that run after you install a new kernel and compared to this process I would call dkms simple. As I see the current or a "multiple packages in one PKGBUILD" solution still leaves a number of problems unsolved: * Handling of kernels in private/AUR kernels(It's need to also fork all the drivers packages) * Problems of delayed drivers builds(new kernel is released but drivers are not build yet) > > all of the above approaches we discussed and rejected. The discussions > were iirc on the developer ml, so they're not available. I think the > still preferred, and simpler method is to implement multiple packages > in one PKGBUILD, which also has other benefits. We held that off for > pacman3... so I suppose it's time to revisit. I might take a look at > implementing this after I get Arch back on my laptop. > > I also mentioned DKMS last time we visited this topic, and AndyRTR who > has dealt with it under mandriva (i think) and said it wasnt suitable. Being an old mandriva user myself I can say most problems with it was ironed out about a year ago, I have been using it for my ATI(No OS drivers at the moment) card for about a 1 year now on mandriva and it has worked pretty much with out problems under many kernel upgrades. What I dislike about the mandriva way DKMS works is that it build the drivers on boot and not on install. The way to fix this is running it on driver install and on kernel install, but if this is too automagical, you could always make the user do it himself: dkms build -m module -v version -k kernel-version dkms install -m module -v version -k kernel-version But even if it's not something you would like to part of current would it be possible to get a hook in the kernel.install script so the modules could be build on kernel install? Troels. _______________________________________________ arch mailing list [email protected] http://archlinux.org/mailman/listinfo/arch
