Comment 7 for bug 257739

15:50 <soren> cjwatson: I've previously had a bit of a talk with BenC about which modules belong in which udebs... What sparked the discussion is the fact that it's currently not possible to netboot if you're using virtio nic's, because not all the required modules are available in a netboot environment.
15:50 <soren> cjwatson: He suggested I talk to you about it.
16:34 <cjwatson> soren: is it just a matter of including virtio-modules in the netboot initrd, then?
16:35 <cjwatson> hmm, although that has the block device too
16:36 <soren> cjwatson: virtio-net, virtio-pci, and virtio-ring, iirc.
16:37 <cjwatson> soren: personally I think I'd prefer virtio_blk to move to block-modules, virtio_net to move to nic-modules, and both of those udebs to depend on virtio-modules
16:37 <soren> They used to be in... err... nic-modules, I think, but BenC removed them to reduce the size of the udebs.
16:37 <cjwatson> I wish BenC would ask me about this stuff
16:37 <cjwatson> 'cos I'd have said "they're tiny, don't bother"
16:37 <soren> cjwatson: Heh :)
16:37 <cjwatson> every time he rearranges udebs, it means more complicated deltas for me to track in debian-installer's list files
16:37 <soren> cjwatson: It seems virtio-net is already in nic-modules.
16:38 <cjwatson> <cjwatson@riva ~>$ dpkg -c /mirror/ubuntu/pool/main/l/linux/virtio-modules-2.6.27-4-generic-di_2.6.27-4.6_i386.udeb | grep virtio_net
16:38 <cjwatson> -rw-r--r-- root/root 23836 2008-09-24 04:16 ./lib/modules/2.6.27-4-generic/kernel/drivers/net/virtio_net.ko
16:38 <cjwatson> huh, so it is. How did he manage to convince kernel-wedge to do that?
16:38 <soren> s/$/ also./
16:38 <cjwatson> that's a bug.
16:38 <soren> It's listed in both nic-modules and virtio-modules.
16:38 <cjwatson> modules shouldn't be in more than one udeb
16:38 <cjwatson> if necessary, one udeb should simply depend on another
16:39 <soren> Certainly.
16:39 <soren> it didn't used to be in nic-modules, that's for sure.
16:39 <soren> It's a recent chance.
16:39 <soren> change.
16:39 <soren> For some reason, gitweb only shows me the last couple of days worth of history :/
16:39 <soren> *days'
16:40 * soren has a bed spelling day
16:40 <soren> Err... bad.
16:40 <cjwatson> I'd suggest http://paste.ubuntu.com/50526/
16:41 * soren nods
16:41 <cjwatson> could you take that up with Ben?
16:41 <soren> cjwatson: Are the list of udebs to be included in the netboot image hardcoded somewhere or is it perhaps based on their priority or something?
16:41 <cjwatson> they are hardcoded
16:41 <cjwatson> (and dependency-expanded)
16:42 <cjwatson> priority is used to decide what anna pulls in at runtime by default
16:42 <cjwatson> we can't use priority for the initrds since different installation methods need different udebs
16:42 <cjwatson> and, honestly, for the most part we'd rather control it in the installer than have to upload the kernel to change it :)
16:43 <soren> Ok.
16:43 <soren> i was just wondering since virtio-modules was priority: extra.
16:44 <soren> while plip-modules is standard. I like to think that virtio is less esoteric than plip :)
16:44 <cjwatson> we could change that (in fact, I will), but that wouldn't help with the initrds
16:45 <soren> cjwatson: Alright. Thanks for clarifying!
16:45 <cjwatson> it would make it possible to install on a virtio block device using a netboot installer
16:45 <cjwatson> overrides changed
16:46 <soren> I'm not sure I understand.. What exactly would make it possible to isntall on a virtio blcok device using a netboot installer?
16:47 <cjwatson> having virtio-modules priority: standard
16:47 * soren ponders
16:47 <cjwatson> well, assuming the partitioner can cope
16:47 <soren> I thought you just said that priorities didn't affect anything?
16:48 <cjwatson> no, I said they didn't affect initrd building
16:48 <cjwatson> 16:42 <cjwatson> priority is used to decide what anna pulls in at runtime by default
16:48 <soren> Well, given that nic-modules depends on it now, it must become priority: standard, though.
16:48 <cjwatson> nic-modules doesn't depend on it *yet*
16:48 <cjwatson> but yes, that is also true
16:48 <soren> Point
16:48 <soren> Oh, right, I missed the part about anna.
16:48 <cjwatson> although actually I wouldn't count on udebs being closed under priority. but anyway.
16:49 <soren> Oh. I sometimes forget how special udebs really are :)
16:50 <cjwatson> debs aren't reliably closed under priority either. policy says they are, but nothing checks optional vs. extra
16:50 <cjwatson> I don't think Debian's ever actually complied with that bit of policy in my memory
16:50 <soren> Isn't there a priority_mismatches.txt or some such?
16:50 <cjwatson> yes, that covers >= standard for debs
16:50 <cjwatson> nothing checks udebs at all
16:50 <cjwatson> (it isn't necessary, so not worth bothering)
16:51 <soren> Oh, ok.