Thanks for your comments. On Wednesday 05 April 2006 00:06, Joey Hess wrote: > > 1) Add a boolean preseedable option debian-installer/no_net > [...] > For netcfg, what is the advantage of skipping the short dhcp test? A > static config would also not be very useful if there is really no > NIC. Yes, but it can be useful if there is a NIC that is just not connected during the install but will be later. I think this matches the old behavior. > It would seem better to skip netcfg entirely, if anything. As for > preseeding, it's already possible to preseed netcfg/use_dhcp=false. Sure, but that means setting two variables. I was also thinking of using d-i/no_net as an easy boot parameter from the boot prompt, not just for preseeding using a file. > For choose-mirror, skipping it would be nice if there is no NIC of > course. But for the rest, you've already provided a way to skip > choose-mirror in preseeding: just preseed the mirror to "don't use a > network mirror". Providing a second way to do the same thing via > debian-installer/no_net is redundant. Again, that works well when preseeding from a file, but not really from the boot prompt. > For apt-setup, if there's really no network set up, then it should > already DTRT; when it fails to connect to a mirror it will leave it > commented out. Although I think it needs some other changes, see below. But why even try to connect to a mirror if you already know there is none available? I agree though that unsetting mirror/http/hostname (and ftp) is cleaner. > > 2) The template debian-installer/no_net will be used internally too: > > [...] > If the aim is to streamline this class of installs to the same degress > as it was pre-beta2, I don't think these changes manage it. Remember, > sarge's behavior on a full CD is that is still let the user do normal > network configuration -- because they will want a network after their > install The no_net option is not intended for this use case, but only for the use case where the system is unconnected to the net. So the user will still configure the network normally. > -- but it detected it was a full CD and just asked the user to > make sure they didn't want to use a mirror too. Your proposal only > approaches that behavior if a user chooses not to configure the > network, which most users will not do. Anyone who configures the > network will still have to go through choose-mirror. But now has the added option to select "don't use", which does approach the old behavior. Whether or not to default to that option for full CD installs is open for discussion. Effectively this implements the old "do you want a mirror too" question. > > [1] A tested patch for this is attached. > > [...] > Or you could just always put the item on the menu and fail with an > error if it's selected on a CD that doesn't contain base. I considered that option as well. Fine by me. > Wouldn't d-i fail anyway later if the CD didn't contain base and the > user said to skip choosing a mirror? Seems that failing early in > choose-mirror is only a convenience. I don't know that this convenence > is really worth the nastiness of making choose-mirror know about CDs. Did you see the patch? I don't think it is that nasty and I do feel that failing early is more user friendly in general. Especially as you prevent the user from getting frustrated _after_ having scratched his partitions. Also, I already committed a change today that needs the same check in order to avoid asking the suite question (r36098) when changing the suite can lead to failed installs. > Counterproposal: > * Add "don't use a network mirror" to choose-mirror and make > choose-mirror unset mirror/http/hostname and mirror/http/directory > when that is selected. Ack. > * Test to make sure that things fail appropriately in cases where no > mirror is configured and the CD doesn't contain base. From my reading > of base-installer, it will already display > base-installer/cannot_install in this case, and the user can go back > and choose a mirror then. Not sure. debootstrap fails in a fairly nasty way if the mirror is there but the architecture is not supported on it for example (a case that is now tested for in choose-mirror). > (base-installer could even invoke choose-mirror for the user in this > case.) Good idea. Anna does the same I noticed today. Personally I prefer checking for correct/consistent setup at the point where the selections are made though (when possible). > * Make apt-setup DTRT and skip mirror generator in this case (which it > currently does not). Ack. > * Give cdrom-detect some way to determine if a CD is a full CD or a > netinst. If it's a full CD, it can set choose-mirror to default to > "don't use a network mirror" (being careful about preseeding of > course). > - base-config used the heuristic of > 200 or whatever packages > being on the CD meant it was a full CD, and we could test that in > choose-mirror. > - Maybe a better way to go would be to add a .disk/self_contained > flag file that is set for full CDs and other CDs that don't need > network to be generally useful, and have choose-mirror test for > that. Especially since apt-setup could also test for it and avoid > adding semi-useless sources.list lines for netinst CDs. I like the second option. A flag like that will be useful anyway. Maybe we should also have a flag that tells is the default tasksel tasks are fully present (to catch the powerpc CD case where these are partly on CD 2)? I'm still not convinced that we should default to "don't use" though.
Attachment:
pgpBwEySZOIOa.pgp
Description: PGP signature