[00:00] sure, i can do that tomorrow. thanks for thinking about it :) [00:00] unless my early script errored, though, i didn't see anything in syslog from it [00:05] At what site can i see changes in the installer ? And how can i profit from them? As i'm working with the liveCD, cause the installer crashes on partition stuff === cjwatson_ is now known as cjwatson [12:40] hi, the version of gutsy kernel was upgraded to 2.6.22-14.47, but Packages file in gutsy/main/debian-installer/binary-i386/ is not aware of this change [12:40] can anybody take care of this? [12:42] gutsy/main/installer-{arch}/current/images/ should be also updated I think [12:42] no, that's not correct [12:43] gutsy wasn't changed - a new version was uploaded to gutsy-security and gutsy-updates [12:43] it would be wrong to change dists/gutsy/ [12:44] I think we would only worry about uploading a new set of installer images if the changes involved a remote security vulnerability that could be exploited during installation [12:46] but is it possible to include 2.6.22-14.47 in gutsy/main/debian-installer/binary-i386/Packages at least [12:48] as current one makes ubuntu mirrors based on Packages files (like apt-mirror) unworkable with installer [12:54] no, I'm afraid it isn't [12:54] dists/gutsy is absolutely fixed and unchangeable [12:55] I've heard some reports of bugs like this, and I think it may be down to a bug in the mirroring tools; can you reproduce the problem with the master archive? [12:55] if a mirroring tool produces a result that doesn't work but the master archive does work, then it is undeniably a bug in the mirroring tool [12:56] ok, i will check this [12:57] is gutsy-security or gutsy-updates included in installer sources.list? [12:58] because 2.6.22-14.47 is trying to be installed now [13:00] the installer does use gutsy-security and gutsy-updates [13:00] if that isn't part of your mirror, then you need to add it [13:02] ok, I think that should solve it [13:17] adding main/debian-installer from gutsy-secutiry and gutsy-updates to mirror fixed my problem [13:17] thanks cjwatson [13:26] good [14:56] cjwatson: i put my syslog up at http://glueless.net/preseed/ [14:59] DHCP shows up a couple times near the end since i was trying to get the logs accesible [15:00] after i let DHCP finish it did complain about the hostname string i have set in my early script [15:00] but it still did DHCP [15:09] Now my x login looks beautiful - completely black except for "Login:" and then "Password:" [15:14] evand, any progress with the patches? [15:14] danp: the problem is that netcfg isn't installed when that preseed file is handled, so all your db_set calls fail because the questions don't exist. I'd recommend instead either just putting all that stuff into the preseed file, or (if it needs to be conditional) piping stuff to debconf-set-selections which will deal with creating the question structures for you. [15:15] Oops, wrong channel. [15:16] xivulon: I merged some of your partman-auto-loop changes yesterday, I'm looking over bug 151579 right now, actually [15:16] Launchpad bug 151579 in wubi "umountfs must check whether a mountpoint contains a loopmounted root file" [High,Fix committed] https://launchpad.net/bugs/151579 [15:17] xivulon: why is the /host/boot situation needed? It seems like quite the special case that boot would be a subfolder of the host filesystem. [15:19] also, why is the full path needed for sendsigs.omit'ing ntfs-3g and ntfs? Is there another process with the same name running at the same time as ntfs-3g? [15:19] the full path isn't needed, I took that out, didn't I? [15:19] I'm assuming it isn't, I left that part of the merge out. [15:19] evand don't see /host/boot in the patch for 15179, what are you referring to? [15:20] xivulon: sorry, that was poor transitioning. I'm asking you about other bug reports. [15:20] ah [15:21] can you remind me which one? [15:21] Specifically, 173659 and... [15:22] I don't see a bug report for the latter, but it's your usage of the full path for sendsigs.omit in your partman-auto-loop.lupin-support branch. [15:22] http://codebrowse.launchpad.net/~ago/partman-auto-loop/lupin-support/annotate/ago%40nbago-20080114235447-9rmsbro57lb2zsru?file_id=hostboot-20071221000351-3qf78yoaep16wys3-2 [15:23] that's the new code for fstab.d [15:23] indeed, I don't see why the use case for that is necessary. [15:24] and I'm asking you to explain why you need /boot to be a subfolder of the host filesystem. [15:24] because in loopinstallations boot has to be an ntfs folder [15:24] so it can be accessed by grub4dos [15:24] ah, argh. [15:24] it cannot be inside the loopfile [15:24] so we need to bindmount in fstab [15:24] fair enough :/ [15:26] so regarding your partman-auto-loop.lupin-support branch, is there any reason why you're specifying the full path or is that just leftover from an earlier version? [15:26] you mean for sendsigs? no reason on top of my head [15:27] ok, I left that out when I merged as its not necessary, I just wanted to check with you to make sure there wasn't some reasoning that I was missing. [15:27] thanks for clearing those up [15:27] re 15179 by the way use the patch mentioned in the last-1 post [15:28] it is more generic and should be closer to what cjwatson suggested [15:29] indeed, that's what I was working with [15:29] in fact it's not a patch, well insertion points should be clear [15:29] echo $skip_devices | grep -qs $DEV"$" && continue [15:29] that goes inside the loop at the top, the rest goes before the loop [15:30] it basically skips all mountpoints that appear in /proc/mounts before / [15:31] cjwatson I assume that is what you suggested [15:32] IIRC yes [15:40] evand note that 173659 involves 2 files in partman-auto-loop/lupin-support: mount.d/70bind and fstab.d/hostboot [15:41] indeed [15:50] in mount.d you might want to also check type: [ "$type" == none ] || exit 1 [15:56] mount.d/70bind and it is [ "$type" = none ] || exit 1 [16:09] cjwatson: so i can send those same answers to debconf-set-selections in my early script and it should work? [16:11] danp: yeah [16:16] awesome, thank you. i'll give that a shot [16:16] it's been driving me bonkers :P [16:24] you might also find it illuminating to cat /bin/debconf-set-selections in the installer; ignoring the rather verbose parsing code, you can see how it uses debconf [16:25] it's looking like something like 'netcfg netcfg/disable_dhcp boolean false' would do it [16:27] evand have added a comment to Bug #151579 [16:27] Launchpad bug 151579 in wubi "umountfs must check whether a mountpoint contains a loopmounted root file" [High,Fix committed] https://launchpad.net/bugs/151579 [16:27] danp: d-i rather than netcfg at the start of the line (for arcane reasons) but yes [16:27] danp: err, and you want true not false? [16:27] ok [16:27] it contains the patch for umountroot, so that any filesystem still standing gets remounted r/o (as opposed to doing that only for /) [16:28] cjwatson: oh, right. i was looking at the output of d-g-s on another machine [16:29] so 'd-i netcfg/disable_dhcp boolean true' to clarify [16:29] I did not test that last patch [16:30] cool [16:31] would that also work for things that are available when the preseed file is loaded? like if i wanted to just generate one big input to d-s-s [16:31] or do i have to use db_set for those things [16:31] debconf-set-selections is what's used for the preseed file itself; you can use it for anything [16:31] evand for autopartition-loop note that I submitted a change 2 days ago', http://codebrowse.launchpad.net/~ago/partman-auto-loop/lupin-support/revision/41 [16:32] it is a superset of db_set [16:32] I would put anything unconditional into the main preseed file, but that's just me [16:32] right [16:33] the issue was that when I called requires_disk_space, I checked for the existance of loopfiles in /host, but /host was not avalaible at that stage, so had to shuffle things around [16:35] xivulon: I already merged those changes. [16:35] good [17:05] cjwatson: any thoughts on how I should proceed with bug 183426 (basically, partman runs ridiculously slow for calc)? Looking over the logs, nothing jumps out, and I'd hate to make him run through another install to get an strace or set -x on whatever part of partman is hanging. [17:05] Launchpad bug 183426 in ubiquity "[hardy] alpha-3: i386 desktop cd - partition gui hangs" [Undecided,New] https://launchpad.net/bugs/183426 [17:08] cjwatson: success! thank you very much [17:08] my simple static test worked, now i can move back to the boot param-based stuff i was doing [17:11] is there a way to mix information from DHCP and static configuration? say, if i wanted to provide a hostname and IP but have everything else (netmask, gateway, domain, etc.) provided by DHCP [17:11] but in the end i want a static interfaces file [17:27] xivulon: What was the reasoning that Szabolcs gave for formatting the file rather than the loop device? [17:31] He said that it was safer to format the file directly without loopdev intermediaries, I never really digged any further simply changed the code [17:32] The patch by the way also add some safety net, if the host device get formatted instead of the loop file people are not going to be too happy [17:33] hrm, I'll have to follow up with him on that as I got an inquiry as to why it was necessary. Is email the best way to reach him, or is he on IRC as well? [17:33] I usually email him [17:33] ok, thanks [17:33] szaka@ntfs-3g.org [17:34] once you are there you may ask if there are any relevant upgrades to ntfs-3g which are worth considering for inclusion (to which of course he'll reply yes) [17:35] I'd be particularly interested in fallocate and fschk and safety of write operations on non-zeroed/non-contiguous loop-files [17:35] I thought fallocate wasn't in the mainline kernel yet? [17:36] something about and endless bikeshed over arguments to it [17:36] an* [17:36] did not follow that, never mind than [17:42] IIRC the suggestion came when I was asking about possible solutions re system freezes users reported during I/O operations. It was one of the things he asked me to change, it did not fix the issue (turned out to be a 2.6.22 kernel problem) but I left the code there anyway since my understanding was that it was safer anyway [17:42] and did not seem to have any negatives [18:31] Can I get another pair of eyes on this, does this look reasonably correct? cjwatson? http://pastebin.ubuntu.com/3610/ [18:32] xivulon: your patch failed in the case of there being nothing before / as NR is the number of lines processed which ends up being all of them [18:35] evand: hmm, the second field of /proc/mounts isn't necessarily unique so I'm unsure about that [18:35] xivulon: actually, that assertion is wrong, but it still fails [18:35] ah, hrm [18:36] hmm it works for me [18:36] your version though returns only / [18:36] xivulon: it fails if the second field is not unique, which as cjwatson points out, so does mine [18:39] awk '$1!~"^rootfs" && $2=="/" {print NR}' /proc/mounts [18:39] evand: instead of reading from /proc/mounts as it does, why not arrange to read from the output of sed -n '/^rootfs/,$p' /proc/mounts ? [18:40] you don't need to mess around with line numbers :) [18:40] just "print from ^rootfs to end of file" is what you want [18:40] anyway, got to go, see you tomorrow [18:41] ah, good call, thanks! [18:41] you mean in the above you have more than 2 mntpoints as / ? [18:41] enjoy your evening, goodbye [18:41] xivulon: yes [18:41] one being rootfs [18:42] can we have a list of devices that are not "real" devices? [18:42] disk devices I mean [18:42] xivulon: for example, I have: [18:43] rootfs / rootfs rw 0 0 [18:43] /dev/disk/by-uuid/0666f72f-99e1-4948-b718-220f48093423 / ext3 rw,errors=remount-ro,data=ordered 0 0 [18:43] rootfs should be skipped [18:43] your command returns 6, the position of the latter [18:43] which is correct [18:44] should return the position of the "real" root-device [18:44] ah, my mistake [18:45] xivulon: you don't need to have a list of devices. you just need to trim everything above ^rootfs out of /proc/mounts before unmounting stuff. no complexity beyond that is required. [18:45] I trimmed everything about "real" / [18:46] as opposed to rootfs [18:47] I trimmed everything ABOVE "real" root device [18:47] i.e. in the case of evand everything above and including /dev/disk/by-uuid/0666f72f-99e1-4948-b718-220f48093423 is NOT unmounted [18:48] I think that is correct [18:49] there's no need to explicitly skip rootfs because the code inside the loop already does that [18:49] case "$MTPT" in [18:49] /|/proc|/dev|/.dev|/dev/pts|/dev/shm|/proc/*|/sys|/var/run|/var/lock) [18:49] continue [18:49] hmm [18:49] anyway, really gone [18:49] the root_line looks for the line number of the real root [18:49] hence it has to ignore rootfs [18:50] that is the point to trim from [18:50] whatever is above that point is not unmounted, if rootfs happens to be above it will be skept [18:51] in fact I think it was your suggestion to ignore rootfs and focusing on / when determining what chunk of /proc/mounts to skip [20:11] evand any q on the code? I think it is correct [20:13] cjwatson re list of devices, that is just another way of splitting /proc/mounts in 2, the list contains the devices of the first chunk of /proc/mounts (containg everything up to and including root) that have to be skept [20:14] looping through the remaing chunk of /proc/mounts would yield same results [20:15] http://paste.ubuntu-nl.org/52193/ for instance does the same thing of the patch submitted [20:28] xivulon: I don't see anything particularly wrong with it aside from it no longer being explicit in what mount points its skipping, but I want to wait for cjwatson to look it over before I commit it as this is a pretty important package we're messing with. [20:31] This is safer: http://paste.ubuntu-nl.org/52195/ [20:32] in case someone has 0 or 3+ root lines in /proc/mounts (for reasons I ignore) [20:33] we can move to the grub patches then =) [20:33] heh ok [20:33] have to go now will be back in 1/2 hour or so [20:33] ok [21:16] evand in autopartition-loop>setup_loop I would keep [ "$1" != 0 ] || return [21:16] it avoids having a recipe with 0 sized /home [21:17] for instance [21:17] ah, then this also needs to be changed: [21:18] if [ ! -f "/host$path" ] && [ $1 -gt 0 ]; then [21:18] as we've already established by that point that the size is not 0. [21:18] ok, fair enough [21:18] the one at the top catches both pre-existing images and images to be created [21:19] the one in the if sentence works only for new images [21:19] right, I'm saying that the if statement can be shorted to just the file test [21:20] as the size check is redundant [21:25] grub-installer: evand * r721 grub-installer.ubuntu/ (debian/changelog grub-installer): * Handle cases where /boot is bindmounted (LP: #181658). [21:27] if the size check is on top sure [21:28] indeed [21:28] partman-auto-loop: evand * r39 partman-auto-loop.ubuntu/autopartition-loop: Fix size test in setup_loop. [21:48] hey can someone help me with a problem [21:52] please don't ask to ask a question, just ask it. [21:54] ok but it's stupid [21:54] when i try to start wubi-cdboot.exe it gives me error message "Could not find any appropriate CD" [21:55] xivulon: ^ [21:58] why does xivulon know all about the installer [21:58] Thugacation, the CD has to be in the tray [21:58] Thugacation: xivulon is the author of Wubi. [21:58] xivulon is one of the wubi team [21:59] why can't i just boot the installer from windows [21:59] wubi *is* booting the installer from Windows [21:59] wubi-cdboot is designed to assist those people that cannot boot from CD [22:00] it basically boots from HD and then continues off the CD [22:00] without editing the bios [22:00] if you prefer to boot it natively and are prepared to repartition, you can certainly also do that; set your BIOS to boot from CD and reboot [22:00] that is indeed the standard way to start Ubuntu [22:00] wubi.exe (downloaded from the website) works off the HD solely [22:01] so i can install it without having to burn any CD's [22:01] because i dont have cd's [22:01] for that, you need wubi proper, not wubi-cdboot [22:01] that cjwatson reminds of bug #181734 [22:01] Launchpad bug 181734 in casper "Prompt the user to insert a CD if a live media is not detected" [Undecided,New] https://launchpad.net/bugs/181734 [22:01] as xivulon says, you can download it from wubi-installer.org [22:02] ok thanks for your helps [22:03] np [22:07] so if i have 2 hard drives... a 200gb for windows and a 80gb for ubuntu... i can install linux to the 80gb via wubi-installer right [22:07] yes [22:08] yes i see the screenshots thank u [22:08] well in fact wubi is for installing ubuntu inside of windows [22:08] without changing the partitions [22:08] but [22:08] it you already have a partition allocated exclusively for ubuntu then a proper installation is better [22:09] a proper installation with a CD huh [22:09] damn guess i shall go buy some [22:10] you can use unetbootin (a spinoff of wubi that allows to install via netboot to dedicated partition) [22:10] I might add netboot to wubi but so far it's not planned [22:11] netinstall I mean