[00:36] <CIA-4> debian-installer: cjwatson * r1097 ubuntu/ (5 files in 2 dirs): Move mainline architectures to 2.6.30-5 kernels.
[00:38] <CIA-4> debian-installer: cjwatson * r1098 ubuntu/debian/changelog: releasing version 20081029ubuntu37
[12:25] <CIA-4> base-installer: cjwatson * r369 ubuntu/ (debian/changelog debian/control library.sh): Use block-attr from di-utils 1.68.
[12:27] <CIA-4> partman-crypto: cjwatson * r679 ubuntu/ (debian/changelog debian/control finish.d/crypto_config): Use block-attr from di-utils 1.68.
[12:30] <CIA-4> partman-target: cjwatson * r762 ubuntu/ (4 files in 3 dirs): Use block-attr from di-utils 1.68.
[12:32] <CIA-4> partman-target: cjwatson * r763 ubuntu/ (debian/changelog finish.d/create_fstab_header): Recommend blkid rather than vol_id in /etc/fstab.
[12:43] <CIA-4> debian-installer: cjwatson * r1099 ubuntu/ (build/pkg-lists/base debian/changelog): Add util-linux-udeb to base, for blkid.
[14:19] <CIA-4> debian-installer-utils: cjwatson * r675 ubuntu/ (14 files in 3 dirs): merge from Debian 1.68
[14:21] <CIA-4> debian-installer-utils: cjwatson * r676 ubuntu/debian/changelog: releasing version 1.68ubuntu1
[15:11] <xivulon> I was investigating the use of unionfs for wubi/colinux 9.10 and stumbled upon this article: http://lizards.opensuse.org/2009/04/28/whats-behind-lzma-compressed-livecds/
[15:14] <xivulon> source: http://git.opensuse.org/?p=people/coolo/clicfs.git
[15:21] <superm1> xivulon, keep in mind aufs (unionfs's derivative) is being discussed at uds to be axed
[15:21] <superm1> you'll want to probably sit in that discussion to see what happens of it
[15:22] <xivulon> is this in favor of unionmount?
[15:23] <superm1> i've not heard what the alternative to be proposed was
[15:23] <superm1> but I think they are discussing axing aufs because upstream has denied it going into the kernel so many times
[15:26] <cjwatson> xivulon: we investigated some of that kind of thing around 9.04 release time; the rather critical problem with that approach is that lzma kills rsyncability
[15:26] <cjwatson> but I haven't specifically looked at clicfs, and will do
[15:26] <cjwatson> axing aufs isn't in favour of *anything* yet, we literally don't know what to do
[15:27] <cjwatson> all the options suck, AFAICT
[15:27] <cjwatson> unionfs-fuse is potentially interesting
[15:29] <xivulon> mini_fo might also be interesting
[15:29] <xivulon> but looks unmaintained
[15:30] <cjwatson> mm, don't want yet another badly-maintained union filesystem :)
[15:31] <xivulon> just googled the topic :) mini_fo is used in openwrt, so it should be used quite widely
[15:31] <superm1> openwrt doesn't rev to current kernels often though
[15:32] <superm1> i'm still running a "current" 2.4 series kernel on one of my AP's
[15:32] <cjwatson> yeah, git repository last updated in June 2006, I think it's extremely unlikely that'll build let alone work
[15:32] <cjwatson> (mini_fo)
[15:33] <xivulon> the source might give some ideas though, apparently has less features/craft (only 2 layers, 1 ro, 1rw...)
[15:33] <xivulon> anyway have subscribed (thanks superm1 had missed that from the title), will be an interesting discussion
[15:34] <cjwatson> much more interested in working solutions than in ideas :-) I totally don't want to get into writing kernel union filesystems, ever
[15:36] <xivulon> I was also curious about device-mapper   snapshots, never tried that, what are the pros/cons?
[15:37] <cjwatson> the source device has to be the same filesystem as the filesystem you want to expose, so basically would need to be ext3; in order to compress that you need either gzip (too big) or the lzma patches (not in kernel yet and totally break rsyncability)
[15:37] <cjwatson> the cons are enough for me to veto it; I don't think it's usable for us
[15:37] <cjwatson> (unfortunately)
[15:38] <xivulon> I'd guess clicfs could be made a bit modular in terms of compression
[15:41] <cjwatson> well, it could, but it wouldn't help
[15:41] <cjwatson> the choices, at that level, are gzip compression or lzma compression
[15:41] <cjwatson> gzip compression won't fit, and lzma compression won't rsync
[15:43] <xivulon> I see thanks for the clarification
[15:44] <cjwatson> (we had high hopes for this and did experiments, only to be disappointed)
[15:47] <superm1> rsync can't be outfitted to support lzma compression too?
[15:49] <xivulon> is the uds schedule automatic based on presence or do I need to poke someone?
[15:51] <cjwatson> superm1: other way round
[15:51] <cjwatson> superm1: basically, lzma gets its high compression ratios because it shuffles things about a lot, which more or less inevitably reduces rsyncability
[15:52] <superm1> ah i see
[15:52] <cjwatson> you'd have to reduce its block size a lot in order to make it rsyncable, and at that point why bother using lzma ...
[15:53] <xivulon> so it cannot really be fixed
[15:53] <cjwatson> not in lzma, as far as I know, no
[16:23] <cjwatson> well, that's interesting
[16:23] <cjwatson> unionfs-fuse manages to mount, but then when casper does 'mount -o move /filesystem.squashfs /root/rofs', most of /root disappears except for /root/rofs
[16:57] <davmor2> xivulon: I've added a comment to bug 375476 can you just check if my thinking is correct please
[17:02] <xivulon> davmor2: commented in the bug and closed it
[17:03] <xivulon> any issue with r134?
[17:05] <davmor2> not had chance unfortunately hopefully latter or tomorrow morning first thing.
[17:06] <xivulon> cjwatson, robbiew mentioned that the swapfile session was scraped becase already agreed upon, could you quickly remind me what was decided re swapfile?
[17:07] <robbiew> xivulon: I believe the whiteboard section of the blueprint has a summary
[17:09] <robbiew> xivulon: whoops, nevermind.  Must have just been an email thread that I was recalling
[17:09] <robbiew> xivulon: kirkland should also be able to provide some background
[17:10] <xivulon> I am mostly interested in implications for hibernation and preseeding (given that wubi obviously uses a swap file)
[17:11] <cjwatson> preseeding obviously needs to work but I don't think it needs a *third* UDS session to decide on it - any implementation must arrange for partman-auto recipes to have an appropriate syntax
[17:11] <cjwatson> if that needs to be discussed, Debconf is a better forum than UDS, and I've committed to bringing this up with the other d-i developers there if I haven't already got something working before then
[17:12] <cjwatson> with regard to hibernation, Documentation/power/swsusp-and-swap-files.txt says it works at the kernel level but I would be entirely unsurprised to find out that there were bugs (kernel or initramfs or something else). The initial work required, though, is in adding swap files as a *non-default* option and there's no need to block that on hibernation support being known to work
[17:13] <cjwatson> indeed, it will be easier to have a meaningful discussion about hibernation once we have installer support for this stuff so that it's easy for people to try it out
[17:13] <cjwatson> I just don't want to spend valuable UDS time going over old ground
[17:13] <cjwatson> when the thing that's blocking swapfile support in the installer is simply lack of time to implement, not lack of time to design
[17:14] <cjwatson> I don't imagine anything would need to change in wubi for the moment; it may be that we can simplify partman-auto-loop later on based on core work
[17:14] <cjwatson> but wubi is not the primary target for the time being, I just want to add the core facility
[17:14] <xivulon> cjwatson thx
[17:15] <cjwatson> oh, I see why unionfs-fuse was behaving oddly for me
[17:15] <cjwatson> the mount -o move rips its read-only filesystem out from under it, and it doesn't have its own namespace
[17:16] <cjwatson> I'll try mount -o bind, and there might also be some trickery that's possible with namespace cloning
[17:27] <CIA-4> debian-installer: cjwatson * r1100 ubuntu/debian/changelog: releasing version 20081029ubuntu38
[17:31] <kirkland> robbiew: hiya, what am I providing background on?
[17:32] <robbiew> kirkland: swapfile UDS session...I think cjwatson handled it
[17:32] <CIA-4> base-installer: cjwatson * r370 ubuntu/debian/changelog: releasing version 1.99ubuntu2
[17:32] <kirkland> robbiew: gotcha;  cjwatson: thanks ;-)
[17:34] <CIA-4> partman-crypto: cjwatson * r680 ubuntu/debian/changelog: releasing version 36ubuntu6
[17:36] <xivulon>  cjwatson in man unionfs-fuse: "Presently it is not possible to create a union for /"
[17:36] <xivulon> not sure what it refers to though, canno find a bugtracker
[17:37] <cjwatson> yeah, I saw that
[17:37] <cjwatson> I'm making a working assumption that it's not referring to the initramfs case; but if it is, I think I have some ideas
[17:38] <cjwatson> it may just mean that you can't do it directly
[17:40] <CIA-4> partman-target: cjwatson * r764 ubuntu/debian/changelog: releasing version 59ubuntu2
[17:40] <xivulon> there is also the issue with sendsigs an fuse, which we handled already
[17:41] <cjwatson> right, it would be prudent to guard unionfs-fuse from being killed by sendsigs
[17:41] <cjwatson> least of the problems though :-)
[17:49] <xivulon> cjwatson: http://groups.google.com/group/unionfs-fuse/browse_thread/thread/1d4cf2a090a60de2
[17:51] <xivulon> also contains an live-cd example
[17:53] <cjwatson> spoiling my fun
[17:53] <xivulon> http://podgorny.cz/~bernd/hg/hgwebdir.cgi/radek-trunk-bernd-merge/rev/2f608ea6557e
[17:59] <xivulon> it would be nice if the cow dir was a (resizable) cow file
[18:00] <cjwatson> I'm not sure that chroot option is needed
[18:01] <cjwatson> as long as unionfs-fuse's original filesystem namespace remains available (which it will, it's the initramfs), then it should be able to just continue operating in it