[00:03] Hello from Lucid! [00:04] It boots way fast, great job guys [00:05] bootup times are just getting worse and worse here, 42 seconds on 1-21 and steadily rising to 1 minute 11 seconds today. guess I need to buy a SSD for this netbook :D [00:06] do you have a bootchart of your boot? [00:07] http://sarvatt.com/downloads/asuka-lucid-20100127-1.png [00:08] Sarvatt: you have to reboot twice after an upgrade :) [00:08] hey Amaranth [00:08] howdy [00:08] it has been a while [00:08] how are you? [00:09] I've been going to bed at a proper time and thus missing you :) [00:09] Amaranth! [00:09] good, doing some contract work on top of my day job, crazy busy [00:09] welcome back [00:09] hehe [00:09] Nafai sweet [00:09] one part of the contract work: make compiz start faster [00:09] Sarvatt, try booting again, it was profiling on this one [00:09] handy, that [00:09] Amaranth, who contracted you for that? [00:10] Amaranth, speaking of which do you still plan to work on cleaning things to install by default? [00:10] hmm, don't really want to say right now [00:10] k [00:10] seb128: dunno if I'll have time for it any time soon [00:10] I planned to look at that with mvo next week since you didn't reply to my pings from some weeks ago [00:10] ok [00:10] It's easy enough, just split the packaging into stuff we use + cube/rotate and stuff we don't [00:11] right [00:11] oh, and wobbly, need to keep wobbly in the default install [00:11] that might explain it, i dont think i've ever rebooted without having upgraded something inbetween unless I crashed.. [00:11] it's just that I didn't want to go through every .xml and .so to figure which ones are used [00:11] I spent all my time on it making separate packages for everything [00:11] especially if you did that already [00:11] is there an easy way to list those used? [00:12] $ gconftool-2 --get /apps/compiz/general/allscreens/options/active_plugins [00:12] the file matches the gconf list? [00:12] and there is no depends system or anything to take in consideration? [00:12] yeah, fade plugin has libfade.so and fade.xml files, etc [00:13] if there are dependencies we have those plugins loaded too so if you're just splitting it in half you don't need to care about those [00:13] s/half/two packages/ [00:13] ok good [00:13] I want a least to delete the not used one there to see what difference it makes on boot [00:14] not used ones rather [00:14] I expect there is several of those ;-) [00:14] you can do that without doing any packaging work then, just remove all the lib*.so and *.xml files that aren't in that list and reboot a couple times :) [00:14] probably gain you 0.5s [00:14] yeah, I will do that [00:15] ok, almost nothing then [00:15] I'm wondering where those 10 seconds of cpu use go [00:15] highest I could see would be 1 second [00:15] probably plugins interacting with all these windows loading [00:15] hack gnome-session to make compiz load last, see what happens [00:16] we sort of do that now, well we don't load it first [00:16] we load it as a standard application [00:16] but I will try to delay it [00:16] really it needs to load (and be done loading) right before xsplash goes away [00:16] I was talking about that with keybuk [00:17] If we have xsplash we can hide the fact that things are ugly while compiz is loading by making them talk [00:17] well officially we don't need to speed compiz since it's too slow anyway and we changed targets [00:17] but still would be nice to have faster desktop boot [00:17] changed targets? [00:17] boot target is une now [00:17] ah [00:18] there we go, a little better that time. http://sarvatt.com/downloads/asuka-lucid-20100127-2.png i think that solar plymouth theme is a little too nice for this atom cpu :D [00:29] 'night everybody [00:34] robert_ancell, I've accepted your ubuntu-desktop post and added you to the whitelist [00:35] robert_ancell, quicker to reply there than by email ;-) [00:36] TheMuso: did you find anything about about webkit accessibility? [00:38] seb128, thanks! [00:38] np [00:38] really off to bed this time ;-) [00:38] 'night [00:50] hey robert_ancell [00:52] chrisccoulson, hi [00:52] how are you? [00:53] good [00:53] dpkg-deb: building package `gnome-power-manager' in `../gnome-power-manager_2.28.1-0ubuntu1.3_amd64.deb' [00:53] woo [00:53] pitti: ^^ [00:54] another gpm update? it's taking up all my bandwidth this week ;) [00:54] chrisccoulson: ;-) this is the final one. [00:54] cool! [00:54] chrisccoulson: final one for the double suspend issue. [00:55] chrisccoulson: would you prefer i just paste the patch here or upload it to launchpad? [00:55] i think it would be better in launchpad [00:55] okay then [00:55] * hyperair reluctantly lets the huge bug page load [01:23] yay, lucid fixed one of my major complaints about the ubuntu desktop [01:32] ArneGoetje, would package should i file that antialiasing issue against, fontconfig? [01:53] chrisccoulson: the patch is attached. [01:53] hyperair - thanks [01:53] i'm going to go to bed in a minute anyway, so i will look at it in the morning [01:54] sure [02:01] LLStarks: if it's a general issue, yes. If it's only with some fonts, it might be a configuration issue in the font package. [02:02] LLStarks: please also consider the application to play a role there. Please do test if it only affects one application or multiple or all. [02:15] ArneGoetje, i'm a bit curious as to what ubuntu falls back on in firefox when tahoma fonts are called and ttf-tahoma-replacement is not installed. [02:22] LLStarks: depends on which other fonts are installed on the system [04:23] TheMuso, ping === robbiew is now known as robbiew_ === bjf is now known as bjf-afk [08:02] good morning [08:47] hey there [08:49] salut seb128 [08:49] ça va ? [08:49] lut didrocks [08:49] ouais, et toi ? [08:50] ça va bien, couché tôt ;) [08:52] didrocks, lucky you ;-) [08:52] hey slomo [08:53] hi seb128 [08:53] slomo, did you read my comments about totem-pl-parser yesterday? [08:53] seb128: heh, I saw that you were still active at 01:40 ;) [08:53] seb128: sorry for choosing a different package name than ubuntu, i wasn't expecting this to be in ubuntu already (i mean, i updated the package yesterday morning very few hours after the release) [08:54] didrocks, yeah, I was out in the evening and did read bug emails to make sure I didn't break anything when coming back [08:54] ok :) [08:54] slomo, we update some minutes after the tarballs if those are duing work hours ;-) [08:54] slomo, no problem, it's just that the name you picked don't match the debian gir policy [08:54] slomo, ie you didn't use the gir version there for example [08:55] Good morning [08:55] (and upstream gir version is incorrect, we have a patch for this release which has been integrated upstream) [08:55] hey pitti [08:55] hello pitti! [08:55] crimsun: if we need to keep it, nevermind; I'll just try to trim it a little then [08:56] * pitti waves bonjour to the French mafia [08:56] seb128: the gir has no version... the only thing that was wrong was the '-' unless i misunderstood the policy [08:56] seb128: which version did you add in ubuntu? [08:56] slomo, the "no version" is an upstream bug that didrocks fixed in ubuntu and upstream git [08:57] ok [08:57] slomo, http://git.gnome.org/browse/totem-pl-parser/commit/?id=2d51c9ad64a0b4b6504d99e2f29716701bcf0c6b [08:57] slomo, or https://bugzilla.gnome.org/show_bug.cgi?id=608176 [08:57] Gnome bug 608176 in General "Gir file aren't properly generated" [Normal,Resolved: fixed] [08:58] thanks [08:59] np [09:00] slomo, btw http://launchpadlibrarian.net/38478987/buildlog_ubuntu-lucid-i386.gstreamer0.10_0.10.25.2-2_FAILEDTOBUILD.txt.gz [09:01] slomo, known issue? [09:01] http://people.canonical.com/~pitti/bootcharts/daniel-lucid-20100128-1.png [09:02] pitti, any special change in that one? [09:02] seb128: no, thanks [09:02] seb128: just the daily; but I wanted to look at g-panel [09:02] with yesterday's change [09:03] pitti, doesn't seem to make a real difference [09:03] *nod* [09:04] still 7 seconds of desktop loading [09:04] almost twice the budget [09:04] good morning everyone [09:04] I'll try the new kernel [09:04] hey chrisccoulson [09:04] rather 8 seconds [09:04] hey chrisccoulson [09:05] hey pitti - i've got something for you to try this morning. we discussed yesterday about starting gconfd from Xsession.d, so I hosted the work I initially did at http://people.ubuntu.com/~chrisccoulson/desktop-startup-speed/crack/ [09:05] hey seb128, how are you? [09:05] chrisccoulson, a bit tired but good otherwise thanks [09:05] chrisccoulson, you? [09:05] pitti - it's not packaged in any way at all, and the helper needs compiling [09:06] seb128 - i'm quite tired too this morning. we have a sick baby, so i didn't get much sleep last night [09:06] oh :-( [09:07] chrisccoulson: nice! will do [09:08] pitti - cool, thanks! [09:08] pitti, seems we won 1 seconds compared to yesterday though [09:08] pitti, do you know what boot speed changes landed yesterday? [09:08] out of the gnome-panel one [09:09] seb128: not using gnome-wm any more [09:09] that made a big difference [09:10] pitti, that should was 2 days ago though no? [09:10] pitti, it should have been in yesterday's chart of yours? [09:10] hm, could be [09:11] right, it was [09:11] seb128: the new wncksync perhaps [09:11] I'm just wondering if after all the gnome-panel did win some 0.1 seconds [09:11] ah, no, that landed Tuesday evening, and was on yesterday's [09:11] bootchart is not reliable enough to be affirmative on such differences [09:12] *nod* [09:13] so next win is the background caching [09:13] I still need to investigate the nm-applet today [09:13] and chrisccoulson's gconf change [09:14] pitti, the "sh" process is not on today's daily btw [09:14] seb128: it sometimes disappears in the noise [09:14] ok [09:18] gosh, is g-s-d still calling xkbcomp? [09:20] pitti, it should not, why? [09:21] http://people.canonical.com/~ogra/osiris-lucid-20100127-3.png [09:26] pitti, it's probably coming from libxklavier [09:28] it is [09:28] xkl_config_get_keyboard [09:29] oh, actually, that's not public [09:29] it's xkl_xkb_activate_config_rec [09:30] chrisccoulson, that one is not used in libgnomekbd or g-s-d though... [09:30] hm, no difference with 2.6.32-12 [09:30] I've been trying to find which call lead to that without luck though [09:31] hmmm [10:08] chrisccoulson: [10:08] old: http://people.canonical.com/~pitti/bootcharts/daniel-lucid-20100128-1-2.6.32-12.png [10:08] with your hack: http://people.canonical.com/~pitti/bootcharts/daniel-lucid-20100128-gconf-xsession.d.png [10:09] no win [10:09] it increased the latency for the apps a tad (could also be noise, though), and there's still a slight CPU drop [10:09] structually it worked, though [10:09] I'll check what the difference is to drop sanity-check [10:19] chrisccoulson: any chance we could sqeeze it before ssh-agent? [10:20] http://people.canonical.com/~pitti/bootcharts/daniel-lucid-20100128-gconf-xsession.d-disabledsanitycheck.png [10:21] this ^ already looks better [10:22] seems to win 0.1 second? [10:22] * pitti tries 91x11-gconfd-helper === kermiac is now known as Kermiac_ [10:27] no real difference === Kermiac_ is now known as Kermiac [10:30] seb128, pitti: wallpaper caching seems to work well, I'll do additional test this afternoon [10:30] * didrocks gets in Paris to meet lool [10:30] didrocks: wow! [10:30] didrocks: where are you doing this? postinst magic? or at runtime? [10:30] pitti: runtime [10:31] * didrocks back in a couple of hours [10:31] o/ [10:32] didrocks: let's look over this when you are back; there are some gotchas that come to my mind [10:36] hello [10:37] eww, gst-plugins-bad now pulls in jack [10:38] is that intended? [10:38] dunno [10:52] pitti - just checking the scrollback [10:52] (sorry, i was away from my desk) [10:53] pitti - is this with my build of gconf too? [10:54] chrisccoulson: no, stock lucid otherwise [10:54] chrisccoulson: shall I reinstall your other packages again? [10:54] i think i need to hack gnome-session to comment out the gconf_init bit, to stop it from spawning gconf-sanity-check [10:54] I just chmodded that to 0 [10:54] also, gconfd has to be loaded with a working session bus, so it can't load before ssh-agent unfortunately === asac_ is now known as asac [10:55] so compiz is much less busy when delayed [10:55] chrisccoulson: I thought it just needs to be dbus-launch gconfhelper ssh-agent gnome-session ? [10:56] (just checking delay desktop boot on the mini to see how that changed) [10:56] pitti, yeah, we could do it that way. i think it is currently ssh-agent dbus-launch gcond-helper gnome-session isn't it? [10:56] seems quite some compiz work is due to things moving on screen [10:56] chrisccoulson: I moved the gconfhelper to 91, that should have done it; no real difference in the chart, thuogh [10:56] seb128_: ah, due to re-rendering? [10:56] pitti, that's what amaranth suggested [10:57] the compiz bar is also twice less busy when using a sleep 5 to start it [10:57] it takes less than 5 second then [10:57] pitti, chrisccoulson: may I ask why you still put ssh-agent everywhere? :-) [10:57] rather than some 9 seconds [10:57] vuntz - good question ;) [10:57] for the non-GNOME sessions and for people who prefer the openssh agent [10:57] vuntz, we don't, the script check if ssh-agent should be used or not [10:57] (not that it'd be a huge win, but well) [10:58] seb128_: well, that was the plan, but I gave up on it [10:58] seb128_: ah, ok [10:58] vuntz, also the openssh maintainer argue than ssh-agent is robust compared to gnome-keyring [10:58] vuntz, so that we should make hard to opt gnome-keyring out [10:58] since asking gconf whether gnome-keyring will have ssh agent support is more expensive than just starting it [10:58] pitti: so, if you're using GNOME, it's not possible to use the ssh-agent launched in the startup scripts [10:58] vuntz, not to mention features gnome-keyring lacks [10:58] vuntz: if it was so easy.. [10:58] ah, yes, unless you look at the gconf key [10:58] hrm [10:59] vuntz: in reality, it's "if you are using gnome _and_ yuo did not disable the g-keyring gconf key" [10:59] vuntz, we were considering using ssh-agent instead of gnome-keyring as an agant [10:59] agent [10:59] pitti: indeed, I forgot about the gconf key [10:59] vuntz: and since ssh-agent has to start before g-keyring, it becomes a game of looking into the future [11:00] on the other hand... [11:00] a workaround would be to have ssh-agent started by gnome-session in the initialization phase, and set the env var via the gnome-session dbus method [11:00] (using a wrapper app, I guess) [11:01] but then, might not be worth the 0.05s win [11:01] or whatever it is [11:01] we still need it to be started out of GNOME though [11:01] seb128_: why? [11:01] XFCE and the like [11:02] ah, I misunderstood. Yes, of course [11:02] although we could just package this script separately [11:02] and pull it in with XFCE, etc. [11:02] but not in UNE [11:02] would help those boxes with several desktops installed [11:02] like une session on ubuntu desktop [11:02] but yeah, in the end, it's probably not worth it, so just discard what I said for now ;-) [11:03] by and large I think the "package separately" idea will work best [11:03] the easier would be to disable gnome-keyring agent and just use the ssh one everybody [11:03] one *for* everybody [11:03] or that [11:03] but you loose a lot of nice integration with that [11:03] I'm not sure how much the agent is "integrated" [11:03] openssh's doesn't cache passphrases by default [11:04] and you have no way to put it into your normal keyring, of course [11:04] pitti - sorry, i didn't see the 3rd chart there when you dropped gconf-sanity-check [11:04] right... [11:04] i just saw it now [11:04] from my POV, g-keyring is so much nicer than the openssh one [11:05] pitti - how are you measuring the improvement on the bootchart? [11:05] right [11:05] * pitti reduces the gconf loading time by 10% with some silly seddery trick and wonders if it's worth it [11:05] what did you sed out? [11:05] the red line doesnt seem to move, but g-s-d starts quite a bit earlier [11:06] chrisccoulson: first, looking at the latency of the general app stage, and second looking at the time between the long gnome-session start line and the red "end" line [11:06] seb128_: tabs, empty string default values, and mtime [11:06] pitti - what about comments? [11:07] what annoys me much more is that this has millions of locale specific defaults [11:07] I would have though that space, etc would win so much [11:07] which would be handled much more elegantly with gettext [11:07] wouldn't [11:07] seb128_: well, that alone reduces it from 2.2 to 1.8 MB [11:07] chrisccoulson: it doesn't have any [11:08] pitti - there's no short and long descriptions in the defaults? [11:08] there are [11:08] do we need them? [11:08] hah, another such thing: %s/short_desc=""//g [11:08] chrisccoulson: unfortunately yes [11:08] :( [11:09] I think the biggest potential win would be to throw out locale specific defualts and get them through gettext [11:09] then you would get the "load the mo files to get the value".. [11:09] which I'm not sure would turn to be a win [11:09] that could be done on the client side only, though [11:09] similar to the translated descriptiosn [11:10] well I though you didn't do it for the values to avoid the mo reading? [11:10] I just didn't consider teh values at all back then [11:10] are you sure? [11:10] I thought they were for real localized default values [11:10] I though we discussed that [11:10] not for translated text strings [11:10] seb128_: I'm not sure, no [11:11] CD 수준의 [11:11] [...] [11:11] this is pure abuse of gconf [11:11] having a gconf value being used as a three-line text value which is human visible [11:11] well usually translation of default are for specific usecases... [11:11] like gedit has a key with default encodings [11:11] yeah, it does make sense for those [11:12] the one you describe seems a bug [11:12] do we have many of those? [11:12] but gnome-media is really abusing it [11:13] /usr/share/gconf/schemas/gnome-audio-profiles.schemas [11:13] that's plain crazy [11:14] urg, yes === seb128_ is now known as seb128 [11:17] seb128: moving away /usr/share/gconf/schemas/gnome-audio-profiles.schemas alone already saves 10% [11:17] and reduces file size from 2.3 to 2.0 MB [11:19] this might be worth fixing in the package itself [11:20] what do you call "fix"? [11:20] drop the audio profile descriptions from gconf [11:20] to put it where? [11:21] using gettext on the value in the app itself perhaps? [11:21] pitti, well those profiles are used in different apps [11:21] ie rhythmbox let you choose the profile to use on cd copies [11:22] well, it was just an idea [11:22] but this is not "configuration", it's a huge translation database [11:22] anyway,just looking at the hogs here [11:22] right, wrong use in any case, but I can understand why they do that [11:22] with the easy seddery we already win 10%, so that might be worth doing [11:22] should rather be a file on disk [11:22] *nod* [11:22] how much are we speaking about btw? [11:23] gconf is 0.5s? [11:23] with a libgnome-media API around [11:23] so 10% is 50ms? [11:23] somethign like this [11:23] little effort, little gain kind of thing [11:23] right, I was rather thinking about either it's worth breaking the profiles thing if that requires apps updates [11:24] I'm trying to check if they use the libgnome-media lib to get those [11:24] or if things get the gconf values directly [11:24] the three compiz-*schemas are huge as well [11:25] l -Sr /usr/share/gconf/schemas/ [11:25] ls -lSr /usr/share/gconf/schemas/ [11:25] sorry [11:25] * pitti checks if compiz is installed by default [11:25] tomboy too [11:26] oh? [11:26] oh, I purged it from my laptop, so I don't see it [11:26] * pitti does on the mini [11:27] pitti, the tomboy schemas seems to not be stripped from translations [11:27] what build component does that again? cdbs [11:27] tomboy doesn't use cdbs [11:28] /usr/share/cdbs/1/rules/langpack.mk [11:28] GETTEXT_DOMAIN="$$DOMAIN" perl /usr/lib/cdbs/strip-schema.pl $$d > $$d.new; mv $$d.new $$d [11:28] urgh [11:28] indeed, tomboy should be fixed [11:29] they use dh7 [11:29] seb128: you could just add the strip-schema.pl call to debian/rules manually? [11:29] yes [11:29] I will do that [11:30] the .desktop doesn't have the gettext domain either [11:30] we should make those cdbs magics work on dh7 too one day [11:30] why does libcompizconfig0 depend on compiz-core? [11:31] ah, nevermind; can be purgd along [11:33] * pitti unseeds compiz from netbook then [11:33] $ bzr commit -m 'drop compiz, we use mutter now' [11:33] that was an easy one :-) [11:34] does it make any difference on boot? [11:34] twiddling thumnbs, waiting on bootchart [11:34] we didn't want to have a GNOME session on UNE install? [11:34] you can install it (apt-get install ubuntu-desktop) [11:35] right, I meant by default [11:35] didrocks: ^ [11:35] I don't konw [11:35] I didn't use the une a lot before this cycle [11:35] but they use to have a desktop switcher [11:35] I know rick says he uses the GNOME mode to hack usually [11:35] and the une mode for other things [11:35] use -> used [11:36] well, but structurally the "netbook" seed should stop pulling in compiz [11:36] pitti, didrocks is not around I think, let him reply later [11:36] if we want to add GNOME to the UNE build, it should be pulled in through the desktop seed [11:36] he's having lunch with lool today iirc [11:36] so I'm not actually sure whether that really dropped compiz [11:36] ok [11:37] but we didn't drop compiz from the netbook seed when changing from netbook-launcher to the mutter plugin [11:37] still worth doing a chart without compiz [11:38] pitti, is mutter going to replace both compiz and nautilus in lucid? [11:39] seb128: nice one; gconf reduced to 60% after purging tomboy and compiz [11:39] kklimonda: no, nautilus stays around as a file browser; it just won't render the desktop (as in previous UNEs) [11:40] erm, I was thinking about metacity.. [11:40] my bad [11:40] kklimonda: yes, mutter is a composite-enabled fork of metacity [11:51] seb128: hmm [11:51] seb128: WDYT about removing the locale defaults for locales which aren't actually available on the system? [11:52] seb128: I thought about writing a "gconf-trim-defaults" script (via dpkg trigger) which removes the tabs, empty defaults, etc.; this could also filter out those localized default values which we don't need [11:54] I'll write such a thing and see what difference it'll make [11:54] ... after lunch [11:54]