ali1234 | someone reported test results for it already :) | 00:00 |
---|---|---|
TJ- | The mailing list said Thursday | 00:08 |
TJ- | haha, I thought I had a bright idea to stop the Xorg on :1 being spawned. "chmod -x /usr/bin/Xorg" -- that killed the Xorg on :0 !! I thought once a program was mapped and running that shouldn't affect it | 00:09 |
TJ- | Interesing denial of service avenue :) | 00:09 |
ali1234 | hmm | 00:21 |
ali1234 | you might find this useful | 00:22 |
ali1234 | sometimes when i am debugging system components that you can't run directly, what i do is replace the binary with a shim that logs when it is started and the arguments and then runs the real binary | 00:23 |
ali1234 | this can help find race conditions if eg something is run twice accidentally | 00:23 |
TJ- | yeah, I use the method sometimes | 00:23 |
TJ- | I usually use a --bind mount over the original | 00:24 |
ali1234 | hmm i am confused | 00:33 |
ali1234 | alt-f1 on the login screen makes the font bigger? | 00:33 |
ali1234 | and alt-f2 switches to high contrast? | 00:33 |
ali1234 | alt-f3 = osk? | 00:34 |
ali1234 | but how do i get to a console? | 00:34 |
TJ- | On an X server it's Ctrl+Alt+F1-Fx | 00:38 |
TJ- | But when on a console VT it only needs Alt+Fx | 00:39 |
ali1234 | yeah, but it doesn't seem to work on the xubunut login screen | 00:39 |
ali1234 | it works after login | 00:39 |
TJ- | works here on 3 laptops, just tested. You have to release the Alt key and re-press it though, you can't sit your finger on it | 00:39 |
ali1234 | not for me | 00:40 |
TJ- | I've got 3 very different laotops here all are OK | 00:40 |
ali1234 | probably vbox being weird | 00:40 |
TJ- | ahhh, yes | 00:41 |
ali1234 | i tried your dbus line, it just prints usage :( | 00:43 |
ali1234 | oh, there should be a space between / and org | 00:44 |
TJ- | yeah, DBus is a pain with it's replication of naming in service, interface, and path | 00:45 |
TJ- | If you ever want to understand it better, install and run the GUI d-feet which allows you to browse the system and session trees as well as execute methods with or without arguments. Great aid in figuring out how to use the dbus-send CLI | 00:46 |
ali1234 | so i noticed that during first login, xfsettingsd uses 100% CPU for several seconds | 00:47 |
ali1234 | that is something that does not happen on subsequent logins | 00:47 |
ali1234 | exit | 00:48 |
ali1234 | hmmmm | 00:51 |
ali1234 | got a thread to pull on... | 00:51 |
ali1234 | log out -> delete ~/.cache -> login - the delay comes back | 00:52 |
TJ- | is it the thumbnailer ? | 00:53 |
ali1234 | that doesn't appear to generate anything in .cache | 00:53 |
ali1234 | or hasn't yet anyway | 00:53 |
ali1234 | oh, it could be | 00:53 |
ali1234 | nope | 00:54 |
TJ- | does 'top' reveal what's running? or you could start "inotifywait -rm $HOME/.cache/" from a tty then login the GUI, see what files/paths are accessed/written | 00:54 |
ali1234 | it's the fontconfig subdirectory that triggers it | 00:55 |
ali1234 | or rather the absense of it | 00:55 |
ali1234 | i will keep pulling :) | 00:56 |
ali1234 | top doesn't reveal as much as i thought it might... seems like everything is running at once | 00:56 |
TJ- | ahhh, yes, GUI font cache.. prerenders the glyphs first time | 00:58 |
ali1234 | but why so slow? | 00:58 |
TJ- | takes a while if it generates bitmaps for all glyphs on all fonts and sizes configured by the theme, etc. | 00:59 |
TJ- | I'm guessing but it makes sense | 00:59 |
ali1234 | so what you're saying is... it's ochosi's fault? ;) | 00:59 |
TJ- | haha no... it could be something entirely different | 01:00 |
knome | that's very likely though | 01:00 |
knome | ;) | 01:00 |
knome | anyway, nighty :) | 01:01 |
ali1234 | we had a missing icon cache before, maybe there is some font cache missing too | 01:01 |
ali1234 | all of the cache files in fontconfig are generated within 1 second of each other | 01:03 |
ali1234 | or... wait, that's minutes | 01:03 |
ali1234 | hmmmm | 01:05 |
ali1234 | remember these numbers: d0972 6333f | 01:08 |
ali1234 | TJ-: is there a way to make inotifywait print timestamps? | 01:08 |
TJ- | yes, use the --format xxxxxxxx option (see "man inotifywait" ) | 01:10 |
ali1234 | it seems to be tied to a specific cache file | 01:13 |
ali1234 | it better not be noto sans. i hate noto sans | 01:13 |
TJ- | at least you've got a good lead now | 01:14 |
ali1234 | i'm deleting half the files and re-logging | 01:15 |
ali1234 | down to [89ab]* now | 01:15 |
ali1234 | down to 3 files now | 01:20 |
ali1234 | hmmmmmm... got you | 01:22 |
ali1234 | deleting just one font cache file delays login by 60 seconds... it's the one that starts 9b... | 01:22 |
TJ- | 9b89f8e3dae116d678bbf48e5f21f69b-le32d4.cache-7 ? | 01:23 |
ali1234 | yes | 01:23 |
ali1234 | NotoSerif | 01:23 |
ali1234 | argh | 01:23 |
ali1234 | and also NotoSans | 01:23 |
ali1234 | i knew it :( | 01:24 |
ali1234 | Noto is google's replacement for Droid font. it supports ALL OF UNICODE, so it is absolutely enourmous | 01:24 |
TJ- | Yes | 01:26 |
ali1234 | 9b* is specifically the CJK variants | 01:26 |
ali1234 | even with how big it is, 1 minute still seems excessive | 01:26 |
TJ- | how big is it though? | 01:26 |
ali1234 | 85MB for the stff referenced in 9b* | 01:27 |
TJ- | I've got the same font selected and that file is only 44KB | 01:27 |
ali1234 | Noto is like 50 different font files though | 01:28 |
ali1234 | /usr/share/fonts/truetype/noto and /usr/share/fonts/opentype/noto | 01:28 |
ali1234 | hmm something is messed up here | 01:29 |
ali1234 | all the other fonts in opentype/ are otfs | 01:29 |
ali1234 | noto is .ttc files... what even are those? | 01:29 |
ali1234 | a "font collection" apparently | 01:30 |
TJ- | treutype-collection / | 01:30 |
ali1234 | well, time to add my findings on the bug | 01:31 |
TJ- | I wonder why it's doing that. I've just deleted the .cache/fontconfig/ directory and logged in again and it's not being recreated | 01:33 |
TJ- | OK, I've found out how to run it: "fc-cache -f" ... and the cache is taken a long time to build | 01:39 |
TJ- | only 1.1M but took about 30 seconds | 01:39 |
TJ- | "fc-cache -vf" is more helpful | 01:40 |
ali1234 | https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=845058 | 01:40 |
ubottu | Debian bug 845058 in fonts-noto-cjk "fonts-noto: Installing the package freezes all graphical applications." [Normal,Open] | 01:40 |
TJ- | and it is sitting on /usr/share/fonts/opentype/noto | 01:40 |
TJ- | says there's only 24 fonts in there | 01:41 |
ali1234 | yeah. the four collections have 8, 8, 4, and 4 fonts | 01:42 |
ali1234 | according to file | 01:42 |
ali1234 | but each one has all of the CJK unicode | 01:42 |
TJ- | the source is 85MB | 01:42 |
ali1234 | the installed files are 85MB too | 01:43 |
ali1234 | the non-CJK part of Noto is at /usr/share/fonts/truetype/noto and is 26MB but it is 156 fonts | 01:45 |
TJ- | fc-list -v | less helps to see the extend of the Noto* faces | 01:45 |
TJ- | What language is that system set to? Mine are set to en_GB-UTF-8 | 01:46 |
ali1234 | same | 01:46 |
TJ- | I don't see those big files being generated | 01:46 |
ali1234 | what big files? | 01:46 |
TJ- | oh, thought you said the cache files were large, but you were talking about the file referenced | 01:47 |
TJ- | as in the /usr/share/opentype/oto | 01:47 |
ali1234 | right. the cache is like 50kb | 01:47 |
ali1234 | it just takes a really long time to generate it when you log in | 01:47 |
TJ- | right, so the delay is worse on slower CPUs presumably, and/or slow I/O | 01:48 |
ali1234 | fc-cache -vf takes 12 seconds | 01:48 |
ali1234 | most of which is spent on noto cjk | 01:48 |
TJ- | is yours generating a system cache in /var/cache/fontconfig/ too? | 01:48 |
ali1234 | i didn't run it as root, so i guess not | 01:49 |
TJ- | no, I'm wondering if any system job is also run first time, that adds to the delay | 01:49 |
ali1234 | deleting user's cache causes 60 seconds of delay | 01:50 |
TJ- | aha, postinst | 01:51 |
TJ- | /var/lib/dpkg/info/fontconfig.postinst: fc-cache -s -f -v 1>/var/log/fontconfig.log 2>&1 || (printf "failed.\nSee /var/log/fontconfig.log for more information.\n"; exit 1) | 01:51 |
TJ- | how about if you only remove the content of $HOME/.cache/fontconfig/ rather than the entire .cache directory? | 01:51 |
TJ- | maybe it's several tasks doing a similar thing? | 01:51 |
ali1234 | that's what i did | 01:51 |
TJ- | oh, thought you meant you were removing the entire .cache | 01:52 |
ali1234 | i only delete the file 9b89f8e3dae116d678bbf48e5f21f69b-le32d4.cache-7 - nothing else | 01:52 |
ali1234 | this delays login by 60 seconds | 01:52 |
TJ- | so the slowness is accessing /usr/share/font/opentype/noto/* | 01:52 |
ali1234 | seems like it | 01:52 |
TJ- | you could monitor that with inotifywait | 01:53 |
ali1234 | hmmmmmmmmm | 01:56 |
ali1234 | i uninstalled fonts-noto-hinted (listed in the "differences" file) and the problem goes away | 01:57 |
ali1234 | not sure what that actually removed | 01:57 |
ali1234 | hmmmmmmmmmm... with that package removed... nothing is generated in ~/.cache/fontconfig | 01:59 |
ali1234 | it also removed xubuntu-default-settings | 01:59 |
ali1234 | hmmmmmmmm | 02:00 |
ali1234 | i reinstalled all the packages and still nothing gets generated in the cache on login | 02:01 |
ali1234 | time to reinstall i guess | 02:08 |
Unit193 | donofrio: software-properties (0.96.24.25) may help your issue. | 02:10 |
ali1234 | hmm | 02:17 |
ali1234 | there is some kind of problem with /var/cache/fontconfig | 02:19 |
Woowoo678 | How/where should I report a bug related to upgrading the distro from version 16.04? | 02:28 |
ali1234 | depends on the bug | 02:29 |
Woowoo678 | it looks like some residual upstart configurations break lightdm after the switch to systemd | 02:31 |
Woowoo678 | prevented me from logging into any sessions | 02:32 |
ali1234 | not sure about that | 02:32 |
ali1234 | oh no | 03:05 |
ali1234 | it isnt the mtime | 03:06 |
ali1234 | the cache files are corrupt | 03:07 |
-SwissBot:#xubuntu-devel- ::xubuntu-default-settings:: [trunk] r677 Launchpad automatic translations update. (by Launchpad Translations on behalf of xubuntu-dev) | 04:48 | |
knome | blog article prepared for announcing the results but don't go wild before we have a confirmation from all winners | 07:29 |
* bluesabre tries being less wild | 09:58 | |
Unit193 | https://launchpad.net/ubuntu/bionic/+queue?queue_state=1&queue_text=, bluesabre. | 10:01 |
bluesabre | Unit193: woohoo numlockx fix | 10:02 |
Unit193 | Sort of. | 10:02 |
bluesabre | ? | 10:03 |
Unit193 | Not accepted. | 10:03 |
TJ- | My issue with a resume-time greeter vt8 Xorg constant restart - on upgrade to 18.04 nvidia-340 had replaced nouveau. Rid the system of nvidia and it works fine. | 14:08 |
slickymasterWork | willem, Im almost sure you're aware of https://lists.ubuntu.com/archives/xubuntu-devel/2018-April/011634.html but in all cases here it is | 16:17 |
willem | ochosi, yesterday you said "willem: thats not default adwaita, by default it has a bright panel" and asked "did you use something like gtk-theme-config?" | 16:19 |
willem | I didn't. Not knowingly that is. Trying to run gtk-theme-config in terminal says "Command 'gtk-theme-config' not found, but can be installed with:" | 16:21 |
willem | Still... you've got more important things to do. I thought this would be a simple fix... | 16:22 |
willem | rebooting now to start testing the final beta... | 16:23 |
willem | slickymasterWork, I had sort of gotten the message ;-) Testing live session now. Does not go to plan unfortunately... will report on the tracker | 16:49 |
flocculant | ali1234: thanks for the font cache stuff - I did read most of it this morning - also saw you pinging in -desktop | 16:58 |
flocculant | Unit193: does the numlock script definitely check for laptops with seperate kbd and turn it on in those cases? if it does could you comment on the bug for me please : | 16:58 |
flocculant | I wondered why I couldn't tab to Unit in here | 16:59 |
willem | testing live session final beta, quick question: shouldn't the live session, after building up the desktop, when it finds wifi networks, pop up a message in the right top corner telling me about wifi networks being present? | 17:11 |
willem | disregard my question... remembered I had another laptop I could use to test; so I found the answer myself | 18:08 |
flocculant | bluesabre: I assume that we'll release this beta with the issues? | 20:09 |
flocculant | don't particularly want to do that - but guessing I'll not have a choice - also don't mean that to sound like it does, but I'm too tired now to dig out the thesauraus :) | 20:10 |
flocculant | read replies at stupid o'clock tomorrow | 20:11 |
Unit193 | flocculant: Hah, and yes it's certainly *supposed* to at least. | 20:43 |
bluesabre | flocculant: I don't think we have a beta release blocker bug | 20:57 |
bluesabre | Unless there's one I've missed. Seems like everything is hunky dory once you login once | 20:58 |
ali1234 | flocculant: will cooke mentioned wanting to get it fixed for the beta, but i think the problem is a little bit tricky due to it being an interaction with the livecd | 21:05 |
ali1234 | the good news is we know exactly the cause of the problem now. someone just needs to figure out the right way to fix it | 21:06 |
bluesabre | ali1234: that's in relation to the bluetooth bug? | 21:39 |
ali1234 | yeah. bluetooth timeout is a red herring | 21:40 |
ali1234 | it was fontconfig all alon | 21:40 |
bluesabre | wowza | 21:40 |
TJ- | hehehe, ali1234 did some great debugging last night | 21:40 |
bluesabre | And it's because we carry the monstrous noto font library? | 21:41 |
ali1234 | i just posted this complete description of the full problem: https://bugs.freedesktop.org/show_bug.cgi?id=103652#c6 | 21:41 |
ubottu | Freedesktop bug 103652 in fc-cache "fontconfig requires nano second precison for fontcache" [Normal,New] | 21:41 |
ali1234 | noto makes it slow but the rebuild should not run at all | 21:41 |
ali1234 | and it seems like fontconfig taking ~1 minute to rebuild the cache delays all other processes from starting, so they all time out too. bluez, and ubuntu has a similar problem with g-s-d | 21:43 |
bluesabre | ali1234: excellent work figuring that all out | 21:45 |
bluesabre | and TJ- too, I saw you also poking at it | 21:46 |
ali1234 | a lot of people have been poking this one because it's so widespread | 21:46 |
ali1234 | someone noticed it happening in debian in november :) | 21:46 |
bluesabre | wowza | 21:47 |
TJ- | could we use a sneaky fix? just 'touch ...' the files to stop fc rebuilding the cache? | 21:48 |
ali1234 | no, touch doesn't work because the timestamps are embedded in the cache files | 21:48 |
ali1234 | and also they have to exactly match, not just be newer | 21:48 |
ali1234 | we could do what ubuntu does and force a font cache rebuild inside the installed system during install | 21:48 |
ali1234 | then only the live desktop will be delayed | 21:49 |
ali1234 | first login on install will not | 21:49 |
TJ- | read the file's internal timestamp and apply it ? | 21:50 |
ali1234 | teres a bunch of different ways to work around it. i'm sure laney will come up with something good :) | 21:50 |
TJ- | the timestamp is at offset 0x30 in the cache file, it's a standard time_t | 21:54 |
ali1234 | yes | 21:55 |
ali1234 | but then there is a second timestamp at 0x38 which is nanosecond accurate. that is the problem | 21:56 |
ali1234 | squashfs doesn't support that, so that part of the timestamp gets zeroed | 21:56 |
ali1234 | (the second field is just the nanoseconds) | 21:56 |
ali1234 | squashfs wipes it from the actual files, but because it is embedded in the cache file, it survives there | 21:57 |
TJ- | ali1234: right, I get that, but for the installed system a .postinst script could use touch to set the correct timestamp on each cache file by reading the internal timestamp | 21:59 |
ali1234 | could do, or it could just rerun fc-cache which is simpler :) | 21:59 |
TJ- | Because the .postinst already has a stanza to trigger an fc-cache if needed | 22:00 |
ali1234 | that will happen anyway as soon as you update any font package | 22:00 |
TJ- | but that is what we're trying to prevent isn't it? the fc-cache run | 22:00 |
ali1234 | well, not ubuntu | 22:00 |
ali1234 | they want to fix it for the live cd, because they already rebuild the cache during install, so they are not affected on first login | 22:01 |
ali1234 | but if it's fixed in the live cd, that will fix it for us as well | 22:01 |
TJ- | the /postinst fc-cache run is ignored because the cache files are copied in but then when fontconfig is called on it sees wrong timestamps and calls fc-cache - do I have that correct? | 22:01 |
ali1234 | yes but you're missing one part - when fontconfig gets called it doesn't have root so it rebuilds it only for one user | 22:02 |
TJ- | for squashs I wonder if an overlayfs would help ... not sure we can do metadata only though | 22:02 |
ali1234 | that's why it happens again on first login | 22:02 |
ali1234 | during the live desktop it rebuilds it for the live user and that's wiped on reboot | 22:02 |
TJ- | yeah, the patch is the way to go I think. I was just thinking for the Thursday deadline for the beta ISOs, depending on what time tomorrow. I read earlier the ISO gen started this morning | 22:04 |
=== GridCube_ is now known as GridCube | ||
ali1234 | the ultimate fix would be to add nanos support to squashfs but of course that would be a huge change during beta | 22:08 |
TJ- | and ultimately not very useful | 22:09 |
TJ- | the assumption by fontconfig is the problem | 22:09 |
ali1234 | the assumption is only that files will never be moved between a filesystem that supports nanos and one that doesn't | 22:09 |
ali1234 | another possible fix would be to strip the nanos before building the iso cache | 22:09 |
ali1234 | they are going to disappear anyway, that would just mean doing it slightly earlier | 22:10 |
TJ- | yeah, it's a weird one though, worrying about the nanoseconds for a font cache | 22:11 |
ali1234 | it looks like fontconfig can be built entirely without support for nanos so that would be another option | 22:11 |
TJ- | I wish I could build a system entirely without support for bugs :D | 22:11 |
TJ- | We could also build the ISO images on ext4 without the extra_isize feature to disable nanosecond timestamps :) | 22:25 |
ali1234 | yes | 22:25 |
ali1234 | that would mean big changes to LP builders though | 22:26 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!