| freemangordon | Wizzup: strip? it is build with -O2 | 07:28 |
|---|---|---|
| freemangordon | and no '-g' | 07:29 |
| freemangordon | ok, on a33 with chimaera, the same test program uses 58280 RES | 08:16 |
| freemangordon | 76704 RES on n900 and 81632 RES on d4 with daedalus | 08:21 |
| freemangordon | Wizzup: ok, what I see as a major strace difference chimaera vs daedalus, is that chimaera uses lots of lstat64 calls when looping through theme images, while daedauls uses readlink | 08:39 |
| freemangordon | see https://paste.debian.net/1371648/ | 08:48 |
| arno11 | freemangordon: those readlink lines are also the main diff between strace running with calculator (stock) vs calculator (qt5ct tweak) FYI | 09:23 |
| arno11 | *on daedalus | 09:24 |
| freemangordon | arno11: by saying "diff" do you mean they are absent or that lstat is used instead of readlink? | 09:25 |
| arno11 | i mean they are absent | 09:26 |
| arno11 | hmm there are not absent in fact, just a lot more without tweak | 09:32 |
| arno11 | *hundreds more | 09:32 |
| freemangordon | ok, have to run now, but later on will put some traces in gtk/maemo style plugins, to see who does that | 09:43 |
| arno11 | ok | 09:46 |
| arno11 | Wizzup: able to get ringtone working: the problem was only @idle, screen off. renicing dbus-daemon @-5 (like fremantle) solves the issue | 13:07 |
| arno11 | but still the sphone crash if you don't click on 'mute ringing' before answering | 13:09 |
| arno11 | so it seems to be 2 different issues | 13:09 |
| Wizzup | seems like a race condition if renice matters | 13:15 |
| arno11 | hm seems -5 is not enough @100% of time, -8 is fine | 13:19 |
| arno11 | yeah | 13:19 |
| Wizzup | I'm worried that just renicing something solves problems | 13:20 |
| Wizzup | We really ought to fix any underlying problems | 13:20 |
| Wizzup | surely dbus-daemon isn't soo slow that it cna't deal with a few things in 5-10 seconds | 13:20 |
| arno11 | ok | 13:21 |
| arno11 | at least, renice could be useful for sicelo to avoid missed calls :P | 13:24 |
| uvos | arno11: im working on fixing the sphone crash | 15:59 |
| uvos | not that this will help you | 15:59 |
| uvos | since after the fix sphone will still exit, it just wont assert, since its in an situation it cant recover from | 15:59 |
| arno11 | ok that's fine | 16:00 |
| arno11 | btw, any idea why if i click on 'mute ringtone' before answering, it avoids the crash ? | 16:01 |
| uvos | sphone crashes because libpulse reports that it lost its connection to the pa server | 16:02 |
| uvos | muteing ringtone causes the hifi stream to be idle when the ucm profile gets switched | 16:02 |
| uvos | presumably this avoids pa breaking somehow | 16:03 |
| uvos | presumably what is happening is that pulseaudio crashes when a voice call is entered while the hifi stream is active | 16:03 |
| uvos | pulseaudio crashing then takes down sphone with it | 16:03 |
| arno11 | ah ok makes sense | 16:07 |
| arno11 | could it be possible to add a function to mute ringtone when clicking on 'answer' to avoid this kind of situation ? | 16:09 |
| uvos | no | 16:09 |
| uvos | switching proflile while the steam is active is fully valid | 16:10 |
| uvos | theres a bug somewhere that needs to fixed | 16:10 |
| Wizzup | yes, as I also stated, pa fails on my d4 regularly in asserts | 16:11 |
| Wizzup | I would not be surprised if this is the same on n900 | 16:11 |
| Wizzup | and this gets logged to /var/log | 16:11 |
| arno11 | i didn't see any other troubles atm but i didn't try mpv (which is usually a bit buggy on n900) | 16:12 |
| arno11 | but yeah maybe something a bit different in PA 16, no ? | 16:12 |
| Wizzup | looks like the ai crawler have found our git.maemo.org and are crawling the wholdr droid4-linux repo | 16:13 |
| Wizzup | so useful | 16:13 |
| arno11 | oh cool | 16:15 |
| arno11 | wow lot of failures in /var/log @PA | 16:17 |
| arno11 | *indeed | 16:17 |
| arno11 | i wonder if our default.pa is correct btw | 16:19 |
| Wizzup | what kind of failures | 16:22 |
| arno11 | [pulseaudio] module.c: Module "module-switch-on-port-available" should be loaded once at most. Refusing to load. | 16:23 |
| arno11 | *when sphone crashes @ringtone | 16:23 |
| Wizzup | that does not seem like a failure | 16:25 |
| arno11 | sorry was not talking about failures i see, just PA msg @crash | 16:25 |
| arno11 | i.e @failure: | 16:26 |
| Wizzup | so you're saying there is no crash message logged? | 16:26 |
| Wizzup | if my PA crashes I see 'assertion failed... aborting' | 16:26 |
| Wizzup | or similar | 16:26 |
| Wizzup | ser | 16:26 |
| arno11 | ok, no 'assertion failed' for me, but let me double check | 16:27 |
| arno11 | indeed, no similar failure on my device | 16:27 |
| arno11 | at least in syslog, maybe visible elsewhere ? | 16:28 |
| arno11 | ah i remember..yeah i see similar failure in sphone logs | 16:31 |
| arno11 | 'assertion...failed at../src/pulse/thread-mainloop.c:166, function pa_threaded_mainloop_stop(), Aborting | 16:32 |
| Wizzup | I think it's user.log but not sure on top of my head | 16:35 |
| sicelo | Wizzup: hoping your battery woes are over? | 16:46 |
| Wizzup | looks like it so far | 16:57 |
| sicelo | lovely :-) | 16:57 |
| Wizzup | quite insane how many /16's I had to block to get rid of the overly aggressive crawlers | 17:35 |
| sicelo | yes those are a real nightmare | 17:44 |
| freemangordon | Wizzup: seems realpath() implementation in glibc has changed | 20:04 |
| freemangordon | https://public-inbox.org/libc-alpha/c2a203a5-5817-b0f4-3f56-1ffca9f971da@cs.ucla.edu/T/ | 20:04 |
| freemangordon | this is called several times by sapwood engine for each image | 20:05 |
| freemangordon | https://git.maemo.org/leste/sapwood/src/commit/e8f249b8a26075467a14d0d93966c6f590611181/engine/theme-pixbuf.c#L188 | 20:10 |
| freemangordon | however, to me it seems we have an issue with fs on n900 | 20:11 |
| freemangordon | that code should not take that much time, even for 1000 files | 20:12 |
| Wizzup | yeah, I doubt that this is the problem | 20:14 |
| Wizzup | @ realpath | 20:14 |
| Wizzup | maybe we can revisit some tunables that we may have set in the past in sysctl - I don't think we do that any more, right | 20:14 |
| freemangordon | like what? | 20:15 |
| freemangordon | also, I may implement some filename caching in sapwood, but not sure what side effects it may have | 20:16 |
| freemangordon | I wonder if we have some kernel issue, or dunno | 20:19 |
| Wizzup | freemangordon: don't know, I just remembered arno tried some things, but if this is a fs regression somehow it'd be kernel side | 20:19 |
| Wizzup | but that should be the same on chimaera | 20:19 |
| freemangordon | having swap on emmc should not slow down the system | 20:20 |
| Wizzup | it could be we have slightly more ram pressure and suddenly vfs is not ok? no idea | 20:20 |
| freemangordon | or that, yeah | 20:20 |
| freemangordon | anyway, calling that code 1000 times should not take 3 seconds, even on n900 | 20:20 |
| freemangordon | (1000 being some arbitrary value) | 20:21 |
| freemangordon | ok, the real value is 409 | 20:26 |
| freemangordon | Wizzup: how to install the new upower? | 20:31 |
| sicelo | just apt update | 20:34 |
| sicelo | in fact, dpkg -l upower ... what does yours return? | 20:34 |
| freemangordon | upower 1:0.99.11.1-1+m7 | 20:36 |
| freemangordon | so, this one has epoch | 20:37 |
| freemangordon | I don;t think we have upower in the repos | 20:37 |
| arno11 | Wizzup: i doublechecked, no more custom sysctl stuff or similar in daedalus and chimaera-devel. still some weird stuff only in stock chimaera | 20:42 |
| freemangordon | how to check on the device? | 20:42 |
| arno11 | you mean for sysctl or upower ? | 20:43 |
| freemangordon | sysctl | 20:44 |
| arno11 | check in /etc/sysctl.d | 20:44 |
| arno11 | for n900 perf something | 20:44 |
| arno11 | the file should be empty | 20:44 |
| freemangordon | there is only a symlink | 20:47 |
| arno11 | yeah so that's 'fine' | 20:48 |
| freemangordon | mhm | 20:48 |
| arno11 | brb | 20:49 |
| freemangordon | Wizzup: btw, noatime has huge impact on n900 | 20:55 |
| arno11 | on daedalus ? i use it and don't feel too much diff | 21:00 |
| arno11 | but maybe depending of sdcard | 21:00 |
| freemangordon | mine is fast, on theory | 21:01 |
| freemangordon | *in theory | 21:01 |
| arno11 | ok | 21:01 |
| arno11 | what kind of improvement you see btw ? | 21:03 |
| arno11 | (just curious, to test) | 21:03 |
| freemangordon | ssh console session becomes way more responsive | 21:03 |
| arno11 | ah ok, interesting (i don't use it most of the time) | 21:04 |
| sicelo | Wizzup: another leste-config MR ... to update the hwclock once ntpdate has updated system clock - https://git.maemo.org/leste/leste-config/pulls/61 | 21:46 |
| sicelo | been driving me mad for some time | 21:46 |
| freemangordon | what the "-rw-r--r-- 1 root root 102598896 Jan 3 2023 /usr/lib/arm-linux-gnueabihf/libLLVM-15.so.1" ?!? | 21:49 |
| uvos | whats wierd about that? | 21:55 |
| uvos | you suprised llvm is big? or that you have an old version installed? | 21:55 |
| freemangordon | big | 21:58 |
| sicelo | freemangordon: upower - weird, it's definitely here, https://maedevu.maemo.org/leste/pool/main/u/upower/ ... but indeed i can't find with `apt-cache policy upower` | 22:00 |
| uvos | freemangordon: llvm is kinda amazingly small given what it supports | 22:01 |
| freemangordon | uvos: ok, won't argue, it is just that 100MB so seems kind of oversized to me :) | 22:01 |
| uvos | i suspect we could make it smaller by limiting the targets we compile in, theres like 50 gpu isa backends in there that are known to be pretty large | 22:02 |
| uvos | we obv dont need to compile shaders for random desktop gpus.. | 22:02 |
| freemangordon | anyway, I don;t think it is related to the issue we see on n900 | 22:02 |
| uvos | sure only a small part of that should end up in ram on n900 | 22:03 |
| freemangordon | there is some general slowness, it feels like device runs on 125 MHz all the time | 22:03 |
| freemangordon | but that's not the case, for sure | 22:03 |
| uvos | i presume the other obvious case is also not present | 22:04 |
| freemangordon | it seems that QT (or general) memory usage has increased | 22:04 |
| uvos | (too mutch memory pressure) | 22:04 |
| freemangordon | yeah, perhaps we hit that | 22:04 |
| freemangordon | as my test program startup time varies between 3s and 10s | 22:07 |
| arno11 | hmm, on my device, i don't see too much diff in term of memory usage, neither general slowness | 22:09 |
| uvos | you run with stuff turned off no? | 22:10 |
| arno11 | only for qt | 22:10 |
| uvos | ok | 22:10 |
| uvos | like if your desperate for ram you could drop conversations/tp/vcm and re enable sphones messageing, at the cost of ritch messaging support ofc. | 22:11 |
| uvos | but that should save quite the chunk | 22:11 |
| arno11 | conversation + vcm etc don't use too much ram honestly | 22:12 |
| arno11 | and new converstions_slim is nice | 22:12 |
| freemangordon | using maemo-launcher strips 1 second off startup time of the test app | 22:12 |
| freemangordon | so it start in 2700ms | 22:13 |
| arno11 | freemangordon: i'm a bit surprised your device seems slow. i mean my daedalus seems smoother and faster (apart qt) thand fremantle | 22:16 |
| arno11 | *than | 22:16 |
| freemangordon | well, maybe I access it mostly over ssh | 22:22 |
| freemangordon | and wifi latency seems big | 22:22 |
| arno11 | ah ok | 22:27 |
| arno11 | btw i doubt ram or swap usage is a problem. at least not more than chimaera (see tg-desktop working) | 22:36 |
| arno11 | and at least with sd card swap | 22:37 |
| Wizzup | freemangordon: I think we have upower withotu epoch now | 22:44 |
| freemangordon | Wizzup: where? https://paste.debian.net/1371784/ | 22:51 |
| freemangordon | do I miss a repo? | 22:52 |
| sicelo | ah, i see ... you don't have -devel enabled | 23:17 |
| sicelo | i also didn't until just now .. the correct upower does show up now | 23:17 |
| freemangordon | oh, I didn't know there is daedalus-devel repo :) | 23:29 |
Generated by irclog2html.py 2.17.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!