**** BEGIN LOGGING AT Fri May 01 02:59:59 2015 May 01 04:07:13 does anyone have experience in using boost-program-options in their yocto projects ? May 01 07:00:20 good evening May 01 08:06:04 morning all May 01 10:46:45 joshuagl: i'm here! May 01 10:52:46 \o/ May 01 10:52:54 * joshuagl is going to run a fido-next build on the AB May 01 10:53:04 fine by me. RP? May 01 10:54:11 rburton, joshuagl: yes May 01 10:57:00 hmm, at least I would if I could figure out where the red boxes in the instructions are May 01 10:57:06 https://wiki.yoctoproject.org/wiki/How_to_start_a_build_on_the_Autobuilder May 01 10:57:17 "You will see a slew of red boxes on your right. Click the one that says nightly defaults" May 01 10:57:25 I don't see a nightly defaults May 01 10:57:26 hm that might be old May 01 10:57:36 joshuagl: will go private May 01 10:58:21 it does look very old May 01 10:59:03 file a bug for beth :) May 01 11:00:35 I don't have my wiki account yet so can't fix it myself May 01 11:04:25 joshuagl: can't you create a wiki account? May 01 11:04:44 RP: I can, but it needs to be approved iirc May 01 11:04:49 s/can/have May 01 11:06:14 joshuagl: confirmed May 01 11:10:50 thanks RP May 01 11:14:53 joshuagl: whoops! May 01 11:28:57 zoiks! May 01 11:29:47 plenty to dig into there May 01 12:31:06 * rburton fixes fido-next and re-spins for josh May 01 16:10:23 seebs: around? May 01 16:11:06 fray: ? May 01 16:11:32 Just noticed a lot of pseudo log files are quite large (hundreds of megabytes) with "dir quasi-mismatch: '/media/build1/poky/build/tmp/work/qemux86-poky-linux/kernel-devsrc/1.0-r0/image/usr/src/kernel/scripts' [9027600] db mode 040755, incoming path had trailing slash. Not unlinking." style warnings May 01 16:11:47 i've seen a ton of those too May 01 16:11:54 didn't realize they were hundreds of megs, though, wow May 01 16:12:06 kergoth: I actually had a 1.3GB log file in one :/ May 01 16:12:11 wow May 01 16:12:15 that's wild May 01 16:12:44 quasi-mismatch eh? hmm May 01 16:13:13 kergoth: I think that was triggered by a corrupt pseudo database but there do seem to be a lot of these and they are taking a lot of disk space May 01 16:14:06 bluelightning: I was hoping seebs could translate this into english :) May 01 16:26:29 Confusingly, the incoming path did not have a trailing slash but should have May 01 16:26:48 The code checks for a mismatch, not an extra slash May 01 17:12:32 as far as I can tell, that test is just too enthusiastic May 01 17:29:04 pseudo should have a silent mode which is default May 01 17:29:14 most of the times folks dont need them May 01 17:29:22 and when they do it should be easy to turn them on May 01 17:29:41 Somone just showed me the overhead of OE May 01 17:35:21 heh, compare a buildroot build to an oe build? May 01 17:35:29 * kergoth hasn't done that in a while May 01 17:55:45 interesting approach - https://github.com/oe-lite/core/commit/6daedcce3a2ce2b33210cbc33423cfcad364a96a#diff-f34b8afd0b4436dc43d75dccfbc74812R184 May 01 18:32:48 kergoth: buildroot May 01 18:32:57 and home grown May 01 18:33:02 python based system May 01 18:33:09 ah May 01 18:33:14 the speeds are staggerring May 01 18:33:30 I wonder how many assumptions those are making about the build environment May 01 18:33:44 much fewer natives to build,i'm guessing May 01 18:33:52 nothng much other than ubuntu 14.04 with some extra packages installed May 01 18:34:18 my sstate is 3.5G minimum per device May 01 18:34:29 now share it with developers across the globe May 01 18:34:36 between that and defaulting to more features enabled in some of the recipes.. i have a prototype script laying around that automatically disables all packageconfigs with additional dependencies, unless approved, need to see if it actually builds.. May 01 18:34:37 nice May 01 18:34:44 downloading that file takes 40+ mins in some cases May 01 18:35:28 A developer did end of end timing from ground zero to him getting everything ready to work on a bug May 01 18:35:39 takes him 5 hours May 01 18:35:40 with OE May 01 18:35:43 good day, I want to initialize my postgres-database within my yocto image. is there any yocto function that is executed only once when the system boots the first time? May 01 18:35:49 and 30mins with non-OE build May 01 18:36:04 some work on 12 devices May 01 18:36:13 and size rquirements are bloated over the roof May 01 18:36:20 for disk space May 01 18:36:51 facing real practical issues May 01 18:37:02 now I see why SDK makes so much sense May 01 18:37:35 khem`: damn, that's terrible. wonder what all we can do to improve the situation May 01 18:38:05 provide binary feeds May 01 18:38:10 mechanism May 01 18:38:27 mixed for image bulding May 01 18:38:32 package feeds, then create images from them directly? or an sstate mirror? May 01 18:38:37 yeah May 01 18:39:06 depends on your needs, i guess. a lot of folks wouldn't get much use out of that since they want to customize, but for others, that'd give them a *fast* baseline to add their own apps to May 01 18:39:17 sstate is big since it have -dev and dbg mixed May 01 18:39:37 most of the times you just need dbg of few packages on target May 01 18:39:47 and -dev almost never May 01 18:40:29 i want to configure the database, e.g. executing "initdb" and executing my databasescript (plain sql) May 01 18:40:50 I also see a trend where folks dont care for much of full system being debuggable May 01 18:40:57 I have lost that argument May 01 18:41:17 there are memory constraints on devices May 01 18:41:29 you cant debug fully blown QTwebkit May 01 18:41:44 with 100s deep stack trace May 01 18:41:51 it just runs out of memory May 01 18:42:45 hb73: look up post install scripts. May 01 18:43:23 real image on w/o debug info is 90M but with debug info it goes to 1.3G May 01 18:43:32 some balance is needed May 01 18:43:38 khem`: split debug fs on host running gdb w/ gdbserver on the target May 01 18:43:41 works nicely May 01 18:43:51 kergoth: yet it does May 01 18:44:02 thank you very much zeddii May 01 18:44:08 people still need to download the dbg.rootfs :) May 01 18:44:14 speaking of, need to get fray's split debug fs patch merged into oe-core May 01 18:44:14 true May 01 18:44:35 fundamentally system works May 01 18:45:02 but what I have discovered that folks have much higher turnaround time with existing systems May 01 18:45:31 oe/yocto is fantastic for maintaining long term distros with package management and whatnot, but for a lot of other use cases, it's not so lovely May 01 18:45:53 very true May 01 18:46:00 hb73. if the docs or examples aren't clear, shout. it's a matter of testing if they are being run during rootfs creation time, and if they are, doing an exit 1. that means they'll be run on first boot. The examples will make that clear. May 01 18:46:49 A prebuilt SDK + way to repackage and image is what will simplify a lot May 01 18:47:50 I am dealing with application and middleware folks hats here and I completely see their pain when they see a kernel module failed to compile May 01 18:48:02 they dont care for that May 01 18:48:11 zeddii I will use the approach described here: http://www.yoctoproject.org/docs/1.4.2/dev-manual/dev-manual.html#usingpoky-extend-addpkg-postinstalls ("first boot"). That should fit my needs May 01 18:51:12 that's the one! May 01 18:54:15 khem`: sounds like the populate_sdk_ext will be of use, since it provides bitbake and locked down sstate May 01 18:54:20 still probably not tiny, though May 01 18:54:59 I think when we think of scale size matters a lot. its a huge cost May 01 18:55:07 now that I am able to see the numbers May 01 18:55:32 we have SSD drives since we build a lot of stuff May 01 18:55:35 with OE now May 01 18:55:45 so build speed is important too May 01 18:56:22 bunch of things stretching in oppposite directions May 01 18:57:42 if you work on platform software then its a tremendous value for what we have. May 01 18:58:07 unfortunately the ratio probbaly 1:100 when it comes for system devs + app devs May 01 18:58:25 system guys are happy May 01 18:58:43 a small fraction though May 01 19:01:54 an android or iOS SDK model is needed May 01 21:28:18 Small question. In conf/bblayers.conf, what exacly is the LCONF_VERSION? I can't seem to find any useful documentation onit. May 01 21:29:48 more or less internal layer configuration version I guess? May 01 21:30:28 So kind of useless for a 'user' May 01 21:30:32 Thanks JEEBsv May 01 21:30:49 more or less, yes May 01 21:31:24 (that said, I'm just a user as well, so this is just what I gathered from looking at things and not having any extra knowledge) May 01 21:32:02 One more question.. if you have multiple layers (meta-x, meta-y...), and each layer has a machine.conf with the same name, how would you go about choosing one over the other? May 01 21:32:53 I assume it has to do with BBFILE_PRIORITY_x, but I'm not really seeing the results I expect. May 01 21:47:59 psidhu: no, BBFILE_PRIORITY only controls which recipes are preferred and the order bbappends are applied and whatnot May 01 21:48:22 psidhu: config file / class order is determined by BBPATH, which is in turn determined by the BBLAYERS order, as each layer.conf appends to BBPATH, generally May 01 21:48:36 The first layer to have a config file will be the one used May 01 21:51:13 kergoth: thanks for that great explanation. so just put meta-y above meta-x and that's the one that will be chosen in the BBLAYERS variable, right? May 01 21:51:21 yep May 01 21:52:25 How funny, but I suppose it makes sense. I always thought of the layers as outside in, didn't realize I was putting meta-y dead last. May 01 21:54:39 khem`: FWIW, you can make OE fly if you turn debug info off May 01 21:55:31 khem`: I did some serious analysis of buildroot and crosstool verses OE. It comes down to native deps and configuration. Like to like, OE does much better with parallelism May 01 21:55:55 but we default to fully featured rather than fast and streamlined (no debugging) May 01 21:56:09 I'm very open to ideas about how we reduce disk footprint and improve speed May 01 21:56:41 If anyone's noticed, we do track those metrics and I'm about the only one continually trying to do better (e.g. all the hardlinking we do) May 01 22:09:44 RP: yes May 01 22:09:50 I have done some analysis May 01 22:09:54 too here May 01 22:10:11 sstate for native pieces is approx 300M May 01 22:10:29 that incudes cross tools May 01 22:11:23 so, I think defaults should be streamlines may be May 01 22:11:28 start using -g1 May 01 22:11:34 instead of -g May 01 22:11:35 e.g. May 01 22:12:26 khem`: I do have http://git.yoctoproject.org/cgit.cgi/poky-contrib/commit/?h=rpurdie/t222&id=cbae151b04d1c1c64777806a09fcbfc0ff044bad May 01 22:12:33 khem`: that helps a bit with size May 01 22:12:55 yes May 01 22:13:09 khem`: need a bit of refactoring to merge that May 01 22:15:21 thats something to try out yes May 01 22:15:36 let me give it a whirl here May 01 22:15:54 khem`: my builds are now stable with it, just need to clean up the patch May 01 22:16:08 khem`: been trying to clean up my patch queue May 01 22:16:15 ok, I will let you know what I see May 01 22:16:48 with monsters like chromium sstate is just aggrevating May 01 22:17:01 unstripped chrome is 1.6G in size May 01 22:17:08 and we have three copies of it in sstat May 01 22:17:15 for 3 tasks May 01 22:17:19 khem`: :( May 01 22:17:28 now think of guys who are working on it and using sstate May 01 22:17:39 and changing chromium everyday May 01 22:17:52 khem`: the question is, what else do we do though? May 01 22:18:59 khem`: the system is clever enough to know that if you're just building an image, it will grab the right package one or just linking against it, it will just grab the populate_sysroot May 01 22:19:44 RP: due to the geagraphy we have to make a tarred up sstate available May 01 22:20:02 so the size of sstate is still an issue doesnt matter how much its used May 01 22:21:04 most of the requests are like provide standard base linux + tools that seldom change and have friedlier workflow for apps May 01 22:22:01 RP: why is INHIBIT_SYSROOT_STRIP = "1" added to glibc May 01 22:22:20 khem`: because we package that separately and it needs to be unstripped May 01 22:22:30 khem`: it hit issues if it was May 01 22:22:45 khem`: don't remember the exact problem but decided to ignore that one for now May 01 22:23:02 khem`: there are patches in that branch which make sstate over http usable May 01 22:23:29 timeouts ? May 01 22:23:32 issues on http May 01 22:23:37 how do you address those May 01 22:23:53 khem`: several http connections in parallel and reuse connections May 01 22:24:01 ah i see May 01 22:24:13 khem`: ugly proof of concept code but basically works May 01 22:24:25 undertood May 01 22:24:25 khem`: just need to figure out something less ugly May 01 23:43:06 The quasi-mismatch thing is probably harmless, but I should look at it more closely. **** ENDING LOGGING AT Sat May 02 02:59:58 2015