**** BEGIN LOGGING AT Fri Sep 02 02:59:58 2016 Sep 02 07:52:02 good morning Sep 02 08:38:35 <_william_> hi all Sep 02 08:39:16 hello Sep 02 08:39:36 <_william_> I'am trying to get both X11 and OpenGL working on an i.MX6 board and vivante driver. Anyone succeded ? all documentation i found are about removing X11 support to get OpenGL demos ans apps working. IS there a way to get both please ? Sep 02 08:58:21 _william_: might have more luck on the mailing list for such a specific question Sep 02 08:58:26 https://www.yoctoproject.org/tools-resources/community/mailing-lists Sep 02 09:14:25 <_william_> Ulfalize: thanks, i'll post to the list Sep 02 09:51:50 Has anyone any experience benchmarking bitbake in terms of gauging constraints and pain-points, such as CPU, memory or disk? Sep 02 09:53:37 We have a build server and we're currently considering upgrading it (e.g. memory). It's not slow today, but it's not particularly fast either. So before investing in server memory, it would be very nice to see if it indeed is bound on memory Sep 02 09:55:50 sveinse: there's buildstats.bbclass, though i'm not sure if it provides what you want Sep 02 09:56:09 i've seen people using performance co-pilot (pcp.io) too Sep 02 09:58:56 sveinse: We've noticed that having 2GiB of RAM per core is helpful; if you have less (and compile lots of C++ code), upgrading memory may help Sep 02 10:00:17 neverpanic: Yes. We have 16GB today and 16 cores, and I do see swapouts. And yes, very much c++ (qt). We're considering adding 64Gb physical. Sep 02 10:01:17 Please note that this server is a VM Sep 02 10:02:26 I've done quite a bit of compile time benchmarking with our server, my native laptop and running virtually under windows. I have been considering publishing the results if there are interest for it Sep 02 10:02:37 If you're on a VM, fast local SSDs also help. Network-based file systems, not the best idea. Sep 02 10:03:27 neverpanic: We use direct phy access to a dedicated SSD. Only output artefacts are stored on the network fs Sep 02 10:04:12 Sounds very reasonable. Sep 02 10:04:24 sveinse: it pretty much depends on the profile of your builds. C++ builds are usually quite slow and take lots of memory. Sep 02 10:05:07 mario-goulart: Yes, that is similar to my experience as well. Thus we're addressing memory first Sep 02 10:05:08 pcp can be handy to spot bottlenecks, yes. Sep 02 10:05:32 For simpler cases, I find htop handy too. Sep 02 10:05:39 And linux handles surplus of memory as opportunity for fs caching, so it benefits in all cases Sep 02 10:07:27 We installed the server (Ubuntu Xenial) using default options, that is with lvm2. I wonder if that is adding an unnecessary bottleneck. Any experience on that? Sep 02 10:07:33 Keep an eye on the system load. If load/ncores > 2 your system will probably start trashing. Tuning BB_NUMBER_THREADS and PARALLEL_MAKE may help to speed things up according to your build profile and resources available. Sep 02 10:08:22 Also, tuning memory manager parameters to reduce disk/SSD I/O may also be effective. Sep 02 10:09:09 mario-goulart: in the host OS? Sep 02 10:10:06 Are there any here that have experience purchasing cloud services for doing yocto building? Sep 02 10:10:42 Both host and VM, if possible. Note that by reducing the writes to disk you may increase the risk filesystem corruptions or data loss. So, maybe only in the VM if it is a throwaway system. Sep 02 10:11:45 sveinse: I don't know. But I'd be interested in that. Hetzner offers some quite powerful machines (real hardware) for a reasonable price. Sep 02 10:13:03 sveinse: if in doubt, get ram Sep 02 10:13:14 sveinse: and i mean, get really much ram Sep 02 10:14:30 sveinse: from my experience, once you get enough of that, the disk becomes m00t. my build machine does not care anymore if the build directory structure is on either HD, SSD or TMPFS Sep 02 10:16:09 LetoThe2nd: What's "much ram" in this context? Sep 02 10:16:19 sveinse: 64GB+X Sep 02 10:16:54 In my world, I am not used to operate with linux boxes with more than 16G. So >16G is much for me :P Sep 02 10:17:03 sveinse: look at your build directory, du -hs, then round up to next 2^X size :-) Sep 02 10:17:35 LetoThe2nd: yes, great, thanks Sep 02 10:19:22 sveinse: i didn't benchmark it thoroughly, but my tests were like, two simple 2tb-hds in a LVM raid 1 versus two SLC ssds in a LVM raid 1 versus tmpfs. the difference in build speed was in the range of test precision, AFAICT Sep 02 10:20:18 because once you got enough ram, everything runs out of the caches anyways. donate an UPS for emergencies, and be set. Sep 02 10:21:35 Hi, is it possible to use the POKY toolchain for cross compiling *and* host compiling? So that it would be a generic toolchain instead of using ubuntu,suse,redhat toolchain packets? For the purpose of testing the software package on the target and on the host with the same toolchain? Sep 02 10:22:48 user134: you can of course build a toolchain targetting a x86 box. but they are usually meant to be used with the supplied sysroot, not against the hosts system Sep 02 10:25:04 LetoThe2nd: That's one of the reasons why this server is virtual. It fits into the data center at work, with UPS, backup systems, and so on and so forth Sep 02 10:25:51 sveinse: #justsayin Sep 02 10:25:59 We're upgrading the host up to 96GB and plan to assign 64G to the yocto buildbox VM as step 1 Sep 02 10:26:03 LetoThe2nd: #grateful Sep 02 10:27:28 Our builds range from 18G to 66G, depending on the sstate cache hits Sep 02 10:28:58 sveinse: should be ok. probably no wonders happening there, but working Sep 02 10:30:42 sveinse: I've been opting in having more builders than a giant one Sep 02 10:31:25 sveinse: so we have been using hertzner and online.net machines for that. Sep 02 10:32:12 sveinse: https://www.online.net/en/dedicated-server/dedibox-classic is a nice one, fast and cheap Sep 02 10:33:19 otavio: A bitbake run, can that be distributed, or is one "atomic" unit one build instance? Thus, you may deploy multiple builds if you have multiple targets? Sep 02 10:34:01 sveinse: you can manage this outside; we use builds and we trigger multiple builds across them Sep 02 10:34:10 sveinse: ci.ossystems.com.br Sep 02 10:34:25 sveinse: and we trigger parallel builds as well Sep 02 10:34:56 sveinse: as an example: Sep 02 10:34:57 otavio@ci-builder-1:~$ uptime Sep 02 10:34:57 12:34:42 up 27 days, 22:30, 1 user, load average: 18.48, 26.40, 29.74 Sep 02 10:36:40 otavio: yes, but do you split bitbake tasks across multiple machines which is then joined into one output artefact, or do you build multiple targets/artefacts in parallel? Sep 02 10:36:55 Because I didn't know there were support for the former Sep 02 10:37:16 sveinse: build in parallel multiple targets Sep 02 10:37:28 otavio: right, thanks Sep 02 10:37:51 sveinse: for fsl-community-bsp, you can see the number of jobs we have Sep 02 10:38:34 otavio: You are with fsl? (We're building based on fsl) Sep 02 10:39:14 sveinse: we work with several; but NXP is one of the oldest :-) Sep 02 10:39:31 sveinse: we are the maintainers, now of i.MX and QorIQ :) Sep 02 10:41:32 otavio: Cool Sep 02 10:44:10 Right now we're struggling with a jethro build for Variscite/fsl 4.1.15-1.2.0 for i.mx6. bash crash when we do export LANG="en_US.UTF-8" in the command prompt, and have no clue in how to fix it :( Sep 02 10:46:00 It sais "loadlocal" and then some jibberish, next the terminal is reset Sep 02 10:47:02 probably loadlocale rather than loadlocal Sep 02 10:47:30 sveinse: Variscite is a problem; we have it working for one customer on top of Krogoth and it works Ok Sep 02 10:47:51 googling "loadlocale" gives some random crash reports Sep 02 10:48:02 sveinse: any reason to keep jethro for it? Sep 02 10:49:15 otavio: Not other than I have only found jethro for the variscite kit. I'd perfer krogoth if I can use it Sep 02 10:52:12 sveinse: http://freescale.github.io/doc/release-notes/2.1/index.html#supported-board-list Sep 02 10:52:47 sveinse: this needs upgrade for the kernel but so far, it works fine Sep 02 11:13:21 sveinse: still regarding build performance, check the temperature of your box and make sure that the heat is not slowing processors down. If you have enough cooling, set the CPU governor to performance. Sep 02 11:15:02 mario-goulart: I assume a rack blade server (HP ProLiant DL380) has proper thermal management. I mean, it shoulds like a jet engine when the rear and front fans kick off :o Sep 02 11:15:18 *sounds Sep 02 11:15:18 :-) Sep 02 11:16:38 If you have enough storage space, disabling rm_work (in case you have it enabled) may also help. Sep 02 11:18:08 i wonder if terabyte will be the new gigabyte soon when i comes to memory Sep 02 11:18:25 probably just a few more revisions of C++... :P Sep 02 11:19:06 I wonder if the CPU governor any impact when running on an VM... Sep 02 11:19:23 No, not in VMs. Sep 02 11:19:47 I mean, you can only apply the settings to the host system. Sep 02 11:20:02 If you can do that for the host system, it's good. Sep 02 11:20:26 VMs' performance will be indirectly affected. Sep 02 11:20:39 My knowledge with windows is essentially nul :P Sep 02 11:21:19 Ah, ok. Sorry, I'm clueless at Windows. Sep 02 11:21:36 no worries, I'll flag this to the data center admins Sep 02 11:22:46 That reminds me to configure the IO scheduler since this is SSD Sep 02 11:27:43 sveinse: and move away from Windows ;-) Sep 02 11:27:45 LOL Sep 02 11:30:51 otavio: Not my choice. :(. Not easy to be one of three people in a small Linux cluster in a large yea-o-windows-the-great environment Sep 02 11:33:19 walk in with a mac :) Sep 02 11:33:28 they'll shoot ya. Sep 02 12:08:19 So this locale issue is related to old jethro and that it has been fixed in krogoth? Sep 02 12:35:57 hi all, Im getting stuck trying to compile a qt5 based recipe "Could not find qmake configuration file linux-oe-g++." Sep 02 12:36:45 I tried adding -> require recipes-qt/qt5/qt5.inc but still the same error Sep 02 12:37:00 how do I fix that ? Sep 02 12:39:45 ok fixed with -> https://lists.yoctoproject.org/pipermail/meta-freescale/2013-August/004358.html Sep 02 13:38:21 I created a bbappend of my image and I added inherit populate_sdk_qt5 , after create the sdk using -c populate_sdk and installed into my pc I'm having problem with dependencies to compile my Qt application. Sep 02 13:38:35 error: gio/gio.h: No such file or directory Sep 02 13:39:23 I will need explicit the glibc into the image recipe? Sep 02 13:50:41 glib* Sep 02 14:00:27 hello, why is curlpp only as native package available ? Sep 02 14:15:21 iskander_work: are you sure about that? looks like it has target and nativesdk versions too. Sep 02 14:15:38 i'm looking at the metaopenembedded recipe Sep 02 14:15:49 *meta-openembedded Sep 02 14:17:17 sveinse: berton is here. Talk to him in a private chat if you prefer Sep 02 14:35:06 Ulfalize: i'm trying to build it for beaglebone, doesn't work Sep 02 14:39:33 iskander_work: how does it fail? what errors do you get? Sep 02 14:39:50 Ulfalize: my bad, sorry, it works, i got strange problöems with bitbake, killing tmp helped, no idea what went wrong, maby because i use cache :( Sep 02 14:40:13 ok, that was easy then ;) Sep 02 14:41:11 thanks Sep 02 14:42:50 iskander_work: if you only want to clean up a single recipe in tmp/ (remove its work files), you can do 'bitbake -c clean' btw Sep 02 14:43:18 i did, it didn't help Sep 02 14:43:28 got strange boost and curlpp messages Sep 02 14:43:39 i think i messed up my tmp Sep 02 14:43:40 'bitbake -c cleansstate' will do the same thing, but also clean up the cached data in the sstate cache Sep 02 14:43:48 ok, dunno then Sep 02 14:43:50 ok, thx Sep 02 14:45:02 the problem is, i built curlpp recipe but the package is libcurlpp, and the package is what is going to be installed into rootfs :) Sep 02 14:45:17 i need a break Sep 02 14:45:37 yeah, mixing up recipe and package names is pretty common :) Sep 02 14:47:03 furthermore, i configured rm_work and then tried to find curlpp source files, totally stupid Sep 02 15:47:23 whats the best way to get some external nodejs packages we use ? Sep 02 17:25:49 hello, there is a way to get a version of a package outside the recipe? Sep 02 18:59:00 Hi guys! Sep 02 18:59:37 Any idea about that while doing repo sync? Sep 02 18:59:38 https://paste.kde.org/pns5khjlx Sep 02 19:05:45 http://source.android.com/source/downloading.html Sep 02 19:07:24 Ok... does not seem to remove the warning. Sep 02 19:12:15 present: That looks like an ssh configuration problem. I believe those are ssh client messages. Sep 02 19:12:41 I suspect your ssh_config is set to reuse connections, but you're connecting to a Gerrit server that doesn't support that. Sep 02 19:13:11 Something like "ControlMaster yes" instead of "ControlMaster auto". Sep 02 19:13:15 bitbucket? Sep 02 19:15:33 I try to change my ssh config file now. :) Sep 02 19:18:29 stryx`, Bingo!!!! Sep 02 19:18:46 ControlMaster yes is the answer!! Sep 02 19:20:49 I guess the original value for: sysctl -w net.ipv4.tcp_window_scaling=1 Sep 02 19:20:52 was 1! :D Sep 02 19:21:35 you should be able to disable it on a per-host basis in .ssh/config Sep 02 19:21:40 to avoid disabling it everywhere Sep 02 19:23:17 I enabled it everywhere ;) Sep 02 19:24:04 kergoth, yes I could but it seems to be a good option anyway Sep 02 19:24:39 I upgraded Ubuntu to 16.04 LTS and forgot the distribution was not tested for Yocto... :/ Sep 02 19:25:15 *yet Sep 02 19:37:08 Thanks guys!! Sep 02 20:43:25 Hey all, is there a way to find out from what package a specific file is from? E.g. I have a file /etc/hostname (I know where this file is from, only a expample) is there a way to query from which recpie is from? Sep 02 20:43:53 after the build is complete, you can use oe-pkgdata-util to do exactly that Sep 02 20:46:47 kergoth: thanks Sep 02 20:49:02 kergoth: can you give me example. I can't find any documentation or examples for that https://github.com/01org/luv-yocto/blob/master/scripts/oe-pkgdata-util Sep 02 20:49:33 are you physically unable to run oe-pkgdata-util --help for some reason? Sep 02 20:49:53 there's plenty of info there. Sep 02 20:51:05 I'm able to run this command but I guess I'm too stupid for that Sep 02 20:51:35 kergoth: ah I forgot the wildcards sorry **** ENDING LOGGING AT Sat Sep 03 02:59:58 2016