**** BEGIN LOGGING AT Fri May 05 03:00:02 2017 May 05 07:10:45 which is the best way to learn yocto (other than official documentation ) ? May 05 07:14:20 nikincn: work in a team/project with somebody who already knows things :-) May 05 07:16:58 that is a good one actually May 05 07:17:47 I can volunteer , someone please use me , I am a fast learner and will surely make sure u benefited by offering me the chance May 05 07:18:16 nikincn: seriously, the yocto documentation is one of the best resources out there if you're not actually in the business already. another option for getting started is attending a yocto project developer day. while it is not perfect, the beginner track certainly can give you some basic understanding in a condensed form May 05 07:18:50 i also have deep connections and practical experience on hardware side I was in Shenzhen for 7 years May 05 07:19:13 ok so documentation is my way forward let me go there if there no other better option May 05 07:19:51 nikincn: plus, the YPDD resources are available at https://www.yoctoproject.org/tools-resources/presentations May 05 07:19:58 thanks a lot May 05 07:20:53 nikincn: maybe pick the intro labs (by challinan), and behanw's stuff should also be pretty neat. May 05 07:21:28 yeah the link is quite useful May 05 07:23:31 nikincn: LF offers courses which are run regularly. https://training.linuxfoundation.org/linux-courses/development-training/embedded-linux-development-with-yocto-project May 05 07:24:41 thanks but a little expensive for me May 05 07:24:46 * behanw is actually just updating the embedded linux courses and embedded drivers courses as we speak. May 05 07:25:09 nikincn: The trick is getting someone to pay for it. ;) May 05 07:25:34 :) May 05 07:25:50 The slides from the Free-electrons yocto trainings are also available if you google for them. May 05 07:25:54 There are many ways May 05 07:26:06 But dealing with experienced people is the best. May 05 07:26:21 There is a lot of misinformation out there I'm afraid. May 05 07:26:38 Stay away from Stack Overflow for Yocto/OE/Poky questions. May 05 07:26:53 Rarely correct, sadly May 05 07:27:17 ok good to know these are practical insights i needed from the community May 05 07:27:51 nikincn: But, not to oversell it, a course will be your fastest leg up. May 05 07:28:19 It's the rocket ship way to get up to speed. May 05 07:29:06 We've had plenty of "experienced people" through our courses, and they often realize they've been doing it wrong (sometimes for years). May 05 07:29:50 behanw: hey, don't tell i'm doing things right. i'll take that as a personal insult! May 05 07:29:51 * behanw goes back to slinging code and tex May 05 07:30:08 LetoThe2nd: You're always wrong. I thought we covered this before. May 05 07:30:33 i will do a cheaper course first , optimise it for the board i am making , see traction around my product then come for that course to know the mistakes I am making May 05 07:30:58 nikincn: Where are you based? May 05 07:31:22 now in India but I have my office in Shenzhen too May 05 07:32:00 nikincn: So, there are licensed resellers of our courses in India (who all do the course for a lot less than we charge) May 05 07:32:05 Same goes for China. May 05 07:32:16 i c May 05 07:33:08 So look around a bit. Resellers are supposed to only use instructors we vette May 05 07:33:24 good if they are available in India though i didnt find anyone May 05 07:33:28 So theoretically good. But you never know for sure I suppose. May 05 07:33:50 I know we have some in India. I just don't know who they are. May 05 07:33:56 Behan how do we connect outside of irc? May 05 07:34:29 I'm easy to find. behanw@$DOMAIN will reach me. Google me. I'm easily found. May 05 07:34:40 unique name and all that May 05 07:34:53 ok May 05 07:37:00 pls check linkedin May 05 07:48:42 Hi, I installed the python-gtk package in my yocto jethro image. When I launche my python script it still miss : module gconf. What can I do ? May 05 08:20:56 What would be a way to get a package list for an image without actually building the image? May 05 08:21:08 hamdy: probably no-one has packaged python-gconf May 05 08:21:34 I'm looking for something like "installed-package-names.txt" which is created under buildhistory. May 05 08:22:22 hamdy: maybe worth looking if you can disable or workaround the gconf usage in your script? May 05 08:26:03 jku : Yes I will try because there is a gconf recipe but this is not python-gconf May 05 08:47:51 nrossi: My (custom) ZYNQ board always dies after (quite) a while. Are you familiar with one such issues on any of the dev-boards? May 05 08:54:00 nrossi: The main reason I ask, instead jumping in and start debugging all sorts of stuff, is because this seems to only happen when there's basically no real activity on the CPU and I noticed that there's something called "Xilinx Zynq CpuIdle Driver" built into the linux-xlnx kernel May 05 08:56:51 JoiF: what do you mean by "dies"? May 05 08:57:42 nrossi: Haha, sounds a bit dramatic when reading it back like this. ;) May 05 08:57:43 nrossi: it becomes undead and can only be shutdown by a silver bullet! May 05 08:58:05 nrossi: The serial console becomes totally unresponsive May 05 08:58:15 LetoThe2nd: Exactly. ;) May 05 08:58:41 JoiF: can you access other interfaces like ethernet? May 05 08:58:52 nrossi: I don't have working, yet May 05 08:59:01 +ethernet May 05 08:59:02 LetoThe2nd, JoiF: So we need a Van Helzynq? May 05 08:59:11 nrossi: YMMD! May 05 08:59:17 Seems like it. ;) May 05 09:00:08 JoiF: has your dt got operating frequencies set? also might be worth setting up a heartbeat gpio May 05 09:00:35 It doesn't seem to be a usb-to-serial hardware fault, because if I rewrite the OS and reboot through xsdb, without doing any hardware resets, it all works. May 05 09:01:17 nrossi: My frequencies may or may not be off. May 05 09:01:38 nrossi: As you may recall, I needed to change the main PS clk from 50MHz to 33.33MHz May 05 09:01:58 I didn't change any other frequencies May 05 09:03:26 JoiF: it might be a case of idle -> cpufreq changes which affect iopll or similar. You can check if the clocks are linked with the clk_summary from sysfs. Also you could test changing the cpufreq May 05 09:04:03 nrossi: Good point. May 05 09:04:06 JoiF: Although i have not seen the behavior you are describing so it could be unrelated and might just be a problem with the serial device May 05 09:04:15 nrossi: I'll look into that May 05 09:04:28 nrossi: True. May 05 09:47:24 https://www.irccloud.com/pastebin/wRq8Tvyh/ May 05 09:47:49 Hi, anyone had seen similar issue on imx platform ? kernel 3.14.38 May 05 09:48:04 micrel phy 9021 May 05 10:25:51 is there a way to clean a package build/recipe in work through devtool? May 05 10:30:52 like, the equivalent to bitbake -c clean May 05 11:05:54 I'm a bit confused right now .. what is the opposite of IMAGE_INSTALL? May 05 11:06:53 I have my own image-recipe, that currently contains just "require recipes-extended/images/core-image-full-cmdline.bb" and then an IMAGE_INSTALL += line, but let's say I wanted to remove something that otherwise comes from core-image-full-cmdline? May 05 11:08:46 So.. instead of cherrypicking->in .. I'm cherrypicking->out May 05 11:09:03 you can use BAD_RECOMMENDATIONS to filter out recommended packages that are by default pulled in May 05 11:09:17 but you can't just say "don't install libfoo even though bar depends on it" May 05 11:09:27 depends means depends May 05 11:09:38 JoiF: ah so you're actually not asking about IMAGE_INSTALL itself, but the += that is often used together with it. May 05 11:09:40 there's an argument that packagegroups should recommend instead of depending, i guess May 05 11:10:05 Yes. But in this case, I'd happily remove all the dependencies. May 05 11:10:09 theoretically you should be able to use IMAGE_INSTALL_remove May 05 11:10:29 So as an example, what I would like to do right now is get rid of NFS-stuff. May 05 11:10:52 It even complains that NFS support isn't compiled into my kernel anyway, and I don't intend to start using it. May 05 11:11:04 wellso what pulls nfs in? May 05 11:11:52 i suggest you start by not just including core-image-full-cmdline May 05 11:11:59 rburton: ++ May 05 11:11:59 copy/paste and then start deleting bits May 05 11:12:19 Ok. So cherrypick->in ;) May 05 11:12:23 full-cmdline intentionally pulls in lots of pieces May 05 11:12:45 Yeah, and I like having most of them .. just not all of them ;9 May 05 11:12:46 ;) May 05 11:13:00 also nfs is a DISTRO_FEATURE, so remove it from there and theoretically the packagegroups should empty themselves May 05 11:13:04 (and if they don't thats a bug) May 05 11:13:26 "empty themselves", that sounds kinda threatening. May 05 11:13:32 haha May 05 11:14:59 grep nfs ./meta/recipes-extended/packagegroups/packagegroup-core-full-cmdline.bb May 05 11:14:59 nfs-utils \ May 05 11:15:04 yeah, was about to paste that May 05 11:15:09 Well, I guess I have my answer :) May 05 11:15:18 see thats why you don't use full-cmdline if you don't actually want a full cmdline May 05 11:15:28 (bonus points for a patch to wrap that in a distro features check) May 05 11:15:41 haha May 05 11:19:46 Well, that's what I get for trying to be lazy. I'll start defining the packages that I *DO* want now. ;) May 05 11:25:51 kanavin: dnf has a funky entry in RRS May 05 12:31:13 rburton: right, yes :) May 05 12:31:19 rburton: also, I don't mean to be rude, but http://recipes.yoctoproject.org/rrs/recipes/2.3/All/?upstream_status=Not%20updated&maintainer_name=Ross%20Burton# May 05 12:31:31 rburton: let's make a deal: you fix it first :) May 05 12:32:27 alimon: RRS hasn't been updated in many, many days... again May 05 13:09:51 For example, the "meta-poky" layer defines a "yocto" collection (presumably because of historic reasons) and that collection depends on the "core" collection from the "openembedded-core" layer: LAYERDEPENDS_yocto = "core" May 05 13:10:03 (from meta-poky/conf/layer.conf). May 05 13:32:47 rburton: why is 0: gettext-0.19.8.1-r0 do_configure - 92s (pid 22011) so epic? May 05 13:33:11 rburton: can we make it get out of the way quickly? May 05 13:49:49 kanavin: that task is the one I studied in some performance tests. Its basically about five configure passes May 05 13:50:52 RP: ouch... it is a bottleneck though :( May 05 13:51:26 kanavin: totally, hence my study. I even asked some questions about it on LKML. Never figured out why its so slow though (other than being a forkbomb) May 05 13:53:27 RP: I guess the only way to fix it is to rewrite the build system to not be autotools May 05 13:54:17 kanavin: I do wonder if there is some underlying sheduling or locking issue wrt the kernel we may be able to mitigate somehow May 05 13:54:27 kanavin: we could also put autotools on a diet May 05 13:59:00 what did the kernel do now? I missed the context. May 05 14:11:11 Hi guys. bugzilla.yoctoproject always confuses me. Where do I report a bug about a recipe (lua) in meta-oe? It's not in oe-core. "Other YP layers"? Component? May 05 14:11:16 paulg: we were discussing why gettext's do_configure is so slow - it's on a critical path, so delays any build May 05 14:13:02 rburton: I've debreakified meta-oe, there was more than just libpng actually May 05 14:13:57 kanavin, ok, well I'll go out on a limb and guess the kernel isn't responsible, else other workloads would be impacted. May 05 14:14:45 I am creating a new recipe for Bazel: https://bazel.build/versions/master/docs/install-compile-source.html, it requires openjdk-8-jdk, thus I added the meta-java layer and packages May 05 14:15:03 now I have to define some environment variables like JAVA_HOME etc May 05 14:15:54 how can I do that? at the moment my do_compile export an absolute path like: export JAVA_HOME=/home/giuseppe/research/projects/zynq/yocto/build/tmp/work/zc702_zynq7-poky-linux-gnueabi/core-image-minimal/1.0-r0/rootfs/usr/lib/jvm/openjdk-8 May 05 14:16:33 I am wondering how I can retrieve some of those paths in bitbake May 05 14:43:48 RP: rburton: I have some AB changes to test, any preference which cluster I use? May 05 14:44:09 joshuagl: none from me but remember maint window is coming May 05 14:44:27 I really should put that in my calendar or something May 05 14:44:37 what time is maint window? May 05 14:44:40 halstead: ? May 05 14:46:57 joshuagl: I can start after a build. Just sometime today. Usualy starts in about 2 hours. May 05 14:47:24 halstead: OK, I'll start a test build then if you don't mind May 05 14:47:29 ? May 05 14:47:36 or we can maybe update a cluster earlier? May 05 14:47:54 joshuagl: Go ahead and start now. May 05 14:48:41 Probably test on both if possible? May 05 14:48:50 halstead: double ack, thanks May 05 14:59:04 getting a "undefined symbol: Perl_xs_handshake" error, https://pastebin.com/2cPmKft6. Seems ulf solved this a few weeks ago. https://www.yoctoproject.org/irc/%23yocto.2016-04-24.log.html May 05 15:02:17 but this isn't working for me. May 05 15:04:08 kanavin: thanks for notice, i'll take care, is a pending topic to develop some notification failures sorry for that... May 05 15:05:59 paulg: i can dig out the lkml posts, it was quite an interesting exercise in pathalogical scheduling May 05 15:06:02 Nevermind, 2016, so a year ago... probably not the same issue May 05 15:06:25 alimon: yeah, please try to keep it running, or at least get notified when it breaks May 05 15:06:27 paulg: it appeared to be round-robin scheduling every fork from configure across my cores, so they were constantly at the lowest speed May 05 15:06:40 paulg: use tasksel to pin it to a cpu and it was a *lot* faster May 05 15:07:07 diego_r: meta-oe doesn't use yocto bugzilla, report bugs on the mailing list May 05 15:09:26 rburton, autoconf/configure doing something pathalogical? I'm so shocked. May 05 15:09:38 paulg: i was shocked too May 05 15:10:54 lets see if 4.9 has the same behaviour May 05 15:11:48 * paulg assumes you can still order "I hate autoconf" t-shirts. May 05 15:17:13 rburton: thanks. So I need to amend my mistake May 05 15:44:15 paulg: http://linux-kernel.2935.n7.nabble.com/Performance-of-forky-workloads-td1255622.html May 05 15:45:36 kanavin: ^^^ May 05 15:45:43 halstead: test run is done, thanks May 05 15:46:08 joshuagl, That was quick. May 05 15:46:29 halstead: well, the build failed but it's not the fault of my changes (he states confidently) May 05 16:06:16 Any thoughts on this? May 05 16:16:30 i just switched to morty branch from krogoth, first time trying to build May 05 16:34:21 er.. | ./run_tic.sh: line 140: unknown: command not found May 05 16:34:28 * kergoth checks current ;master May 05 16:44:15 AB maintenance is complete on both clusters. May 05 16:54:49 sigh, ncurses-native didn't end up in the native recipe sysroot for ncurses, despite being in DEPENDS. lovely May 05 16:57:12 Has bash any restrictions running under fakeroot? May 05 16:57:45 I'm having problems with spu-image-1.0-r0 do_image: /home/sveinse/yocto/build-sp-image-dev/build/tmp/sysroots/x86_64-linux/usr/bin/vs2: line 744: /dev/fd/62: No such file or directory May 05 16:58:35 Which is due to constructs like: "while { read -r line; } <&3; do" and "done 3< <(cat "$in")" May 05 16:58:41 Anything ring a bell? May 05 16:59:01 is vs2 a script? May 05 16:59:08 Yes, a bash script May 05 17:01:27 Are any /dev files masked or tampered with when running under fakeroot/pseudo? May 05 17:02:02 Google tells me that something is missing. E.g. by running ln -s /proc/self/fd /dev/fd would fix it May 05 17:02:22 Can it be due to the limited environment under fakeroot? May 05 17:02:29 paging seebs for possible pseudo behaviour issues May 05 17:02:50 presumably it works outside of fakeroot May 05 17:03:08 Yeah, let me test that May 05 17:04:56 Yes, purred like a kitten May 05 18:05:25 rburton: gcc 7.1 released few days ago, I have updated recipes May 05 18:05:40 rburton: this should hopefully be last update to gcc7 May 05 18:11:44 * zeddii_home has a 4.1 kernel backport for that as well. I just ran out of time. I’ll have to grab those changes and confirm as well. May 05 18:24:28 khem: awesome May 05 18:25:37 khem: pushed already or should i wait for patches? May 05 18:26:15 So, long story short: pseudo doesn't special-case /dev or /proc May 05 18:26:29 Worse, it produces "fake" device files. May 05 18:26:37 So if you actually need /dev to work, it will not be your friend. May 05 18:26:45 (Reason is, non-root users can't make "real" device nodes.) May 05 18:27:00 We could probably special-case a thing for "use /dev for /dev" maybe? May 05 18:44:52 seebs: could pseudo pass through access to /dev with an overlay for modifications it needs to remember? May 05 18:50:36 seebs: not sure I'd say I'd need /dev directly, but using custom fd's in bash should perhaps be possible to do, yes? May 05 18:51:17 We could conceivably do something super fancy to distinguish between stat and other calls, or something. May 05 18:51:49 Custom fds don't need /dev, i don't think? N>&M, etcetera. May 05 18:52:03 Although the "make this fd look like a file" thing probably relies on /dev/fd/* working. May 05 18:52:12 How bash manages these fd's and thus requires access to /dev/fd/62 is unknown to me May 05 18:52:37 Huh. I wouldn't have thought it would, except for... there's a bash-specific syntax for "pass a filename which corresponds to this fd". May 05 18:52:46 But that's insane and you shouldn't use it. :P May 05 18:53:15 Did you see my post above? May 05 18:54:39 That it is this bash construct which makes it fail: "while { read -r line; } <&3; do" and "done 3< <(cat "$in")" May 05 18:54:44 huh. <&3 shouldn't require/dev/fd3. I don't know what "3<<(...)" means, but I think that's a bashism. May 05 18:54:52 It is May 05 18:55:22 What does that even do? Like, how is that different from 3<$in May 05 18:55:50 https://www.gnu.org/software/bash/manual/html_node/Process-Substitution.html#Process-Substitution May 05 18:56:28 its possible the easy fix is to rewrite the script to not use it :) May 05 18:56:59 okay so, that's equivalent to doing cat "$in" > file [...] ... 3 which means it should be equivalent to 3<$in May 05 18:57:13 which shouldn't need the fancy thing May 05 18:57:25 yeah, I think so too in this case May 05 18:57:29 and <&3 shouldn't require /dev/fd/* May 05 18:59:09 But it does imply that there are limitations to bash when in yocto/fakeroot May 05 18:59:59 yup May 05 19:00:12 and yeah, confirmed that i can do that as long as i don't attempt the process substitution thing May 05 19:00:35 a general fix for that is surprisingly hard. We have to actually use the in-tree path for most operations, like stat and so on, or mknod, or we can't actually do what we're trying to do. May 05 19:00:52 it's just for actual opens that we suddenly want the path to actually get to a real raw device. May 05 19:00:53 hmm. May 05 19:01:18 I'm running 'while { read -r line; } <&3; do' and 'done 3<"$in"', and that didn't help May 05 19:01:50 forget it.. host tool leakage May 05 19:03:14 * kergoth grumbles May 05 19:05:35 The unelegant failure with this in bash is that it complains about missing /dev/fd/62 and then proceeds the loop as if the file is empty. May 05 19:06:06 I spend a few hours on this today trying to figure out why our installer misbehaved :P May 05 19:10:16 Are you always separating metas from the actual sources, even if the sources and the meta/recipes are internal to your organization? May 05 19:12:25 I'm kinda curious, because we've tried having separate recipe with a meta layer from the actual application code, but it failed with loss of synchronization on a multiple occation. May 05 19:13:50 So we've combined them, but writing recipes is surprisingly hard. It's apparent that yocto isn't tailored for this IMHO. May 05 19:17:49 isn't tailored for what? May 05 19:18:56 having (large) sources alongside the recipies May 05 19:20:05 What is the best way to assign a static ip in my image? I have done a bbappend to interfaces.bb May 05 19:20:24 it makes up from a quite long list of SRC_URI, doesn't it? May 05 21:33:00 Does the lowercase 'conffiles' in base-files do anything? I can't find a user of that variable. May 05 21:38:59 RP: nobody answered that? May 05 21:39:07 (the gettext kernel thing) May 05 22:04:42 kanavin: no May 05 22:31:04 I cannot seem to override the interfaces file in openembedded-core/meta/recipes-core/init-ifupdown/init-ifupdown-1.0. Does anyone know of the correct way to do this. I am trying to make a static ip in my image. I can do it by changing the file directly, but I want to keep it separate in my custom recipe as opposed to changing the source. May 05 22:34:57 riz_: I think you could add an init-ifupdown_%.bbappend in your layer, then add a new /init-ifupdown/init-ifupdown-1.0 file May 05 22:37:17 clsulliv: Yeah, I tried that. I wasn't sure what to put in the append file. I basically copy an dpasted what was in the original file May 05 22:37:39 And then I put my interfaces file in /init-ifupdown/init-ifupdown-1.0 as you said May 05 22:38:39 riz_: you might need to add to the FILESEXTRAPATHS variable, but besides that I don't think you should need anything in the bbappend May 05 22:39:09 ahh. Sounds about right May 05 22:39:16 Will try. Thanks! **** ENDING LOGGING AT Sat May 06 03:00:04 2017