**** BEGIN LOGGING AT Wed Mar 25 02:59:57 2020 Mar 25 07:52:08 Hi, its me again :) i wrote a story for replacing that Makefile with CMake or at least clean it up a lot. Hope somebody is happy here. Mar 25 07:54:42 I managed to tell the Makefile dynamicly that it should use the library path. But now i ran into something else. : http://dpaste.com/0XNM7WE Mar 25 08:18:34 Guest5213: sounds like its missing/filtering the sysroot somewhere. Mar 25 08:24:51 New news from stackoverflow: Yocto does not install files to /usr/share Mar 25 08:37:16 LetoThe2ndq strange :/ Mar 25 08:39:55 hi guys. I'm getting this error in my build: "ERROR: Postinstall scriptlets of ['docker-ce'] have failed. If the intention is to defer them to first boot,then please place them into pkg_postinst_ontarget_${PN} ().Deferring to first boot via 'exit 1' is no longer supported." Mar 25 08:40:10 is there a way to get more debugging info about the reason why it failed? Mar 25 08:43:28 (found it in the logs update-rc.d: /home/ubuntu/nuvlabox-os-raspberrypi/tmp/work/raspberrypi3-poky-linux-gnueabi/core-image-base/1.0-r0/rootfs/etc/init.d/docker.init: file does not exist. thanks anyway) Mar 25 08:49:23 http://dpaste.com/05A4QED Next one. Mar 25 08:52:23 undefined references. Now its getting strange. Mar 25 08:53:38 that can be missing dependencies, cc versus cxx... Mar 25 09:07:23 LetoThe2nd ok. Mar 25 10:13:53 Guest5213: btw, you mentioned you "wrote a story"? Mar 25 10:53:34 Hi there! Mar 25 10:53:34 I'm wondering if I can place build directory inside my meta- directory Mar 25 10:53:34 e.g: oe-init-build-env meta-mymeta/build Mar 25 10:54:08 copycat_88: technically yes, practically it is massively discouraged. Mar 25 10:54:14 copycat_88: what would the usecase be? Mar 25 10:59:14 Well, I'm trying to optimize my CI for Yocto, and want to keep in git only files that was really managed by me. It's meta- directory and build/conf/local.conf. As I know, there is no way to move some essential variables from build/conf/local.conf to the machine configuration file, so I thought that I can move whole build/conf/local.conf insise my meta- Mar 25 11:00:20 copycat_88: if you're keeping local.conf in the repo, you're doing it basically wrong anyways. Mar 25 11:00:57 copycat_88: have a look at what kas does, for example. Mar 25 11:02:43 hi guys. I'm struggling to understand the class poky/meta/classes/update-rc.d.bbclass. If I understand it correctly, in short, if systemd is enabled, then run "update-rc.d ". And as far as I know, update-rc.d will fetch the INITSCRIPT_NAME which in my case seems to be docker.init. But the question is: "why is it docker.init?" cause Mar 25 11:02:44 if I look at meta-virtualization at https://git.yoctoproject.org/cgit/cgit.cgi/meta-virtualization/tree/recipes-containers/docker/docker-ce_git.bb?h=zeus#n105 , docker.init should not be installed *and in fact it is not...that's why my build is failing with """core-image-base-1.0-r0 do_rootfs: Postinstall scriptlets of ['docker-ce'] have failed . Mar 25 11:02:44 update-rc.d: /home/ubuntu/nuvlabox-os-raspberrypi/tmp/work/raspberrypi3-poky-linux-gnueabi/core-image-base/1.0-r0/rootfs/etc/init.d/docker.init: file does not exist """ Mar 25 11:03:20 pushing build into the layer essentially leaks/mangles your build infrastructure with the meta data, and that should be avoided. options are either to make up a combined distribution, like khem's yoe-distro or rburton's example do, or using a build automation like kas, or to some extent repo Mar 25 11:03:35 LetoThe2nd: thank you. I suspected that I was doing something wrong. Going to look at kas Mar 25 11:03:51 but the thing is, why is INITSCRIPT_NAME set to docker.init? in https://git.yoctoproject.org/cgit/cgit.cgi/meta-virtualization/tree/recipes-containers/docker/docker.inc , the INITSCRIPT_NAME should be set to '' right? Mar 25 11:05:18 unless I have to explicitly tell the build not to consider sysvinit? Mar 25 11:05:50 cjdc3: here you go https://git.yoctoproject.org/cgit/cgit.cgi/meta-virtualization/tree/recipes-containers/docker/docker-ce_git.bb#n114 Mar 25 11:06:22 LetoThe2nd yes but I shouldn't end up in that instruction Mar 25 11:06:32 cause I have systemd distro appended Mar 25 11:07:25 cjdc3: and systemd is actually in use and functional? Mar 25 11:07:58 LetoThe2nd I'm "almost" sure. It builds well without the docker-ce package Mar 25 11:08:13 cjdc3: where did you do the append? Mar 25 11:08:26 conf/local.conf Mar 25 11:08:40 DISTRO_FEATURES_append = " systemd virtualization"DISTRO_FEATURES_remove = "x11 wayland"VIRTUAL-RUNTIME_init_manager = "systemd" Mar 25 11:08:59 DISTRO_FEATURES_append = " systemd virtualization", DISTRO_FEATURES_remove = "x11 wayland", VIRTUAL-RUNTIME_init_manager = "systemd" Mar 25 11:09:09 ok, then i'd say do bitbake -e docker-ce | less and have a thorough look. Mar 25 11:12:09 LetoThe2nd INITSCRIPT_NAME_docker-ce="" Mar 25 11:13:24 and DISTRO_FEATURES? Mar 25 11:13:57 ok I found something Mar 25 11:14:29 if I don't set DISTRO_FEATURES_BACKFILL_CONSIDERED = "sysvinit", INITSCRIPT_NAME_docker-ce="docker.init" Mar 25 11:14:32 and that will fail Mar 25 11:15:00 which is acutally stated explicitly in the manual, right at the point where switching to systemd is explained. Mar 25 11:15:15 _remove works only without spaces IIRC Mar 25 11:15:36 not exactly sure but I definitely remember there was something about perfect match Mar 25 11:16:04 but it is not consistent Mar 25 11:16:07 and i'll check to correct myself if needed :) Mar 25 11:17:27 cause the docker-ce recipe will not populate if systemd is in DISTRO....while the INITSCRIPT_NAME will be set based on sysvinit Mar 25 11:19:19 I am trying to use PR service for auto incrementing the version no on changing source Mar 25 11:19:47 However i am not getting the verison no updated in .ipk files generated .. Mar 25 11:20:24 What are the entries to be added in .bb files and build/local.conf files for PR service to work Mar 25 11:20:38 I think a better condition for https://git.yoctoproject.org/cgit/cgit.cgi/meta-virtualization/tree/recipes-containers/docker/docker-ce_git.bb?h=zeus#n105 would be to check if systemd in DISTRO and sysvinit NOT in DISTRO Mar 25 11:20:45 wouldn't your agree LetoThe2nd? Mar 25 11:21:54 cjdc3: not necessarily. my take is that it should be consistent with other similar places. (which i do not know nor intend to check at the moment, sorry) - so if a similar situation is already handled otherwise in another place then you are right, otherwise not. Mar 25 11:23:06 ultimately this conditions is tightly coupled with https://git.yoctoproject.org/cgit/cgit.cgi/meta-virtualization/tree/recipes-containers/docker/docker.inc#n56, so the conditions should be similar Mar 25 11:23:15 but ok Mar 25 11:24:54 cjdc382: if in doubt, send a request for clarification to the mailing list and the maintainer. Mar 25 11:25:03 it might have a proper reason, it might be a bug. Mar 25 11:25:25 New news from stackoverflow: Bitbake: How to only fetch the sources? Mar 25 11:27:01 LetoThe2nd (y) Mar 25 11:30:21 Hi I see that you are talking in area where I also have some problems Mar 25 11:31:54 basically for some reason I don't see login prompt in my initramfs image, I set _login_manager ?= "busybox", _init_mamager ?= "sysvinit" and initscripts =? "initscripts" in my config Mar 25 11:32:23 last line on boot is: [ 3.346647] Run /init as init process Mar 25 11:33:26 when I changed to systemd I had the same problem, So kernel prints were visible, prints from systemd too and last line was something like "Reached multiuser ..." Mar 25 11:34:05 Does anyone know how to fix this or how to trace such problem ? Mar 25 11:35:19 rokm: if i had to guess, then the login console is set to another tty than you expect. Mar 25 11:35:43 it should be in /etc/inittab right ? Mar 25 11:36:57 maybe, i haven't used such in a long time, and never since i'm using yocto Mar 25 11:59:51 rokm: not sure, but have a look at SERIAL_CONSOLE(S?) Mar 25 12:00:47 SERIAL_CONSOLES = "115200;ttyS0 115200;ttyS1" Mar 25 12:01:30 like I wrote I see dmesg logs Mar 25 12:02:36 rokm: these are two different things. Kernel uart and userspace uart can be different Mar 25 12:02:54 try to check on other uarts Mar 25 12:03:07 but I run this on qemu Mar 25 12:03:52 these settings I have in machine.conf for qemu-x86 Mar 25 12:08:47 Could it be related to this print: [ 3.208231] Warning: unable to open an initial console. Mar 25 12:13:27 rokm: looks like /dev/console is missing in your image. I'm not 100% sure, but probably you need to add initramfs-framework to your image Mar 25 12:14:13 for now I added mknod console.... to do_install() in my initramfs package Mar 25 12:16:24 yep that was that :) Mar 25 12:16:35 I'm wondering what's the policy about PACKAGECONFIG defaults upstream? (wrt latest patches for sssd on the mailing list) do we aim small featureset or full featureset or "whatever the first committer decided"? Mar 25 12:18:19 "whatever the first committer decided" as long as all the dependencies are satisfied with the layer and layer depends where the recipe is Mar 25 12:22:19 JaMa: alrighty, thanks Mar 25 13:11:45 LetoThe2nd: just want to say thank you for kas suggesting, impressive tool. Completely covered all my needs Mar 25 13:12:12 copycat_88: YW Mar 25 13:21:06 i've got one recipe to create for building one binary for STM32F0 (cortex-M0, no MMU), needs its own toolchain and some additional libs from stm. I find multiconfig (if even possible, I merely found an example in meta-zephyr of one board running on m0) quite overkill for that. What about putting everything inside a recipe? What are your suggestions? Mar 25 13:21:33 qschulz: overkill! overkill! Mar 25 13:21:43 LetoThe2nd: When would you like me to do a guest live stream about building with containers? Mar 25 13:21:57 JPEW: whenever you feel like. Mar 25 13:22:18 I remember discussing this with one or two people at ELCE last year and I remember some people saying I'd be better off building it myself and putting it pre-built in a recipe but I'd rather have more people sharing their thoughts on that Mar 25 13:22:18 It's every Tuesday? Mar 25 13:22:30 JPEW: every Tuesday except when it's not :D Mar 25 13:23:14 JPEW: no. the initial plan was every second tuesday a month. but since i think october i had to skip randomly, and march+april are being dropped for personal reasons. Mar 25 13:23:33 JPEW: in a nutshell, its like this: fix a date, announce it a couple of days before, done. Mar 25 13:23:57 LetoThe2nd: lol, OK. I'll have to figure out this twitch thing that all the kids rave about ;) Mar 25 13:24:28 JPEW: there is no hard rule at all when it has to happen. of course longer announce timeframes as well as something compatible with EU+US timeframes will give you more attendance, but thats about it. Mar 25 13:26:48 LetoThe2nd: we talked about that right? I read my notes again and I might have drunk one too many beers because I wrote "Check Tuesday talk on Zephyr" but read the slides and.... either I'm missing something or it's not that helpful Mar 25 13:26:48 JPEW: i'd start by 1) getting a twitch account 2) installing obs 3) do some testing and tinkering. Mar 25 13:27:07 qschulz: the latter. "its not that helpful" Mar 25 13:27:20 LetoThe2nd: Ok, Thanks. Mar 25 13:28:52 LetoThe2nd: then what's your opinion? put all the shit into one big mess of a recipe? Mar 25 13:29:28 qschulz: nope. my vote clearly is on multiconfig. Mar 25 13:30:32 LetoThe2nd: aaaaah I thought you said overkill! in the abonding sense not in the sense of telling me to go the overkill way Mar 25 13:31:19 qschulz: you obviously didn't get the reference. Mar 25 13:31:20 https://www.youtube.com/watch?v=MlzTET_8SQg Mar 25 13:31:22 then question is, how well does multiconfig work with an m0 no-mmu machine with custom "libc" and custom toolchain Mar 25 13:31:40 qschulz: look at the newlib stuff Mar 25 13:31:47 LetoThe2nd: working on growing the beard but not there yet Mar 25 13:51:01 I have a problem with a warrior build, with python3-native being compiled with bmi1 extension, thus failing with "illegal instruction" on our Xeon X5650 CI server - does that ring a bell to anyone ? (that's a 2.7.1 release, but I can't see anything new in 2.7.3 that would impact this) Mar 25 13:51:26 illegal instruction being a "andn" in this particular case Mar 25 13:54:45 yann: python3 or python2? you said python3-native but then 2.7.1 Mar 25 13:55:59 New news from stackoverflow: What’s the purpose put a .so file before execution file? Mar 25 13:57:22 python3 on poky 2.7.1 :) Mar 25 14:04:05 I don't even see a -march=... in python3-native/3.7.2-r0/temp/log.do_compile that could explain the use of this extension Mar 25 14:05:16 how can I prevent generation of -src package of some recipe? Mar 25 14:06:54 I'm trying to create a patch according to https://wiki.yoctoproject.org/wiki/TipsAndTricks/Patching_the_source_for_a_recipe but when I finish with devtool and try to bitbake it, I get "can't find file to patch" Mar 25 14:07:14 what am I/devtool doing wrong here Mar 25 14:10:00 dv|2: what's the actual issue you have? Mar 25 14:10:21 stuom1: exact commands you ran and what you did Mar 25 14:10:34 LetoThe2nd scrum Mar 25 14:10:44 stuom1: otherwise it's hard to help Mar 25 14:11:01 qschulz, I have several closed-source recipes. don't want to generate -src package for them. Mar 25 14:12:27 Guest5213: so nothing we could share for your fame? Mar 25 14:12:43 previously I made PACKAGES = "${PN}-closed ${PN}-dbg ${PN}" and did "FILES_${PN}-closed="usr/src/debug*" in my recipe, but with Zeus it is not working Mar 25 14:13:11 qschulz: devtool modify , modifying CMakeLists in workspace, bitbaking as a test from workspace all fine, devtool update-recipe , devtool reset . Then removing workspace and bitbaking, and patch cannot find CMakeLists Mar 25 14:14:02 LetoThe2nd not yet Mar 25 14:14:14 dv|2: but the src packages aren't actually installed on your image right? Mar 25 14:14:21 I don't understand the issue actually Mar 25 14:14:23 Guest5213: think about it! Mar 25 14:14:26 something to do with CMakeLists not being in the root of that github repo but one folder deeper, probably, but dont know how to tell that to devtool Mar 25 14:14:38 dv|2: https://www.yoctoproject.org/docs/latest/ref-manual/ref-manual.html#var-PACKAGE_DEBUG_SPLIT_STYLE Mar 25 14:15:08 dv|2: -dbg and -src packages are automatically generated without any input from the recipe, so if you want them to not be, set that Mar 25 14:15:25 qschulz, right. But I do rsync of all packets from pfx.../rpm to my server. Don't want that packets to appear or make that packets empty during generation time Mar 25 14:16:32 i have to ask again Mar 25 14:17:00 rburton, is PACKAGE_DEBUG_SPLIT_STYLE may be used per-recipe? Mar 25 14:17:05 dv|2: yes Mar 25 14:17:09 ideally its a distro wide thing Mar 25 14:17:09 http://dpaste.com/05A4QED any suggestion Mar 25 14:17:11 s ? Mar 25 14:18:02 Guest5213: could be linker order, we pass --as-needed which means if the link order is wrong you get libraries removed before the symbols that use them appear Mar 25 14:18:05 stuom1: I would think patches are applied on top of ${S} Mar 25 14:18:41 so if the path of your patch relative to ${S} is wrong, that won't work? Mar 25 14:19:09 stuom1: and just to be sure, it's an actual error in do_patch and not that the change didn't make it to the sources right? Mar 25 14:19:15 It seems unlikely that I am the first person to attempt to build a custom cpython module with pokey but I can't find any decent prior art. There are four recipes in my sumo tree, but they all point to github for their source and perhaps use easy_install instead of setup.py Mar 25 14:20:47 The work directory fills up with thousands of files when I 'inherit distutils3-base ' in the recipe for my two-file module. Mar 25 14:21:01 It's a bit of information overload. Mar 25 14:26:11 qschulz: yes error is from do_patch. S = "${WORKDIR}/git/source", and CMakeLists is in source folder. Can I change the path of the patch? Or what should I do? Mar 25 14:41:54 rburton: looks like target compiler flags leaking to host binaries, or something close enough Mar 25 14:44:28 qschulz: to answer my own question, setting ;patchdir=.. after the patch file did the trick Mar 25 14:45:07 Hi, I support a compagny layer and would like to be notify if there's some recipes that point to internal repository that needed to be updated. Do you guys know if there's a feature within yocto that could do the trick? Mar 25 14:45:20 rburton i use LD line from yocto but before we just used $CC as linker Mar 25 14:45:53 arm-fslc-linux-gnueabi-ld: cannot find -lgcc -.- i dont get it Mar 25 14:49:25 stuom1: indeed but that is weird, how is a change to this git/CMakeLists required but the S directory where the configure and compile tasks should be run is a subdir? Aren't you "calling" this Cmakelists directly? anyway, glad it's fixed :) Mar 25 14:49:50 dk78: would it be ok if it worked the other way around? Mar 25 14:50:18 The recipes automatically get the latest code from the upstream repos? Mar 25 14:51:46 https://pastebin.com/VFTGFQjk : looks like a pretty bad value of PYTHONPATH for an host run Mar 25 14:52:16 dk78: there is a thing called AUH I think (auto-upgrade helper)? Mar 25 14:52:52 @kriive the only issue i see is how could I track the changes for the changelogs? Mar 25 14:53:12 qschulz I'll check that Mar 25 14:54:25 hm, for this it may be that 099d627 in 2.7.2 deals with the issue Mar 25 14:54:32 dk78: https://www.yoctoproject.org/docs/latest/mega-manual/mega-manual.html#var-UPSTREAM_CHECK_GITTAGREGEX and the two below are used here and there. Don't know if it's related but here it is Mar 25 14:55:41 cool thanks for your input Mar 25 14:56:26 qschulz and kriive think it will do it with all that! Mar 25 14:59:51 My task is so simple, it's got to have been done before. I'm looking at some other python libraries that inherit the pypi.bbclass file; this causes their source to be pulled from pypi.python.org Mar 25 15:00:50 But my two files are just local to the recipe, so it wouldn't make sense to inherit pypi and override everything about it Mar 25 15:10:25 rburton, I add PACKAGE_DEBUG_SPLIT_STYLE = "debug-without-src" and everything remains the same. Mar 25 15:11:44 Hi guys, if I Mar 25 15:13:44 HI if update a package thats is included in my rootfs but the root fs is not rebuilding it nor reconstrukting the rootfs tarbal Mar 25 15:13:51 it looks like libgcc.a is generated too late... is this possible? Mar 25 15:14:44 Guest5213: if you don't have your dependencies set straight, then yes. Mar 25 15:15:04 LetoThe2nd what is the dependency for libgcc ? just libgcc? Mar 25 15:15:51 @hmw1 If you clean you receipe or increment the PR it should rebuild the rootfs with the new package Mar 25 15:16:00 Guest5213: i think so, yes. Mar 25 15:16:50 PR="+git${SRCPV}" so if it changes in git it should recomplile but it wont Mar 25 15:17:40 LetoThe2nd i added libgcc as depends. seems to work but i still get the same error. Mar 25 15:18:37 Guest5213: and you are sure it is not messing up some pathes again? Mar 25 15:18:55 hmw1 in you work directory can you see if it's really updated? Mar 25 15:18:57 i double checked the paths ... Mar 25 15:19:02 but iam not 100% sure Mar 25 15:23:32 LetoThe2nd there is something really fucked up there. If i use GCC as linker (AFAIK that should just call LD) then it works! Mar 25 15:23:35 Theres a binary Mar 25 15:24:31 Hehehe. Yay for handwritten magic. Mar 25 15:24:44 dk78: to be shure i dit a cleanall of the package and after that i did a rebuild of the rootfs ( no git pull nor build did happen) Mar 25 15:24:57 i only have clean logs in the dir of the package Mar 25 15:26:23 the git string in dir name of the package is the last i pushed to git Mar 25 15:26:49 * the git string is the same in the dir name and last 1 in git Mar 25 15:27:37 LetoThe2nd it works, it feels 'dont touch it, then its fine' Mar 25 15:27:57 *like Mar 25 15:28:49 Guest5213: in a nutshell, it is just in the canonical Makefile state. Mar 25 15:29:02 qschulz: : GOT IT! Mar 25 15:29:07 i will try to replace it with cmake... Mar 25 15:29:38 besides the WORKDIR issues we discussed, the key was importing setuputils3, not distutils Mar 25 15:32:14 The recipe that builds it is ~5 SLOC Mar 25 15:47:20 ecdhe: nice! Mar 25 16:08:14 LetoThe2nd actually its very easy to integrate a plain Makefile project into yocto... If you try to keep your makefile clean Mar 25 16:08:48 i need 35 (formatted) lines in my recipe Mar 25 16:10:10 Guest5213: cool. compared to the, um.... 5 in my cmake recipes, a clear winner! Mar 25 16:10:22 *grins* and puts on a big IRONY tag. Mar 25 16:10:39 i have more then 5 lines in whitespaces :) Mar 25 16:11:06 but i agree with you, this thing needs to be fixed / or switch to cmake Mar 25 16:11:35 sadly cmake is not installed on our machines, i would like to introduce that without saying a word :) just merge and it works Mar 25 16:11:49 Guest5213: don't fix it if it ain't broken (anymore) Mar 25 16:12:32 Guest5213: but for new project i strongly encourage to start out with cmake, or meson, or whatever. just don't hand-craft s**t. Mar 25 16:13:08 anyway i need to get rid of round about 50 lines of unnecesseray variables in that Makefile Mar 25 16:14:05 hand crafted is worse then no build system.. cmake is terrible.. but acceptable Mar 25 16:14:11 (I loathe cmake) Mar 25 16:14:25 and i need to find out why we use CC for everything Mar 25 16:15:07 fray actually this is what i learned: every Buildsystem is PITA . But if you chose one then it should be fine. Mar 25 16:15:35 In my opinion, Makefiles would be ok if we would work with a little bit standardized Makefiles Mar 25 16:15:44 fray: it gets the job done for me, hence i use it. since i've switched to cmake i didn't have a single crosscompilation-induced headache anymore. Mar 25 16:15:53 my problem with cmake is that it tries to "know too much" Mar 25 16:16:22 it really shouldn't have any knowledge (internally) of the system configuration.. it should use the stuff the program or user provides and just perform actions/scheduling.. Mar 25 16:16:32 i also used SCONS in a project... It works but then somebody hacked a lot of python around it and it worked but nobody knowed why :) Mar 25 16:16:46 the number of times over the past 10+ years I've had to fight with cmake because it though something was right that clearly wasn't is too numerous to count Mar 25 16:17:10 (and don't get me started on 'UI configuration interfaces built with cmake' ... please make it stop) Mar 25 16:17:29 fray: we have built our own build system in our cmake recipes, so yes... too many possibilities (and imo, not the good ones :) ) Mar 25 16:17:38 i thought qmake also worked very well :) Mar 25 16:18:12 everything is bad anyway Mar 25 16:18:18 I like standard make syntax myself.. but that then requires the 'GNU' approach of generating makefiles or complicatign things by having dynamicly generated stuff.. which has it's own raft of problem Mar 25 16:18:21 qschulz some of my former colleagues joined later to that scons project. He just wrote a buildsystem wich created Makefiles with Python scripts Mar 25 16:18:32 I think we dont need more buildsystems Mar 25 16:19:01 qmake has pretty poor cross compile defaults Mar 25 16:19:04 imho Mar 25 16:19:06 actually i didnt like autotools. That was a lot of work Mar 25 16:19:27 did anyone triend meson? I have yet to write something with it Mar 25 16:19:33 try* Mar 25 16:19:34 but cmake worked for, its very fast in setup. If you dont want too specific stuff Mar 25 16:20:51 yeah, thats the point. Mar 25 16:21:15 I have a generic directory full of files that are the result of an external build process. I want to copy them to my image without naming them, but unfortunately, bitbake complains that I need to list all the filenames in "FILES-{PN}" Mar 25 16:21:20 Is there a way around this? Mar 25 16:21:22 and lets be honest: 95% of everyday work is the run-of-the-mill stuff, and not the highly complicated situation. Mar 25 16:21:46 ecdhe: just add the containing directory to FILES_${PN} Mar 25 16:22:45 LetoThe2nd what this, really complicated, Makefile does is: Take c,cc make objects, link objects to one binary. Done Mar 25 16:22:58 you cant do something more basig Mar 25 16:23:01 basic Mar 25 16:25:06 ecdhe - specify the files using globbing if you don't know the names.. or as Leto says, the directory they are contained in.. Mar 25 16:25:26 containment is key! Mar 25 16:25:33 containment is key! Mar 25 16:25:34 Mar 25 16:25:35 Makefiles is weird. They are too weird for such simple tool. This tinkering with correct spacing, no way to use loops.. Mar 25 16:25:49 thanks fray, building now, we'll see if it works! Mar 25 16:29:02 But if you want to see real weird sh*t - look into openjdk-8 build scripts. And then into meta-java. I spended 2 full weeks trying to create bitbake recipe for openjdk-8, and this is the craziest build system ever Mar 25 16:29:23 copycat_88: i just don't look at anything java. Mar 25 16:29:25 problem solved. Mar 25 16:30:12 aaaaah java Mar 25 16:33:07 LetoThe2nd: Yeah, I would like to be able to do the same.. Sometimes customers doesn't know anything besides java, and resulting image is: 90% JDK and 10% of kernel + busybox +apps ¯\_(ツ)_/¯ Mar 25 16:34:15 worked on a project once that was a safety critical device that was 90% java + 10% kernel.. god that was a scary piece of equipment Mar 25 16:35:01 robotics to move oil drilling pipes.. I sure hope they planned for the sudden stop when java decided to garbage collect Mar 25 16:38:33 nodejs on an embedded product because the startup only had one developer and they only knew nodejs... Mar 25 16:38:58 Would be funny if oil prices blew up because of an OOM kill, just before the GC striked Mar 25 16:39:32 Oil price up because Java, nytimes head line Mar 25 16:39:36 kriive less funny, these were deployed all through the Gulf of Mexico on drilling rigs in the early 2000s Mar 25 16:39:54 omg Mar 25 16:39:56 (I had nothing to do with the Java part BTW) Mar 25 16:40:28 yeah yeah everyone says so Mar 25 16:41:23 ...fromt he original press release... "next-generation oil rig controller, its e-Drill product, promises equipment that has a far higher level of automation and functionality (through sophisticated robotic algorithms); is simpler to set up and maintain (through thin-client, web browser-based "screens"); and leverages the Internet to allow remote diagnostics, monitoring and control of these systems." Mar 25 16:42:15 "The first systems ... are operational today in the North Sea and the Gulf of Mexico ..." (2001) Mar 25 16:58:27 I have a question about custom layer... Is it correct put my custom images in recipe-core/images or should I put in other place? Mar 25 17:03:21 I always put them there, I think recipes-core/images is the established standard Mar 25 17:04:38 thanks because I see on yocto channel in youtube that someone use images folder in root of meta-custom Mar 25 17:04:50 I'm improving my skills on yocto :) Mar 25 17:08:57 guerinoni: They are just recipes so you can put them any where you would like, as long as the layer.conf has the BBFILES set to find it, but modeling others helps people find what you are trying to provide. The standard location is recipe-core/images. A number of thematic layers place the recipes in somethig like recipes-/images. Similar with packagegroup recipes.. Mar 25 17:11:55 fray: heh, I worked on a telephony product that initially used Java on arm5 for SIP stuff. Real-time Java was/is a thing ;) Mar 25 17:12:33 fray: they eventually rewrote that part in C++, mostly to save on the licensing costs Mar 25 17:12:45 thanks I prefer keep standard path and method because is more clear for other work on that layer Mar 25 17:14:05 smurray ya.. it was 'thing'.. not a good thing.. but a thing Mar 25 17:16:29 fray: heh, indeed. I was glad to not be working on that part of the product Mar 25 17:32:31 Hi there, I'm triyng to make a recipe for scikit-sklearn python module, but i'm facing an error in the do_compile (ImportError: numpy is not installed.). However, numpy is already installed in my image, so i'm not sure where to look. Did someone already get this problem ? Mar 25 17:33:27 Looks like this guy had the same issue: https://stackoverflow.com/questions/59065058/yocto-create-recipe-for-scikit-learn , but this was not resolved Mar 25 18:01:06 yohboy: If you want it to be available during build time you need to add it to DEPENDS in the recipe Mar 25 18:01:22 I did it Mar 25 18:01:32 `DEPENDS_${PN} += " \ Mar 25 18:02:05 is it the correct way ? Mar 25 18:02:17 yohboy: No, just to DEPENDS Mar 25 18:02:55 The per-package is only for runtime dependencies (RDEPENDS) Mar 25 18:03:49 did not give better result, same error :/ Mar 25 18:06:22 yohboy: give the exact content of you change to DEPENDS and the error you see Mar 25 18:06:38 DEPENDS += "python3-numpy" Mar 25 18:08:02 https://pastebin.com/0Y2xXhpz Mar 25 18:09:29 Try adding python3-numpy-native too Mar 25 18:10:09 https://pastebin.com/csSCALTd Mar 25 18:10:12 my recipe Mar 25 18:10:15 ok i try Mar 25 18:15:14 yohboy: pretty sure erbo's right. What happens is that the python code run at compile time cannot find numpy. Since it's *executed* at compile time, therefore on the host system and not the target, you need the -native variant of numpy. Mar 25 18:15:40 erbo: well ok, native is ok, i have now some other error related to other library, but I'll add them :) Mar 25 18:16:08 qschulz: ok nice to know thanks ! Mar 25 18:54:53 looks like scikit-learn need also scipy Mar 25 18:55:38 there is not recipe for scipy too ? :/ Mar 25 19:01:39 yohboy: no there is not Mar 25 19:02:04 gonna be painful I guess Mar 25 19:02:51 I do have a somewhat hacked recipe Mar 25 19:03:25 i'm interested Mar 25 19:03:44 looks like it need openblas too Mar 25 19:04:45 yup and this need fortran compiler Mar 25 19:05:32 Openblas => https://raw.githubusercontent.com/intel/intel-iot-refkit/master/meta-refkit-extra/recipes-convnet/openblas/openblas_0.2.20.bb Mar 25 19:06:27 nice Mar 25 19:07:31 https://jumpnowtek.com/yocto/Add-Fortran-support-to-a-Yocto-build.html Mar 25 19:08:09 ... for the 'FORTRAN_forcevariable = ",fortran" ' in local.conf Mar 25 19:08:44 the fortran compiler is for openblas ? Mar 25 19:10:03 I do have FC (Fortran Compiler) referenced in my scipy recipe Mar 25 19:10:16 not the openblas recipe sorry Mar 25 19:10:41 ok Mar 25 19:10:51 I have an error in the openblas recipe then Mar 25 19:11:09 in oe_run_make Mar 25 19:11:23 I also have in my local.conf 'RUNTIMETARGET_append_pn-gcc-runtime = " libquadmath"' Mar 25 19:13:34 ok Mar 25 19:13:45 yocon: did you get this error https://pastebin.com/85mbBAMK ? Mar 25 19:13:57 with openblas Mar 25 19:14:29 No I did not :/ Mar 25 19:14:37 :/ Mar 25 19:16:47 My WIP scipy recipe : https://pastebin.com/Sg84rQ5K Mar 25 19:18:51 lapack replace openblas ? Mar 25 19:18:53 Hi guys, quick question Mar 25 19:19:30 yohboy: I kind of remember you need both? Mar 25 19:19:39 maybe yes Mar 25 19:19:50 you did not include openblas then ? I don't see it Mar 25 19:21:22 If I set CROPS ENTRYPOINT to entrypoint: ["/usr/bin/dumb-init", "--", "/usr/bin/poky-entry.py", "--workdir", "/opt/yocto", "--id", "1000:1000"], can I rely that a shell gets spawned inside /opt/yocto? Mar 25 19:21:37 can I rely on the fact* Mar 25 19:22:00 Because CROPS + Gitlab-ci is giving me tons of headaches Mar 25 19:22:01 You can find the OpenBLAS recipe I used in the link I sent from the intel-iot-refket repo Mar 25 19:22:13 are* Mar 25 19:22:22 well ok Mar 25 19:22:30 cross compilation seems broken though :/ Mar 25 19:22:49 same error in this thread https://yocto.yoctoproject.narkive.com/9fJuNa0I/scipy-recipe Mar 25 19:23:15 yeah sorry, "Batteries are definitely not included" Mar 25 19:23:52 well Mar 25 19:24:05 using scikit seems compromised :') Mar 25 20:31:04 one more question. I'm copying the contents of a source folder to destination on the target folder (no build step required, just some binary blobs). I have told git to ignore the contents of this folder, so when I clone my yocto layer, the folder isn't created, and when I go to build, bitbake complains that it can't glob that directory Mar 25 20:32:04 I could tell git to track a README file but I'd prefer to just tell bitbake to create the directory within the layer if none exists... is that possible, or total insanity? Mar 25 20:32:21 ... or can I have it try/except the glob failure? Mar 25 20:38:03 I could attempt to prevent the parsing of the recipe if no source folder is found, but this would cause no package to be built Mar 25 22:27:43 New news from stackoverflow: How to increase disk size when building yocto || How to cross-compile an application with library dependencies for Yocto Linux? Mar 25 23:18:38 RP: sent a binutils patchlet, which fixes Ross's problem of upgrading autotools, plus it should save some time during configure as we no more require to run autconf Mar 25 23:32:31 Where in the yocto project, we compile dtb, where is the rule recipe to create a dtb file. My problem is that the dtb getting generated doesn't have __symbols__ Mar 26 00:09:12 sj634, I think that is via the kernel Mar 26 00:10:01 I believe the device tree needs to be defined in the machine file or in your kernel recipe Mar 26 00:10:48 for example see: http://git.yoctoproject.org/cgit/cgit.cgi/poky/tree/meta-yocto-bsp/conf/machine/beaglebone-yocto.conf Mar 26 00:40:13 where is the recipe to convert dtc to dtb for e.g. - /opt/poky/2.7.3/sysroots/x86_64-pokysdk-linux/usr/bin/dtc -o dtb -o phyDriver.dtbo -@ ./phyDriver.dts /opt/poky/2.7.3/sysroots/x86_64-pokysdk-linux/usr/bin/dtc -o dtb -o phyDriver.dtbo -@ ./phyDriver.dts Mar 26 00:40:48 The problem is that my yocto project isn't generating the symbols file and I believe it is missing -@ Mar 26 00:41:12 So, I have to search in yocto to find where the dtb rule and add -@ Mar 26 00:41:43 Thus, I am unable to use my dtbo since there is no __symbols__ generated by the device tree blob. Mar 26 00:42:08 One more question is dtb platform independent ? Mar 26 00:49:48 Like where are the entire logs of yocto to know from where it is reading the dts file and what command it is running to generate a dtb file Mar 26 01:19:35 I can see the output file - ls -l ./tmp/deploy/images/microzed-zynq7/zynq-microzed-microzed-zynq7.dtb Mar 26 01:19:56 but I don't know how it is getting generated where is the microzed dts file? Mar 26 02:19:21 sj634, its part of the kernel build. look at kernel.bbclass or one of the other kernel bbclasses Mar 26 02:20:24 the file is generated during kernel build. the base files could be in a BSP layer that then patches the kernel Mar 26 02:20:37 also peek at kernel-devicetree.bbclass Mar 26 02:21:31 bitbake -e | grep KERNEL_DEVICETREE, might also give you a thread to pull on **** ENDING LOGGING AT Thu Mar 26 03:04:25 2020