**** BEGIN LOGGING AT Thu Jan 16 02:59:59 2020 Jan 16 05:29:03 New news from stackoverflow: php://input stream is empty on lighttpd and yocto based image || using EXTRA_OECMAKE doesn't have any effect in cmake variable in bitbake recipe Jan 16 07:39:29 Hello, I have on git repo with several lib sources inside, the question is : what is the best way to configure a recipe to choose which library we want to install on the target ? PACKAGECONFIG or something else ? Thx. Jan 16 07:44:18 PinkSnake: are they all mutually exlusive? Or just different libs doing different things? Jan 16 07:50:26 erbo Several "individual" libraries, nothing in common. I would like just to checkout once all the repo and maybe select which I want to install. I don't know if PACKAGECONFIG is made for that. Any idea ? feel free to ask if it's not clear :) Jan 16 07:57:28 PinkSnake: So the idea of not build all would be to save build time? Otherwise I would just build them all and put into different packages so it's easy to just pull in those you want in the images. Otherwise I guess PACKAGECONFIG could be used to just build the ones you actually want. Jan 16 08:07:00 erbo it's totally right. I have a repo like this : https://pastebin.com/2bzhpshB and I would like to do something like that: https://pastebin.com/ZcMn84W5. Looks dirty :( Maybe I have to configure a CMakelists/ autotools ? Doesn't matter what I have to do I would like someting proper :) Jan 16 08:26:14 there are few ways to handle such cases, one is packageconfig route and it seems you can add --enable/--disable options or some other component build system to provide knobs which can be instrumented via packageconfig Jan 16 08:27:38 other way is you write separate recipe for each one of the libs while checking out the source once in work-shared/ much like what gcc-source recipe does or linux-yoto recipe does as an example, Jan 16 08:28:29 suppose if your libs are shared objects and there are apps which link to them, then shlibs can automatically add the needed solibs to image once the app is added via IMAGE_INSTALL Jan 16 08:29:37 New news from stackoverflow: Difference between ldconfig and ldconfig.real Jan 16 08:29:46 so app <- libfoo.so and libfoo.so is built by libfoo.bb then only libfoo.bb will be built, in packageconfig approach whatever you enabled as a set of packageconfig all of that will build Jan 16 08:34:22 Thank you for advice, I'm not sure to understand how to made all-in-one recipe, so as the main trouble is not checking out all sources for each lib, i'm going to take look to clone in work-shared, but have a top Makefile and pass EXTRA_OEMAKE should to the trick alos isn't it ? Jan 16 08:36:09 PinkSnake: IIUC the backlog, you build all libraries in one recipe. Then you create a package for each lib (`PACKAGES +=` or even DYNAMIC_PACKAGES if you're feeling it (I never used it)), and set FILES_${PACKAGEXYZ} = "" for each package = lib Jan 16 08:37:18 PinkSnake: dependency at build time is on the whole recipe but rdepends is either pulled automagically by Yocto or you add an explicit RDEPENDS_${PN} += " " etc. Jan 16 08:38:09 qschulz Nice, that looks like the better idea! Thank you to all of you for your time ! Jan 16 08:38:42 PinkSnake: OR, if your source code is actually under git, you can use one recipe per lib and use "subpath" parameter in SRC_URI => https://www.yoctoproject.org/docs/2.6/bitbake-user-manual/bitbake-user-manual.html#git-fetcher Jan 16 08:41:44 qschulz Haa, as I have already do all recipes, "subpath" looks like a really good alternative to not waste to much time on that (y) Jan 16 08:43:06 well, you still will carry full .git folder with subpath Jan 16 08:44:02 khem: could you explain a little bit more, I actually considered subpath a few times but never used it, so eager to learn what are the consequences of such a parameter Jan 16 08:44:53 not sure how big your monorepo is, if its not that big I think using simple recipes are ok, since it would fetch whole repo once and thten untar it with recipes ( if you write multiple recipes ) Jan 16 08:46:55 qschulz khem In fact the repo is not really big, but as I made a refact of all Yocto project I just want to do that proper. I will try to test subpath and packageconfig to take a look Jan 16 08:49:09 qschulz: it uses git-read-tree Jan 16 09:00:00 khem: hey what are you8 doing in my timezone?!? Jan 16 09:09:38 khem: ok I need to do some reading first :) thanks Jan 16 09:43:00 Hi, I'm trying to build an old version of busybox that is failing with : In file included from applets/usage.c:11:0:| include/applets.h:268:17: error: 'login_trivial_usage' undeclared here (not in a function)| IF_LOGIN(APPLET(login, BB_DIR_BIN, BB_SUID_REQUIRE))..... Jan 16 09:43:12 scripts/Makefile.host:104: recipe for target 'applets/usage' failed| make[1]: *** [applets/usage] Error 1 Jan 16 09:49:13 ifan: if i had to guess, then i'd say broken config / compile state Jan 16 10:06:38 LetoThe2nd: rebooting the system, cleaning fixed the problem Jan 16 10:06:56 If you dont mind me asking, how could you see that Jan 16 10:07:28 ifan: intuition. it felt like parts had been compiled with one config, parts with another Jan 16 10:12:21 Okay, thanks again :) Jan 16 10:12:48 have fun! Jan 16 10:17:04 what does this mean exactly "Package version for package libgcc went backwards" ? Jan 16 10:18:04 angelo__: that you built the package at least twice, and PV was lower in the later build that in the former. Jan 16 10:18:07 (AFAIK) Jan 16 10:18:09 yes Jan 16 10:18:30 if you're jumping around releases and testing upgrades, it's normal and thats why i turn it off Jan 16 10:18:45 qschulz: devtool modify devtool build devtool target-deploy worked like a charm. the question is now how I can edit and commit my actual application. I found devtool finish which supposedly pushes my changes to the recipe and removes it from the workspace but hoiw do I actually make changes and commit them? also, when I open the devshell via bitbake I can edit the sources and then got from there but that seems awfull ycumbersame as I'd always have to open the Jan 16 10:18:45 devshell, edit, close the devshell, build, deploy to target, and then iterate this ... Jan 16 10:18:46 if you're not doing that then something went wrong Jan 16 10:18:55 rburton: cue: "jump around" - house of pain Jan 16 10:27:18 Yatekii: I don't use devshell, I don't even know what it's supposed to do :D Jan 16 10:28:09 when you did a devtool modify you should have gotten a path to where the sources are extracted Jan 16 10:28:48 so you go there and you commit things there? Jan 16 10:31:02 Why in my Yocto image */proc/config.gz* is missing? Jan 16 10:31:39 nacknick: probably its not set in the kernel config Jan 16 10:32:06 (as long as proc is mounted and not completely empty. but thats pretty much an edge case) Jan 16 10:32:51 LetoThe2nd: Why should I do to set it? Jan 16 10:32:54 what* Jan 16 10:33:09 its a kernel config like any other Jan 16 10:34:13 i've witnessed your struggles yesterday and sorry, i do neither have the time nor energy to get invested in that topic again. all i can do is give you the reason and where to start looking. Jan 16 10:36:36 I added `CONFIG_SECCOMP_FILTER=y` to *.scc* file like qschulz explained to me yesterday, it seems like the kernel was rebuilt, but I want to make sure this feature was added to the kernel. Without kernel's config file I don't know how to do it Jan 16 10:37:25 nacknick: https://wiki.gentoo.org/wiki/Kernel/IKCONFIG_Support Jan 16 10:38:02 nacnick: Look in something like `tmp/work-shared//kernel-build-artifacts/.config` after the kernel was built Jan 16 10:38:24 nacknick: ^^^ Jan 16 10:39:28 LetoThe2nd: Thanks. Actually I tried today to add `CONFIG_IKCONFIG_PROC=y` hopefully it will add the config file - but it didn't. I will read if there is something new in the link you sent Jan 16 10:40:18 paulbarker: I found it inside `kernel-source/tools/testing/selftests/seccomp/config:CONFIG_SECCOMP_FILTER=y` - but I want to see it in the final image on the RPI3 Jan 16 10:40:55 The .config file used in the build should be in kernel-build-artifacts Jan 16 10:41:25 Assuming you're on a recent yocto version Jan 16 10:42:34 No. I use yocto 2.4.4 Jan 16 10:43:14 paulbarker: "rocko" Jan 16 10:43:28 Ah. Have a look in the kernel workdir then Jan 16 10:44:38 nacknick: I suggest to check that config options do not depend on other options Jan 16 10:45:32 does yocto support/are the plans to support SRC_URI[sha512sum] ? Jan 16 10:46:20 milloni: doesn't seem to be used/supported in current state at least. any particular need/usecase? Jan 16 10:46:34 not really, just curiosity Jan 16 10:46:44 someone on my team suggested using it instead of 256 Jan 16 10:46:48 more bits aint bad :) Jan 16 10:47:26 *shrug* Jan 16 10:48:24 milloni: well it actually would double the length of the checksum in the recipes, which are pretty ugly already. Jan 16 10:49:20 LetoThe2nd: i see you're an aesthetician like me Jan 16 10:49:20 milloni: and given that its just for integrity detection (i don't think we need to fear hash collisions or forged archives), i can't actually see any benefit, only cost. Jan 16 10:50:29 LetoThe2nd: i thought it was, among other things, for integrity detection in the security sense, that is for protection against "forget archives" Jan 16 10:50:50 forged* Jan 16 10:51:14 milloni: sent a PoC that you can forge an archive that still compiles :) Jan 16 10:51:21 not that it is a likely thing to happen but it is in my threat model Jan 16 10:51:27 (exploiting a hash collision, of course) Jan 16 10:51:45 i cant with sha256sum Jan 16 10:52:04 no one can, as far as i can tell Jan 16 10:52:15 milloni: real world answer: if your threat model is so advanced that it cares about this, then you should certainly be able to push the needed changes upstream and maintain them. Jan 16 10:52:45 ack Jan 16 10:52:58 it's not so advanced Jan 16 10:53:12 my collegaue's argument was just "more bits aint bad" Jan 16 10:53:50 would be a trivial patch to support it as an option Jan 16 10:54:03 we should just have a number of hashing schemes and just ask that at least one is set Jan 16 10:54:15 milloni: Make him use signatures several GB in length. That will cure the "more bits aint bad" syndrome Jan 16 10:54:23 like i said. not bad, but they come at a cost. and at the moment probably nobody cares to pay the cost, how ever little it is. Jan 16 10:55:31 paulbarker: i think we just need to make a blockchain out of it ;) Jan 16 10:55:38 using AI, in the cloud, of course Jan 16 10:57:37 blockchain jokes as so 2010s Jan 16 10:57:42 s/as/are/g Jan 16 10:57:57 what's the new shiny technology of 2020s then? Jan 16 10:58:20 can't tell after two weeks yet. ask me the day after tomorrow, mkay? Jan 16 10:58:55 milloni: gimmie a minute to train an AI to name 2020s tech... Jan 16 10:59:26 oh hello ecclescake, i didnt know you were here Jan 16 11:34:39 Hello, I'm trying to create an image that includes another image (I'm making a USB install stick), but I cannot get the dependencies right. Jan 16 11:35:46 So I have two recipes . and I have tried adding to RDEPENDS in the recipe Jan 16 11:36:41 BoJones: thats no dependency usecase, but a multiconfig one Jan 16 11:37:03 LetoThe2nd, multiconfig .. Jan 16 11:37:17 yep Jan 16 11:37:19 Interesting .. I'll have a look Jan 16 11:39:48 because it's a pain to google anything related to "install" command, can somebody tell how to set the file executable with install? Jan 16 11:41:00 stuom1: install -m xxx takes the rwx mode as xxx :) Jan 16 11:41:31 stuom1: and "man install" in google gives me this as #1: https://linux.die.net/man/1/install Jan 16 11:42:04 so is it "install 644+x" then? Jan 16 11:42:25 stuom1: i guess you should rather already finish the calculation. Jan 16 11:43:27 i mean in setting excutable in chmod is "+x" Jan 16 11:43:42 you know what "644" means? Jan 16 11:44:28 LetoThe2nd, do you have an example on how to use multiconfigs to include another image ? My google-foo is not good on this one. Jan 16 11:44:29 ah yeah now i was mixing with the other format Jan 16 11:44:58 BoJones: not yet, sorry. Jan 16 11:45:07 LetoThe2nd, Ok Jan 16 11:50:40 hello.. I have added the meta-browser layer and depended on the chromium recipe. This successfully builds chromium. I have added 'IMAGE_INSTALL_append = " chromium-x11"' to my custom layer's layer.conf. The package, however, doesn't get installed in the resulting image: it is present in deploy/, but it is not installed. What else do I need to do? Jan 16 11:51:11 dexterlb: 1) stuff like this should not go into the layer.conf Jan 16 11:51:33 dexterlb: 2) check bitbake -e if it actually end up there Jan 16 11:51:55 dexterlb: 3) check your image if it actually honors IMAGE_INSTALL, there are ways around it. Jan 16 11:55:13 1) I'm sorry: in fact, it is in my image recipe (which inherits core-image). I don't know why I said layer.conf. Is that correct? Jan 16 11:55:13 2) it appears in many places in bitbake-e, including IMAGE_INSTALL and PACKAGE_INSTALL Jan 16 11:55:13 3) how do I do that? Jan 16 11:57:15 dexterlb: if its in the image file, shows up in bitbake -e and your image inherits core-image, then all should be fine Jan 16 11:58:25 but it is not installed (none of the files that are in the package appear in the target filesystem, and 'opkg list' doesn't show it) Jan 16 11:58:44 dexterlb: what does the manifest of the image say? Jan 16 11:59:11 LetoThe2nd: where do I find this manifest? Jan 16 11:59:24 dexterlb: right next to the image file Jan 16 12:00:15 the manifest contains the package Jan 16 12:01:11 dexterlb: that doesn't fit Jan 16 12:02:21 if the manifest contains it, then it is installed. chances are that the package either does not contain what you think (dpkg can be used to inspect .ipk contents) or you are looking at the wrong image. Jan 16 12:02:27 you're right, it doesn't.. I think I've done something utterly stupid, like inspecting the wrong image - let me verify Jan 16 12:04:24 qschulz: devshell opens a new shell and loads the env for your recipe and cds into it's source folder Jan 16 12:04:40 qschulz: I thought I was supposed to use it but i am more than happy if I don't need to Jan 16 12:05:06 qschulz: ERROR: output path /home/yatekii/repos/ecarup/build-openstlinuxtkrt-stm32mp1-raichu-cubemx/workspace/sources/raichu-core already exists and is non-empty // I did a devtool reset raichu-core and then devtool modify raichu-core ... Jan 16 12:05:46 ah hmm reset said: NOTE: Leaving source tree /home/yatekii/repos/ecarup/build-openstlinuxtkrt-stm32mp1-raichu-cubemx/workspace/sources/raichu-core as-is; if you no longer need it then please delete it manually Jan 16 12:05:55 notr sure what I am supposed to do with this info tbh Jan 16 12:18:11 * rburton finally timed a build on the server he never uses Jan 16 12:18:21 30 minutes for core-image-sato from scratch \o/ Jan 16 12:19:10 rburton: but hot download, i guess. Jan 16 12:19:31 right Jan 16 12:19:52 * LetoThe2nd is only half impressed, then. Jan 16 12:19:58 haha Jan 16 12:20:55 don't actually know how fast the connectivity is where it sits Jan 16 12:21:05 kernel downloads at 27M/s Jan 16 12:21:09 not bad. Jan 16 12:21:16 better than my home connection for sure Jan 16 12:21:19 by about 27 times in fact Jan 16 12:21:36 but seriously. if i would have been completely impressed, then you would have no mission in life anymore. Jan 16 12:21:57 so i noticed this build was blocking on gstreamer finishing so obviously i'm doing a clean build with the mesonification patches in Jan 16 12:25:01 Hi, i have a question regarding a recipes for a pypi python package, with "inherit pipy" is the recipes suppose to download also the dependencies automatically? Jan 16 12:25:41 for instance I'm trying to use python-sphinx in the meta-virtualization Jan 16 12:25:42 khem erbo FYI subpath works like a charm. Thh for the hint ! Jan 16 12:26:50 the recipe is very simple. inherit pypi, package name, checksums, licence and various variable for the descrition Jan 16 12:27:19 no RDEPEND Jan 16 12:27:51 Mat80: it probably should. why are you asking? Jan 16 12:28:13 the recipes build and create the package, but of course sphinx need some dependencies and if I try to run sphinx on the target machine it fails because of missing dependencies Jan 16 12:28:14 @Mat80 it doesnt Jan 16 12:28:32 at least not with the few pypi packages I have experience with Jan 16 12:29:11 Mat80: it does not download depends, you need to package those too Jan 16 12:29:12 bitbake doesnt complain but trying to run on target, you get import errors Jan 16 12:29:35 I'm a yocto newbie and this seems still a lot black magic...but I don't understand where the magic stops and where I should do something Jan 16 12:29:35 meta-python is your friend Jan 16 12:29:51 rburton what do you mean? Jan 16 12:29:59 meta-python is full of pypi recipes Jan 16 12:30:04 ah ok Jan 16 12:30:07 i'll have a look Jan 16 12:30:09 ty Jan 16 12:30:16 oh, just read again Jan 16 12:30:30 if its giving missing deps then the RDEPENDS needs improving in the recipe Jan 16 12:30:39 just add what it moaned about until it works Jan 16 12:30:52 Yatekii: well, remove it :)? but do it only after you have made sure that your changes are in the layers. I think devtool creates a bbappend with the patches btu I'm not 100% certain. Check if there's any change in any layer, check those are the ones you want, then you can get rid of the sources for raichu-core Jan 16 12:31:50 Yatekii: what devtool does is that it creates a new layer with new recipes whose source code is in workspace/sources. When you do devtool reset it "disables" this recipe but leaves you with the sources so that it does not remove something you wanted to keep Jan 16 12:33:15 nacknick: I don't know exactly how does one create config fragments but I can explain how *I* would do it. Jan 16 12:33:17 but can it be that a recipe on the virtualization layer build the package but doesn't work? What is the "QA" situation on the various openembedded meta layers? can be a layer "problem" or can be something in my configuration? Jan 16 12:34:26 I try to rephrase: should I trust what comes in the layer and search the problem somewhere else or not? Jan 16 12:34:27 Mat80: if its just an import error then its the recipe at fault. it assumes some python bits are present, and they are not. Jan 16 12:34:37 Mat80: most people forget that you can't rely on the full standard library Jan 16 12:34:49 ok ty Jan 16 12:35:03 so ensuring that your image has python3-modules installed gets that and may just work around the problem Jan 16 12:36:18 this is the error pkg_resources.DistributionNotFound: The 'sphinxcontrib-websupport' distribution was not found and is required by Sphinx Jan 16 12:37:03 nacknick: I would try to run first make savedefconfig to get the defconfig of the .config in use, copy that one somewhere safe, then run make menuconfig for the kernel. Find a way to enable CONFIG_SECCOMP_FILTER from there (hit "/" key and enter SECCOMP_FILTER. Then if you don't have the option to select it, work out the depends on and enable the needed ones. Once that's done, enable SECCOM_FILTER option. Jan 16 12:37:09 Exit and save. Then run make savedefconfig and put the diff between the two defconfigs in.scc file Jan 16 12:37:17 My guessing is that i have to create a recipe for https://pypi.org/project/sphinxcontrib-websupport/ Jan 16 12:40:02 nacknick: Kconfig options have a dependency mechanism (`selects` and `depends on`). If you're putting CONFIG_SECCOMP_FILTER=y but SECCOMP_FILTER actually `depends on` another disabled Kconfig option, then it's a no-op. You need to enable the depends on AND your Kconfig option. Jan 16 12:45:27 nacknick: from a VERY quick look, it looks like you should make sure that your arch supports it (i don't recall which rpi you're building an image for but if it's aarch64, then you're fine on that side) then check CONFIG_SECCOMP and CONFIG_NET are enabled. Check this with menuconfig honestly and if your SECCOMP is not enabled follow my process or a better one if there's one (neevr used kernel config Jan 16 12:45:33 fragments before so I don't know how's one supposed to create them) Jan 16 12:50:01 happy to hear if there's any other documented way to create fragments :) Jan 16 12:53:00 Can someone explain how a recipe like this can work at all? http://git.yoctoproject.org/cgit/cgit.cgi/meta-cloud-services/tree/recipes-devtools/python/python-pyparsing_2.0.1.bb Jan 16 12:53:40 Mat80: what bit confuses you? Jan 16 12:54:03 there is not even the name of the package to build Jan 16 12:54:05 pypi recipes take the package name from the name of the recipe Jan 16 12:54:26 stuom1 ah thats why Jan 16 12:54:28 pypi class uses PN to fetch Jan 16 12:54:31 and distutils does the build Jan 16 12:54:41 its missing RDEPENDS, which is naughty Jan 16 12:54:41 i'm really sorry for this stupid questions Jan 16 12:54:53 look at the pypi class, its really simple Jan 16 12:59:03 qschulz: armv7 Jan 16 13:04:16 I tried to run `bitbake -c menuconfig virtual/kernel` and last lines I get after run it are: `menubox.c:(.text+0x3c3): undefined reference to `wrefresh'collect2: error: ld returned 1 exit statusscripts/Makefile.host:116: recipe for target 'scripts/kconfig/mconf' failedmake[3]: *** [scripts/kconfig/mconf] Error Jan 16 13:04:17 1/home/user/yocto/poky/build/tmp/work-shared/raspberrypi3/kernel-source/Makefile:541: recipe for target 'menuconfig' failedmake[2]: *** [menuconfig] Error 2Makefile:150: recipe for target 'sub-make' failedmake[1]: *** [sub-make] Error 2Makefile:24: recipe for target '__sub-make' failedmake: *** [__sub-make] Error 2Command failed.Press any key to Jan 16 13:04:17 continue... ` Jan 16 13:04:21 hip hip hooray green build https://autobuilder.yoctoproject.org/typhoon/#/builders/83/builds/639 Jan 16 13:04:25 RP ^^^ Jan 16 13:04:42 "send qa email" seems not always working though Jan 16 13:07:37 nacknick: *shrug* Jan 16 13:08:10 aren't you missing libncurses-dev package on your host system by any chance? I don't know if that's the error returned by menuconfig usually but maybe? Jan 16 13:10:27 qschulz: you're right :) Jan 16 13:10:48 Is there a recipe target which only runs if the package is installed through a package manager? That is, I'd like to reload udev or systemd units if the package was installed via opkg but have no need to do this on the initial boot. Jan 16 13:11:45 pkg_postinst_ontarget_${PN} is close but those run on a freshly flashed device on the first boot. Jan 16 13:12:26 fullstop: there is a difference between "no need" and "shouldn't absolutely happen" Jan 16 13:12:55 Can I rephrase it to "absolutely should not happen" ? Jan 16 13:13:16 fullstop, why is pkg_postinst() not suitable? Jan 16 13:15:24 fullstop, for systemd you generally need to just inherit systemd and it will do the right thing Jan 16 13:15:42 when does pkg_postinst() run? Jan 16 13:16:42 fullstop, on rootfs generation, and on target when packages are updated via package manager Jan 16 13:17:10 but as I said, if you just want the standard, tested behaviour for systemd and udev, then inherit systemd without further hassle Jan 16 13:17:23 kanavin: Here's one example.. and I'm certainly open to accepting that I am doing this wrong. One of my packages requires that another service is disabled.. so the postinst_ontarget disables the other service. Jan 16 13:17:44 This makes the first boot fail / hang forever. Jan 16 13:19:30 you probably should not be installing the other service to begin with. Disabling them from pkg_postinst doesn't seem right. Jan 16 13:20:23 or write package dependencies such that your package replaces the other service, then it will be deinstalled. Jan 16 13:24:54 kanavin: not quite 100% green but definitely better than many builds :) Jan 16 13:29:50 hmm Jan 16 13:30:38 New news from stackoverflow: Unable to login with root user on yocto build image for beagle bone black Jan 16 13:31:05 So I want dnsmasq but I don't want to install the service. I guess I should append the dnsmasq package to skip the service part instead of touching it in mine. Jan 16 13:32:58 fullstop: or just disable it through your distro Jan 16 13:33:30 something like SYSTEMD_ENABLE_pn-dnsmask = "false" or similar should probably work Jan 16 13:33:33 How do I go about doing that? Jan 16 13:33:44 wow, I'll take a look at that option Jan 16 13:34:18 you have to iron out the details yourself, but looking at the systemd class should give an idea of the needed variables Jan 16 13:35:07 fullstop: INITSCRIPT_PARAMS_${PN} = "remove" ? Jan 16 13:35:23 arf, that might be sysvinit-specific Jan 16 13:35:36 but i'm sure there is something more or less in that shape Jan 16 13:36:42 https://www.yoctoproject.org/docs/2.6.2/mega-manual/mega-manual.html#var-SYSTEMD_AUTO_ENABLE Jan 16 13:36:54 so what LetoThe2nd said basically Jan 16 13:37:39 * LetoThe2nd always says basic things. Jan 16 13:38:32 qschulz: ah i see. so whilst in development i basically never devtool reset then? Jan 16 13:39:03 Yatekii: if you're doing you're development with devtool yes. You could have used the AUTOREV thing which you set up yesterday Jan 16 13:41:17 oh wow *your development* Jan 16 13:42:27 qschulz: i use autorev Jan 16 13:42:44 Yatekii: well, not really, you're using devtool first :) Jan 16 13:42:57 AUTOREV is not the mechanism used by devtool Jan 16 13:46:37 well i need to use it for deployment ;) but that only works if i did update before Jan 16 13:48:58 you could perfectly have a bumped version of your sw and work with devtool for it, create the patches with it, send a patch or an MR and then bump the recipe. Each their workflow :) Jan 16 13:49:15 thats just tedious lol Jan 16 13:49:44 i want "build && deploy" everything else is completely inacceptable for development Jan 16 13:49:58 why would i fuck around with versions or patches? Jan 16 13:50:26 devtool modify xx, keep in that state, then repeat "devtool build xx && devtool deploy-target xx" Jan 16 13:51:53 yep thats what i am doing and thats also why i need the devtool Jan 16 13:52:24 :) Jan 16 13:52:44 thats why i said autorev doesnt help at all :P Jan 16 13:53:17 devtool modify and autorev don't have anything to do with each other Jan 16 13:55:08 thanks everyone! Jan 16 13:57:52 Yatekii: FYI, you shouldn't use AUTOREV in builds. So when you have something decently ready, use commit hash instead. Build systems are supposed to provide the same build if it's built today or in 10 years. AUTOREV does not allow that. Jan 16 13:58:55 Yatekii: what I was saying is: 1) no you don't actually use AUTOREV, you use devtool. Devtool overrides your recipe with its own. Devtool does not use your AUTOREV thing after it's gotten the sources. (it uses externalsrc mechanism) Jan 16 14:00:13 2) some people use AUTOREV for fast moving repo during heavy development when in a team. if you're alone, devtool is perfectly fine. I personally see devtool as a debug tool only but that's my opinion Jan 16 14:00:17 anyway Jan 16 14:03:02 oh c'mon ltp do_package_qa taking 21 minutes and counting is silly Jan 16 14:07:16 rburton: I did try and speed that up at one point :/ Jan 16 14:07:23 what the hell is it doing Jan 16 14:09:20 rburton: don't remember :( Jan 16 14:10:33 can't find the speedup patch I had Jan 16 14:17:03 definitely doing quality assurance Jan 16 14:18:11 fullstop: i certainly hope so Jan 16 14:19:30 It's gonna be so quality Jan 16 14:20:48 i wish that was the case Jan 16 14:21:28 qschulz: ok thanks for the explanation :) well so you are telling me when you are developping your user space application you make a commit for every small feature you implement and then change the has and fully rebuild? Jan 16 14:21:34 I can waste my time in better fashion ;) Jan 16 14:23:18 I was just saying in other words what LetoThe2nd said. devtool and AUTOREV are two different things. Each their use. Jan 16 14:23:50 I don't work in userspace. Kernel is where I stop. Yocto is my foot in the userspace world. devtool isn't so much fun for debugging the kernel :) Jan 16 14:27:49 hmm ok I see :) Jan 16 14:27:57 thx for all the help :) Jan 16 14:28:11 imma code a bit on my application and then I'll see what workflow goes best Jan 16 14:30:36 pleasure, happy coding Jan 16 14:49:51 Hi, Im trying to do a 'bitbake -c populate_sdk core-image-minimal' but it doesnt work because I use dpkg for packagemanagement. Is there a fix/patch for this problem? Jan 16 14:51:14 wasn't aware it was broken Jan 16 14:51:16 can you file a bug Jan 16 14:51:36 dpkg is the least-tested option so i do suggest just switching to opkg or rpm Jan 16 14:52:01 Thanks. But I cant, its a customer demand :( Jan 16 14:52:14 do they actually use the feeds though? Jan 16 14:53:03 Sorry, I dont understand. (Im super new with Yocto) Jan 16 14:56:50 Linus_SWE: if you are only producing images and sdks, then the actual package management tool does not matter, as it appears nowhere Jan 16 14:56:56 customer may be asking for dpkg but unless they're actually using dpkg on the target there's no point Jan 16 14:57:00 and for a sdk, its irrelevant anyway Jan 16 14:57:10 but please do file a bug Jan 16 14:57:37 Linus_SWE: it only matters if they really need runtime package management on the target (which in turn requires at least some form of package repository server) Jan 16 14:57:50 has anyone tried to build midori browser for wayland? Jan 16 14:58:31 Thanks. They have local repos for deb-packages and want to use the workflow for the device we are building for them now. Jan 16 14:59:48 Ill try it on a clean state and if it is the same Ill file a bug. Jan 16 15:06:53 I tried both opkg and rpm.. opkg is _significantly_ smaller in terms of image size. Jan 16 15:09:16 it is for sure Jan 16 15:16:27 why does using opkg result in smaller images? Jan 16 15:17:32 dnf + rpm + python + other deps is bigger than opkg, from what I can see. Jan 16 15:18:19 plus, the opkg db is heavily size optimized Jan 16 15:18:28 Question on dependencies and sysroots: if P1 depends on P2 which depends on P3, I'm seeing P3 files (headers) in the sysroot for P1. How does bb determine that these headers are needed? In some cases, I've been able to exclude them via DEPENDS_${PN}_remove, but in packages with many (60+) auto-detected dependencies, it's non-trivial. Jan 16 15:19:02 I didn't compare the db or package sizes, but that makes sense. Jan 16 15:19:13 Related question: what's the best way to generate an ascii dependency graph like the one at the bottom of this page? The tool is looking for pn-depends.dot, but -g currently generates tasks- and recipes-depends.dot. Jan 16 15:19:46 stacktrust: enable buildhistory Jan 16 15:19:57 that will give you full graphs for about each and everything Jan 16 15:20:02 thx for the info :) Jan 16 15:20:46 Sorry, got that dependency question wrong, should be: if P1 depends on P2 which depends on P3, I'm seeing P3 files (headers) in the sysroot for Jan 16 15:22:20 Third try: if P1 depends on P2 which depends on P3, I'm seeing P1 files (headers) in the sysroot for P3. How does bb determine that these headers are needed? With shared libraries, it can look at binary files to determine dependencies. What is it doing for header files? Jan 16 15:22:42 LetoThe2nd: will look at buildhistory, thanks Jan 16 15:23:19 stacktrust: not sure if i got your question right.. so are there headers you don't expect? Jan 16 15:25:24 stacktrust: that sounds odd, if the dependencies are that way it seems really odd that P1 headers end up in P3 sysroot. How can they, when P3 have to be built before P1 can be built, due to dependencies? Jan 16 15:26:18 Sounds a bit circular Jan 16 15:27:10 Yes, one package has internal and external headers. Some other packages depend on internal-only or external-only headers, so some headers were broken out into a standalone package. At some point, an indirect chain get established back to the base package with all headers. For some reason, OE puts the "external-only" headers back into the sysroot for the base package, at which point they conflict. Jan 16 15:27:14 right, check the dependency tree to see if there are some 'spde channels' bringing those files in Jan 16 15:27:47 Yes, I'll work with the dep tree. If needed, will come back with a more specific example. Thanks. Jan 16 15:44:30 * zeddii makes a note to figure out how to increase znc's history buffer Jan 16 16:01:10 New news from stackoverflow: Switching WiFi connection fails with dbus timeout Jan 16 16:02:11 kanavin: your valgrind ptest fix also made the documentation build, so i presume there's another S!=B bug in there too Jan 16 16:02:19 packages/corei7-64-poky-linux/valgrind/valgrind-doc: PKGSIZE changed from 0 to 23152221 (+100%) Jan 16 16:10:30 rburton, right, I didn't spot that, but I guess having documentation is not a bad thing :) Jan 16 16:10:46 right Jan 16 16:11:01 I was a bit disappointed that it didn't resolve all of the ptest failures, there's still about 6 that fail for other reasons Jan 16 16:11:18 I kind of want to resolve everything and drop @expectedfail from ptest Jan 16 16:11:40 Hey, am trying to get what's going on on the server once it gets started. Just got here https://git.yoctoproject.org/cgit/cgit.cgi/poky/tree/bitbake/lib/bb/server/process.py?h=zeus&id=cf92a2d567260b91a259652bad0ecd790750f710#n451 Jan 16 16:12:42 Anyone knows if the concepts for the `Cooker`, `ProcessServer` and `Bitbakeserver` are top level concepts to understand for this level of abstraction? Jan 16 16:14:00 I'm guessing `Bitbakeserver` is the central server that manages the whole process. It likely spins up `ProcessServers` for individual task to run the build in parallel... Jan 16 16:15:02 (The `ProcessServer` inherits from multiprocessing.Process) Jan 16 16:15:34 What I'm not clear about is the purpose of the `Cooker` Jan 16 16:16:19 If anyone could shed some light on things, greatly appreciated :) Jan 16 16:17:37 (my end goal is to understand the overall runtime of the build and what needs to be done so that an execution of a single task can be stopped for debugging without that causing havoc in the build) Jan 16 16:17:38 read cooker.py Jan 16 16:17:58 MeanEngi: if you just want to stop a single task for debugging then just stop a single task for debugging Jan 16 16:18:22 https://github.com/rossburton/meta-ross/blob/master/classes/pdb.bbclass Jan 16 16:18:43 inherit that, in a python task just call pdb() somewhere, then attach your pdb Jan 16 16:19:41 (you'll need rpdb installed) Jan 16 16:21:41 rburton: I did something similar to that already it worked once I tried the first time but have started to get some strange errors. So just wanted to dive a bit see if I can get the big picture also :) Jan 16 16:22:14 what are the errors? by attaching a pdb you're just causing the task to take longer Jan 16 16:24:13 I've let it go, will post them if I manage to get them again. Am now trying to get the concepts from the codebase Jan 16 17:31:26 New news from stackoverflow: Run u-boot command at startup Jan 16 17:53:23 khem: FWIW its finding the host ld.so.conf and this is breaking things. We need to have our own with our own paths in to be found first Jan 16 17:53:31 (nativesdk-bintuils that is) Jan 16 17:53:34 moto-timo: ^^^ Jan 16 17:54:58 w00t. systemap works on v5.4 and mips64 Jan 16 17:55:00 onto the next. Jan 16 18:09:10 RP: hmm perhaps define LD_LIBRARY_PATH in env file Jan 16 18:09:36 or maybe own entry in /etc/ld.so.conf.d/ Jan 16 18:27:18 zeddii: \o/ Jan 16 18:27:40 RP, khem: nice to see we are getting closer to the root cause Jan 16 18:31:39 New news from stackoverflow: PowerPC is unable to write/read on PCI MMIO space (Unrecoverable Machine Check) Jan 16 18:38:47 Hey, I'm trying to create a recipe. Upon building I get the error | /bin/sh: curl: command not found. But adding DEPENDS += "curl" doesn't resolve the problem Jan 16 18:39:53 nhartman: look like you're missing curl on your host machine Jan 16 18:40:41 I appear to have it with curl --version Jan 16 18:43:01 paths such as /usr/bin are not available when building so that host binaries do not get mixed in with buildroot ones. Jan 16 18:43:12 er.. yocto ones Jan 16 18:44:26 nhartman: perhaps try something like curl-native Jan 16 18:45:10 Will do, thanks Jan 16 18:49:20 fullstop: Great thanks, that fixed the issue. If you don't mind me asking, how did you know to use curl-native. Referencing layers.openembedded.org, it doesn't even list that as a recipe Jan 16 18:51:25 I'm not sure if it can be added to any package or not, but -native means to build it for the arch/platform of the machine being used to build the target. Jan 16 18:52:25 so opkg would build for target, opkg-native would build it for the host. In my case, this would result in aarch64 binaries for opkg and x86_64 binaries for opkg-native Jan 16 18:53:00 Good to know, thanks Jan 16 18:54:05 Is it possible that curl-native would have different/non-compatible flags than the Makefile specified? Jan 16 18:54:36 The actual curl command appears to be failing with "Error 77" now Jan 16 18:55:46 Pretty sure that has something to do with tls certificates Jan 16 18:57:25 Hmm... I can run the curl command locally with no issues Jan 16 18:58:27 curl-native exists because the curl recipe has BBCLASSEXTEND=native in, and that magically creates a curl-native for you Jan 16 18:58:57 and yes, curl-native doesn't know where the ssl certs are stored, so if you need them you need to tell it Jan 16 18:59:05 its a bit of a mess tbh Jan 16 18:59:18 mysteries unraveled Jan 16 18:59:19 do you actually need curl? wget is always available Jan 16 19:00:37 It's not my Makefile unfortunately. Managed to get it working by manually copying my certs in Jan 16 19:00:47 why is a makefile fetching during compile Jan 16 19:00:48 bad makefile Jan 16 19:00:59 rule is, download in do_fetch Jan 16 19:01:05 I think it's being ran during do_configure Jan 16 19:01:09 then you get proxies set, and the system knows what is being downloaded Jan 16 19:01:19 useful when someone asks for your GPL compliance tarball Jan 16 19:01:38 download something in do_configure and its not in the archive of sources used Jan 16 19:01:51 also doesn't get cached in DL_DIR Jan 16 19:02:27 That's not a bad option, nhartman. I assume that the Makefile only downloads if needed? Jan 16 19:03:10 You could fetch it in do_fetch and put it in the appropriate location for the Makefile Jan 16 19:03:25 by which he means just add the right file to SRC_URI Jan 16 19:03:33 yes. Jan 16 19:03:35 you can tell the fetcher to put it in the right place Jan 16 19:04:06 opencv is a great example of how to beat a dumb makefile with the bitbake stick of beating Jan 16 19:04:07 Could be tricky if there is Makefile Magic constructing the URL. Jan 16 19:04:10 Oh that's great. Is it safe to assume that the cert will be generated for a given machine Jan 16 19:04:38 Does yocto/bitbake use the native wget? Jan 16 19:04:50 by native I don't mean -native, I mean wget from the OS. Jan 16 19:04:54 fullstop: yes Jan 16 19:05:02 fullstop: can't build wget without fetching wget :) Jan 16 19:05:27 python http.. :-) Jan 16 19:05:47 yeah, true Jan 16 19:05:59 *so tempted* to just make bitbake depend on python requests and ditch the wget fetcher Jan 16 19:06:07 because the wget fetcher has to work around wget being wget Jan 16 19:06:18 oh use curl Jan 16 19:06:21 :) Jan 16 19:07:20 rburton: I think there is push for same thoughts in js, go, rust community, they implement everything in the respective language, its good and bad Jan 16 19:07:27 i also have a branch somewhere porting wget fetcher to curl Jan 16 19:07:39 requests is pretty damn good Jan 16 19:07:40 yeah that will be a good change Jan 16 19:07:57 I like requests a lot. It makes me wonder what the python folks were thinking with urllib. Jan 16 19:08:04 rburton: yes I agree, I am not opposed to it Jan 16 19:08:10 fullstop: urllib is very much building blocks Jan 16 19:08:17 though duplo not lego Jan 16 19:09:00 I wonder what the mega bloks equivalent is Jan 16 19:09:03 I am not expecting bitbake to be re-implemented in other than python in near future so depending on pythonic features while adds more dependence on modules is ok Jan 16 19:10:30 rburton: requests is awesome Jan 16 19:10:51 fullstop: urlib /facepalm/ Jan 16 19:11:34 khem: especially when those modules are actively developed, mature, robust :) Jan 16 19:16:09 heh sure Jan 16 19:23:32 With that being said, bitbake has fetched billions and billions of bytes with wget and it clearly works. Jan 16 20:01:13 I'm toying with some cool experiments: I'm testing Yocto compilation in Azure. All up to 64-core machine. Trying to find the balance between cores, time and cost. Jan 16 20:02:13 I don't know how azure does it.. but be sure to factory in sstate-cache, network and downloads.. I know some folks had done this with AWS and certain configurations ended up costing only pennies a build Jan 16 20:02:41 (and if the result of the build needs to be exported out of there -- that usually costs more.. at least on AWS it does) Jan 16 20:03:57 fray: Yes, network ingress of source, egress of build artifacts and storage of sstate cache is extremely important to factor in. The latter esp because it has so huge impact on the build times. Jan 16 20:05:54 In azure you pay as you go (at least our argreement is), so the plan is to build a system that power up the bigdog build server, executes the build task, exports the results and shuts down. The lesser more cheaper servers gets to handle the front-end and serving of contents. Jan 16 20:07:29 I've created a small repo for testing basic build perfomance: https://github.com/sveinse/yocto-benchmark Jan 16 20:07:43 Does azure have anything like the s3 service in AWS? When I was involved that was where all of the artifacts when.. and it made it cheaper for the egress since you could access them that way Jan 16 20:08:27 fray: not sure, as I'm not into aws terms. (nor azure tbh, because this is all very new to me) Jan 16 20:09:34 fray: I've been playing with Backblaze B2 & Cloudflare lately. If you set it up right there's no egress charges and only $5/TB/month storage charges Jan 16 20:09:36 There are storage blobs, so I would assume they are s3 like. I wouldn't be surprised if they were Jan 16 20:10:35 Got some more polish and testing to do but once it's done I'm planning to write somewhere how to set it up Jan 16 20:13:38 hello. I know that LetoThe2nd told me to use my brain capacity to learn to write layers instead of trying to get sdk and extensive sdks, but my head won't let it go that easily. I guess I am stuborn. So, I am trying to run bitbake some-image- -c populate_sdk_ext but I get some error when executing a python function in exec_python_func() autogenerated..... any hint on how to debug further or where to find Jan 16 20:13:44 more information? Jan 16 20:13:57 I have tried to google about it but can't seem to find anything that helps Jan 16 20:14:21 nemgti-og: Could you post the full bitbake output via a pastebin? Jan 16 20:14:30 will do. Jan 16 20:17:46 Where does SRC_URI place files? Trying to copy a file with install ${WORKDIR}/file and bitbake cannot find it Jan 16 20:19:28 He paulbarker, here it is Jan 16 20:19:33 sorry.. Jan 16 20:20:01 hey paulbarker now it IS here: https://pastebin.com/u8m4Vfen Jan 16 20:21:54 That seems like a bunch of bitbake runs tagged together and I can't see the exact bitbake command which failed Jan 16 20:23:58 The actual failure seems to be that nothing matches the glob '*-buildtools-nativesdk-standalone-*.sh' in the SDK_DEPLOY directory Jan 16 20:24:41 But finding why that is the case may be a little more tricky Jan 16 20:25:06 that must be because it is ran insided a docker image I created... the bitbake command is: bitbake meta-ivi-image -c populate_sdk_ext (I don't know if this is the command you were referring to) Jan 16 20:25:37 And does a plain 'bitbake meta-ivi-image' work? Jan 16 20:25:59 I'd also test if it works outside a docker image. Docker is great for automation but not for debugging Jan 16 20:26:11 basically, i wrote a script that run (or tries to) the bitbake process inside a container of a crops/poky image Jan 16 20:26:15 ok Jan 16 20:26:21 yes Jan 16 20:26:34 a plain gitbake meta-ivi-image works fine Jan 16 20:27:17 nemgti-og: Are you inheriting testsdk.bbclass? Jan 16 20:27:42 And, do you have TESTIMAGE_AUTO set? Jan 16 20:28:30 jeeeezz..... sorry but I don't know. I started yocto last week... Jan 16 20:28:35 lol, Ok Jan 16 20:28:44 Do you know what SDKMACHINE is? Jan 16 20:28:52 yes I have seen that Jan 16 20:29:13 i am not explicitly setting that variable in my conf/local.conf Jan 16 20:29:34 Try: bitbake -e | grep ^SDKMACHINE= Jan 16 20:30:58 yeap! that equals "x86_64" Jan 16 20:32:20 Ok, can you use the same method to see if "testsdk" is in the INHERIT variable? (you can also dump to a file `bitbake -e > env.txt` and look manually instead of pipe + grep) Jan 16 20:32:21 MACHINE = qemux86-64 in my local.conf Jan 16 20:33:17 nemgti-og: bitbake -e is pretty useful for tracing what and why a variable is set Jan 16 20:33:27 "testsdk" is not in the INHERIT variable Jan 16 20:33:50 nemgti-og: OK Jan 16 20:33:56 thanks JPEW. -e stands for environment I guess? Jan 16 20:34:03 nemgti-og, correct Jan 16 20:34:45 whenever I see an unsuscribe from the various mailing lists, I wonder if I'm getting a heads up on someone changing roles... Jan 16 20:34:48 nemgti-og, You can also be more specific if you want the environment for a specific recipe (bitbake -e my-recipe), multiconfig (bitbake -e mc:my-multiconfig) or recipe in a multiconfig (bitbake -e mc:my-multiconfig:my-recipe) Jan 16 20:35:24 so... should I add testsdk to the INHERIT variable? Jan 16 20:35:31 :q Jan 16 20:35:51 nemgti-og, No, I was just curious because I think I've seen testsdk fail in containers before Jan 16 20:36:05 ha ok. so you're suggesting me to try to find out what what the environment for my recipe is? Jan 16 20:36:08 got it Jan 16 20:36:32 hey that is cool! Jan 16 20:36:42 zeddii: nah its usually someone i finally managed to scare away! Jan 16 20:36:46 nemgti-og, Ya. Might be worth trying to figure whats different inside vs. outside the container Jan 16 20:39:09 alright. I will try to find out where's wally in this mess Jan 16 20:39:14 thanks JPEW Jan 16 20:46:47 ;) Jan 16 20:50:41 zeddii: but for reasons unknown, i fail very often in this noble task. Jan 16 20:50:54 I'm running into a similar issue with go. Added it with DEPENDS and yet one of the install scripts complains about 'go: command not found' Jan 16 20:51:36 install scripts? Jan 16 20:51:52 still, pretty sure you need to inherit a class for go, not just add to depends? Jan 16 20:53:30 kergoth: "go away" Jan 16 20:54:15 kergoth: or "inherit gone" Jan 16 20:55:20 kergoth: managed to fix it by depending on 'go-native' instead of 'go'. Not sure what inheriting a class means though sorry Jan 16 20:56:39 inheriting a class is bitbake recipes 101. every recipe in oe-core does so Jan 16 21:22:31 kergoth: can you also bbappend a class now? Jan 16 21:24:16 no, that's never been supported. bbappend goes based on recipe filename and only recipe filename Jan 16 21:24:53 ah crap. my tested is suspended for a rm -rf. Time to call it quits for the day. Jan 16 21:36:41 kergoth: aw, but what if you need to? Jan 16 21:38:23 two to three options, you can override the bbclass by providing in your own layer and listing it before the other in BBLAYERS, or you can create a new bbclass that alters what the othre did after the fact, where possible, and ensurei t gets inherited as well hwere needed, or you can do the same in the recipes affected via bbappending them Jan 16 21:46:55 khem: so i want to rip py2 out of core and move it to meta-python2 Jan 16 21:47:10 khem: the impact on meta-oe is going to be ... fun Jan 16 21:47:33 not sure if you want to just add meta-py2 as a dependency first, or immedately blacklist everything that uses py2 for now? Jan 16 21:48:07 LetoThe2nd: LOL Jan 16 22:02:16 anyone happen to know the binutils source well? specifically how the scripts in ld/emulparams are created, and what might cause one to be missing? Jan 16 22:13:08 Has there been made any efforts in optimizing the heuristics for task execution parallelization? E.g. to maximize core utilization. I still observe a few "choke" points where most cores are idle. Jan 16 22:14:12 I do understand why these choke points occur, e.g. a base library or toolchain or similar. Jan 16 22:18:54 sveinse: no effort in saying "do this as soon as you can because it takes for ever so the sooner you start the better" Jan 16 22:19:16 would be useful for something like gettext which is a bit of a bottleneck and takes forever to build Jan 16 22:31:47 rburton: yeah. I'm benchmarking various number of CPUs against compilation time, and it is a function of how many such bottleneck there are. If measured as efficiency in "CPU units" vs time, the optimum is lesser than the highest number of CPUs because of this. Jan 16 22:34:58 On that note: I don't think the problem with gettext is that it takes forever, but the number of dependencies on it that makes it into a constriction. Jan 16 22:41:41 its both Jan 16 22:41:53 it takes forever to configure and its a huge number of deps Jan 16 22:42:05 if it didn't take forever to configure it wouldn't be a really bad bottleneck Jan 16 22:47:38 rburton: perhaps blacklist Jan 16 22:47:57 world builds will be smaller, so I have incentive Jan 16 22:48:02 :) Jan 16 22:57:44 FWIW I've started a wiki page to try and collect a bit more info about who and where YP is being used: https://wiki.yoctoproject.org/wiki/Project_Users Jan 16 22:58:12 Anyone feel free to add things they know about, I've started with a really basic set of easy things like the members list Jan 16 23:01:13 Hmm, I'm getting a troublesome do_package_qa error when trying to build some meta-sdr stuff Jan 16 23:01:14 ERROR: uhd-3.15.LTS-r0 do_package_qa: QA Issue: /usr/bin/uhd_images_downloader contained in package uhd requires /mnt/DATA_SSD/yocto/build/tmp-glibc/work/core2-64-oe-linux/uhd/3.15.LTS-r0/recipe-sysroot-native/usr/bin/python3-native/python3, but no providers found in RDEPENDS_uhd? [file-rdeps] Jan 16 23:01:29 Anyone know how something could think it needs a native build of an executable like that? Jan 16 23:02:02 tgamblin: some script has it as the path at the top of the file? Jan 16 23:02:30 tgamblin: have a look at the top of /usr/bin/uhd_images_downloader Jan 16 23:02:48 RP: yeah. I tried correcting that with some sed magic in do_install_append, but after pointing it to /usr/bin/python3, it reports the same issue, even though python3 is in the RDEPENDS Jan 16 23:03:13 tgamblin: I'm guessing your magic isn't working Jan 16 23:03:48 This is the result: ERROR: uhd-3.15.LTS-r0 do_package_qa: QA Issue: /usr/bin/uhd_images_downloader contained in package uhd requires /usr/bin/python3, but no providers found in RDEPENDS_uhd? [file-rdeps] Jan 16 23:03:49 5:33 Jan 16 23:03:53 was it the same or a different file the second time? Jan 16 23:04:13 tgamblin: do you actually rdepend on python3-core Jan 16 23:04:20 ok, that error is different Jan 16 23:04:31 python3 is an empty package, and that check doesn't follow dependency chains iirc Jan 16 23:04:42 rburton: ah, let me try core... Jan 16 23:06:47 * tgamblin pats his poor 2700X on the case as it chugs along with the build... Jan 16 23:07:01 success! Thanks rburton and RP Jan 16 23:07:33 tgamblin: you want to point out to management how much more efficient a decent machine would make you ;-) Jan 16 23:08:06 RP: Oh, we have lots at work. I'm just separating hobby contributions out onto my former gaming PC :) Jan 16 23:09:45 tgamblin: fair enough, I was a little surprised Jan 16 23:09:57 RP: Let me check with my management if our company and product can be listed. In the field of the Medical industry (yet not clinical use). Jan 16 23:10:32 sveinse: thanks. I know this can be tricky but it is useful to show people using the project! Jan 16 23:11:30 RP: I agree. But yeah, getting understanding to support the community cannot be taken for granted unfortunately Jan 16 23:13:42 Oh, is WSL Yocto based? Jan 16 23:13:54 sveinse: built using it, yes Jan 16 23:14:00 cool! Jan 16 23:14:02 sveinse: see how interesting this stuff can be :) Jan 16 23:15:47 Actually this can be used both ways. I don't know how often I have to defend the seeming strangeness of Yocto (build ALL of the tool chain, required copious amount of disk and cpu), and actually show that this is relevant, not an oddity. Jan 16 23:19:14 sveinse: it does have a number of benefits :) **** ENDING LOGGING AT Fri Jan 17 02:59:57 2020