**** BEGIN LOGGING AT Mon Aug 22 03:00:00 2016 Aug 22 07:02:08 pidge: ping ? Aug 22 07:59:31 The docs state nfs is better for sstate sharing than http, but at the same time I got the impression from discussion here on this channel that NFS is not optimal. Why is that? In what way? Aug 22 08:29:42 aaaand one more hardcoded reference to gcc (glibc this time) Aug 22 08:37:40 sstate only saves the hashsum of the recipe, not the actual contents, does it? Aug 22 08:38:21 Is it possible to see why a recipe is being rebuilt? Aug 22 08:38:45 sveinse: the sstate caches task output. the "key" is a hash of all the inputs (variables affecting) the task. Aug 22 08:39:47 sstate tasks just copy the output they want to cache in the sstate cache into a directory Aug 22 08:41:07 sveinse: there's a way to get a list of all the variables a task depends on, but it's currently not documented very well Aug 22 08:41:13 i have a pending bug for it Aug 22 08:41:20 see http://www.yoctoproject.org/docs/2.2/ref-manual/ref-manual.html#usingpoky-viewing-dependencies as well Aug 22 08:41:27 I have a kernel recipe which is retriggered on builds. E.g. wipe tmp/ and it rebuilds, not loading from sstate like all the rest. This recipe opens with a warning "Unable to get checksum for linux-dr SRC_URI entry defconfig: file could not be found". I suspect that this might be the cause Aug 22 08:43:03 sveinse: my usual debugging technique is to remove stuff until it stops rebuilding, and then working backwards from there :) Aug 22 08:43:52 sveinse: if you don't remove tmp/ and rebuild, does it say that no tasks needed to be rerun? Aug 22 08:44:57 Ulfalizer: Yes, I know about the -g option, but it (or -e) cannot show the vars in which is stored into sstate, can it? Aug 22 08:45:17 Ulfalizer: Rerunning bb without erasing /tmp does not rebuild the kernel Aug 22 08:45:30 *err erasing tmp/... Huuuge difference Aug 22 08:46:50 sveinse: https://bugzilla.yoctoproject.org/show_bug.cgi?id=10141 has instructions for how to see what variables went into creating the input checksum Aug 22 08:46:51 Bug 10141: normal, Medium, 2.2, srifenbark, NEW , Suggested fleshing out of the sigdata/siginfo documentation Aug 22 08:47:41 Something changed on the pypi front? I'm gonna submit a patch for python-cassandra-driver, but seems that the pypi default url doesn't work or something like that Aug 22 08:48:11 oh, perhaps it does. .siginfo contains the vars in a pickled (?) object. So I need something that can diff them between sstate and the current recipe then Aug 22 08:48:45 sveinse: short version is to find the siginfo file for the task in SSTATE_DIR and run bitbake-dumpsig on it Aug 22 08:49:10 there's bitbake-diffsigs as well for comparing the signature between different versions Aug 22 08:49:11 Ulfalizer: Yeah, I'm reading the bug report now. Looks promising Aug 22 08:49:36 i think the doc maintainer is on vacation. should hopefully go in after that. :) Aug 22 09:09:33 I think I'm ready to conclude: If you are using EXTERNALSRC, please note that bb leaves two symlinks in the path, oe-workdir and oe-logs. Please be aware that these two symlinks must be deleted if tmp/ is erased, otherwise bb will bork when generating sstate cache for that recipe. Aug 22 09:12:12 ^^ is there a place this information should be put? E.g. bug report? Aug 22 09:14:05 sveinse: https://bugzilla.yoctoproject.org/enter_bug.cgi Aug 22 09:14:25 So this /is/ a bug? Aug 22 09:14:34 i have no idea. i've never used EXTERNALSRC. ;) Aug 22 09:15:11 Lucky you. It's either EXTERNALSRC or fixup hg-support for our part. Aug 22 09:16:07 you could send an email to one of https://www.yoctoproject.org/tools-resources/community/mailing-lists as well Aug 22 09:16:18 asking if what you're seeing is intended behavior Aug 22 09:16:26 and suggesting improvements, etc. Aug 22 09:16:38 that has the advantage of being visible to google. the bugzilla doesn't seem to be. :/ Aug 22 09:17:26 ok, I will. (Personally not to fond of email lists any more. It drowns in all the spam I get.) Aug 22 09:19:50 Hello, I try to run PyQT without x11. Anyone has some experience with that? Aug 22 09:32:57 rubdos: I think pypi changed, yes. There's a different URL that can be used now, let me check. Aug 22 09:33:43 neverpanic, thanks. I was panic'ing. ;p Aug 22 09:33:48 I hardcoded the URL for now Aug 22 09:39:04 rubdos: https://bitbucket.org/pypa/pypi/issues/438/backwards-compatible-un-hashed-package Aug 22 09:40:09 rubdos: seems files.pythonhosted.org/packages/source/${d.getVar('PYTHON_PACKAGE', True)[0]}/${PYTHON_PACKAGE}-${PV}.tar.gz is now the way to go Aug 22 09:40:38 reading the issue, it seems to me that the bitbake class should get changed? Aug 22 09:41:01 There always was the convenient "import pypi" statement, which magically downloaded the tar.gz and compiled it... :P Aug 22 09:41:50 rubdos: yes, is it not changed yet? Aug 22 09:42:23 Don't know for sure; I'll check it. I changed my cassandra version to 3.6 (coming from 3.0), and it didn't work out of the box Aug 22 09:42:49 Perhaps it's not backported to krogoth? Aug 22 09:43:46 mmm, krogoth is the latest stable, so don't think that's the problem Aug 22 09:45:28 seems like I didn't fully update my distro, I'll report back Aug 22 09:47:06 neverpanic, this is current pypi.bbclass: https://github.com/openembedded/meta-openembedded/blob/master/meta-python/classes/pypi.bbclass Aug 22 09:47:09 (in master) Aug 22 09:47:35 for reference: http://cgit.openembedded.org/cgit.cgi/meta-openembedded/tree/meta-python/classes/pypi.bbclass?h=master Aug 22 09:52:32 I'm sorry, that url has been changed indeed Aug 22 09:52:38 wow. I'm slow today Aug 22 09:53:24 Doesn't seem to be backported to krogoth though Aug 22 09:53:28 http://cgit.openembedded.org/meta-openembedded/tree/meta-python/classes/pypi.bbclass?h=krogoth Aug 22 09:53:51 http://cgit.openembedded.org/meta-openembedded/tree/meta-python/classes/pypi.bbclass?h=krogoth-next same for krogoth-next Aug 22 09:53:59 It's in master, but not yet in krogoth Aug 22 09:54:02 rubdos: So suggest a backport on the list, then? Aug 22 09:54:10 Yes, I'll dothat Aug 22 09:54:15 (fill in spaces where needed) Aug 22 11:10:52 ello, I try to run PyQT without x11. Anyone has some experience with that? Aug 22 11:11:55 Hunk, so... Wayland? :P Aug 22 11:12:27 Or just the non-gui elements? Either case I cannot help... Just trying to gather information for the guys that can. Aug 22 11:14:21 never installed wayland with yocto. Is that complicated? Aug 22 11:14:46 no, it's provided by poky Aug 22 11:14:58 Hunk: I think you just need to add that to distro features, if it's not in there... Aug 22 11:15:49 what is the package name? The problem is that pyqt require x11 Aug 22 11:16:18 does it really need x11 or does it need qt, on something Aug 22 11:16:26 as qt5 can run on wayland Aug 22 11:16:49 i guess the real question is "do you need graphics at all" Aug 22 11:17:11 yes i need Aug 22 11:17:24 so what were you going to replace x11 with Aug 22 11:17:31 as this isn't a choice to make without any planning Aug 22 11:18:05 I used the default QWS before Aug 22 11:18:09 for qt c++ Aug 22 11:18:13 that work well Aug 22 11:18:19 but pyqt needs x11 Aug 22 11:18:20 # depends on qt4-x11-free REQUIRED_DISTRO_FEATURES = "x11" Aug 22 11:18:33 oh right, qt4. i believe qt4 does need x11 yes. Aug 22 11:18:40 so... I finally compiled yocto without gcc, and now I need to upstream patches to glibc, busybox and the kernel :P Aug 22 11:18:50 boucman_work: \o/ Aug 22 11:19:11 Hunk: port pygt to qt-embedded? Aug 22 11:19:16 Hunk: can you switch to meta-qt5? Aug 22 11:19:18 (tbh, busybox and the kernel have the same bug, it's in the kconfig infrastructure...) Aug 22 11:20:58 I use meta-qt5 Aug 22 11:21:33 but pyqt is inside meta-openembedded Aug 22 11:22:28 google says pyqt for qt5 exists, so maybe you just need to write a recipe or something Aug 22 11:22:59 my system uses pyqt, on qt5 Aug 22 11:23:04 with wayland Aug 22 11:25:19 Hunk: https://github.com/GENIVI/meta-genivi-dev/blob/9649b132ca5d6ed0b03565ff3642a035fd87891e/meta-genivi-dev/recipes-devtools/python/python-pyqt_5.3.1.bb Aug 22 11:25:36 tsk tsk that should be in meta-qt5 or something ;) Aug 22 11:28:02 I can't comment on the origin, but yes I agree Aug 22 11:36:26 Yow, I have trouble building chromium for my image (OOM killer, or so). I got the built rpm from a colleague, can I override the build systems to just use it? dropped it in tmp/deploy/... is not enough.. Aug 22 11:38:43 thank you i will check it :) Aug 22 11:40:12 Hello, my poky version has two versions of the mesa recipe: mesa_git.bb and mesa_10.6.3.bb Aug 22 11:40:29 mesa_git.bb has DEFAULT_PREFERENCE = "-1" Aug 22 11:41:00 and in my local.conf I set PREFERRED_VERSION_mesa = "10.6.3" Aug 22 11:41:16 So everything should be set that bitbake chooses the 10.6.3 version Aug 22 11:42:24 Nonetheless I get the following warnings, then the build fails: Aug 22 11:42:51 NOTE: preferred version 10.6.3 of mesa not available (for item libegl) Aug 22 11:42:51 NOTE: versions of mesa available: 2:10.5.4+gitAUTOINC+ea0d1f575c Aug 22 11:43:35 I've checked the dependencies of libegl, and it does _not_ specify any mesa version to depend on Aug 22 11:43:56 So what's really the problem here? I've been trying to debug it with no luck :-( Aug 22 11:45:02 the mesa..bb will require the .git file Aug 22 11:45:53 CTtpollard, sorry; could you elaborate further? :-) Aug 22 11:45:56 actually maybe not Aug 22 11:46:49 hmmm ... Aug 22 11:46:50 it's common for recipes to depend on a related .git file, but mesa.bb's pull in tarballs Aug 22 11:47:47 CTtpollard, yes .. it's an http tarball fetcher Aug 22 11:48:05 CTtpollard, what's really frustrating is that if I just remove mesa-git.bb .. everything work as expected without any errors! Aug 22 11:50:55 correction: mesa_git.bb .. Aug 22 11:57:23 darwish: are you sure there's nothing else in your system that is explicitly setting the version in meta_git as required? it might also be useful to use the mesa-megadriver Aug 22 11:58:12 CTtpollard, how can another recipe explicitly state the version? I can grep that expression if there's one Aug 22 11:58:15 or the preferred provider of libegl somewhere Aug 22 11:58:36 (that is, how a recipe mandates using a certain version of a dependency) Aug 22 12:02:48 darwish: it seems to be telling you it can't find the 10.6.3 recipe at all Aug 22 12:05:24 jku, it's there .. at the end bitbake complains that its sees both mesa_git and mesa_10.6.3 and that I must choose one Aug 22 12:05:48 jku, and I've already chosen one using PREFERRED_PROVIDER, and it's showing in bitbake -e :-( Aug 22 12:06:52 huh Aug 22 12:08:09 jku, yeah .. it claims that some other dependencies are pulling for mesa_git, and I've chosen mesa_10.6.3 .. and thus it gets confused Aug 22 12:08:23 (two mesa versions included in the same build) Aug 22 12:09:01 How can I see if some random recipe is pulling mesa_git.bb version (instead of mesa_10.6.3.bb) behind my back? Aug 22 12:09:15 Can even a recipe mandate a certain version!? Aug 22 12:16:42 And if any recipe really depends on mesa_git.bb instead of mesa_10.6.3.bb .. why just removing mesa_git.bb makes everything succeed (the mysterious recipe should've bailed out instead I guess?) Aug 22 12:25:33 doesn't mesa offer multiple PROVIDES ? is the _git version providing something 10.6.3 doesn't? Aug 22 12:25:41 hmmmm Aug 22 12:25:55 a recipe can't mandate a specific version of a dependency, that's a distro-level policy decision Aug 22 12:26:21 great; I'll diff the two Aug 22 12:27:57 joshuagl, both include the same include file with the same provides :-( Aug 22 12:27:58 https://github.com/01org/luv-yocto/blob/master/meta/recipes-graphics/mesa/mesa_10.6.3.bb Aug 22 12:28:03 https://github.com/01org/luv-yocto/blob/master/meta/recipes-graphics/mesa/mesa_git.bb Aug 22 12:53:33 darwish: hmm, not entirely sure what's going on tbh Aug 22 12:53:48 yeah, neither do I :-( :-( Aug 22 12:53:51 same error for different MACHINE? Aug 22 12:54:18 will try Aug 22 14:14:53 hello Aug 22 14:18:37 I want to give someone a archive of the cross development sdk. I did bitbake -c populate_sdk myimgname and it did indeed build a sdk which my coworker could install, but it created an archive which is looking for file in my /home dir where I built the code. This url http://www.yoctoproject.org/docs/1.6.1/adt-manual/adt-manual.html does not say anything about the environment file have references to where the Aug 22 14:18:43 build is made or how to remove the references. ie. other options to provide. Aug 22 14:20:18 that sounds like a bug, the SDK should be able to be installed independently of the build location Aug 22 14:22:46 nvm, yah one thing I need try Aug 22 14:24:53 well 1.6 is really outdated by now. you might consider looking into a recent release, which also offers the extensible sdk facilities. Aug 22 14:26:50 no, i confirmed, he took the .tgz i gave him, sourced the script, it built the code and put in an install dir, in the install dir there is a env script. Aug 22 14:27:02 . ./the-env-script Aug 22 14:27:43 and to test we tried to build something and it complained that th ecompiler oculd not make exe's. Aug 22 14:28:15 i look at the env script and it definitely has refs to my dir with -f directives Aug 22 14:29:46 ok, perhaps I screwed up. Aug 22 14:30:16 i tar'd deploy/glibc/sdk and I gave him that. I should have sourced the script and give him the results. Aug 22 14:30:31 nvm, let me build it on my box and give him the output. Aug 22 14:30:34 my bad Aug 22 14:37:07 hmm. no Aug 22 14:58:21 hi everyone Aug 22 14:58:33 hi aurele Aug 22 14:59:09 is it possible to build a 32bit sdk on a 64 bit machine Aug 22 14:59:20 darwish: recipes don't control versions of dependencies. versions are specified using PREFERRED_VERSION Aug 22 15:00:08 aurele: it is http://www.yoctoproject.org/docs/2.1/ref-manual/ref-manual.html#var-SDKMACHINE Aug 22 15:00:16 kergoth, yeah .. which makes me wonder why Yocto is not able to recognize PREFERRED_VERSION_mesa = "10.6.3" :-( Aug 22 15:00:38 first, use bitbake -e to confirm its set the way you think it is, and wasnt overridden by a config file parsed after local.conf, like distro or machine Aug 22 15:00:43 bitbake -e | grep PREFERRED_VERSION_mesa= Aug 22 15:01:01 (I have a build error with qt5/qtwebengine seems to be due to address space and "oomkiller", but only on 32 bit machines) Aug 22 15:02:21 joshuagl thanks Aug 22 15:03:55 aurele: that would be because webkit can be bigger than 4gb when linking, so good luck holding that in memory on a 32-bit host. using gold as linker may help, or pass —no-keep-memory to the linker Aug 22 15:06:02 rburton, many thanks I knew there was something but can't retrieve it, I will try this Aug 22 15:08:18 having trouble with BB_ENV_EXTRAWHITE, anyone used this? Trying to import CVSROOT environment variable from outside bitbake Aug 22 15:09:10 rburton, aurele: you can also adjust the compiler options to make it consume less ram at link time, we have a workaround of that sort in meta-mentor. not always enough, but might help. see https://github.com/MentorEmbedded/meta-mentor/blob/master/meta-mentor-staging/recipes-sato/webkit/webkitgtk_%25.bbappend Aug 22 15:10:14 nillerbrun: adding a var to BB_ENV_EXTRAWHITE should get it into the metadata, but won't re-export it. check bitbake -e to see if it's set there. also, iirc oe-init-build-env sets it, it doesn't append it, so your alterations to it have to be done after sourcing that Aug 22 15:10:24 at least that was the case in older versions, i think that's improved somewhat Aug 22 15:11:14 s/sets it/sets BB_ENV_EXTRAWHITE/ Aug 22 15:13:01 Thanks kergoth, I'll check. I have a custom build-env supplied by my hardware manufacturer, perhaps that's my problem. Aug 22 15:13:08 ah, could very well be Aug 22 15:16:53 What alternatives are there to a upload syncing of sstate without using NFS? Would a manual additive rsync work? Aug 22 15:18:47 sstate is just a pile of tar files Aug 22 15:18:50 so yes, rsync can handle it fine Aug 22 15:29:31 kergoth, thanks I will try --no-keep-memory first as I would like to keep debug symbols, but it worth the try Aug 22 15:29:41 * kergoth nods Aug 22 15:34:48 I notice that my local SSTATE use a mix of two hex-prefix and a lot that is put there without any prefix. I won't push the latter to the server. I used to have two-letter only, what have I done? Aug 22 15:36:21 I have SSTATE_DIR ?= "/srv/yocto/sstate" and SSTATE_MIRRORS ?= "file://.* http://our-sstate.local/yocto/sstate/PATH" Aug 22 15:38:38 sveinse: if you use PATH on the remote side, it'll include the path prefix Aug 22 15:39:30 kergoth: So no PATH then? I'll find the 8a/ dirs itself without PATH? Aug 22 15:39:43 I don't understand the question Aug 22 15:40:18 if your remote server doesnt' include the prefix, then you don't want to use PATH in the mirror, you'd adjust the mirror replacement to replace the subdir. if the remote server does include the prefix, then you're fine and don't need to do anything Aug 22 15:40:53 What is an example of a prefix in this context? Aug 22 15:42:38 My question is two-fold: When rebuilding with a blank sstate cache locally, I see some setscenes, but I also do a lot of rebuilds. So I'm wondering if I got the mirror statement right Aug 22 15:43:03 the only prefix we've mentioned is the subdir hte sstates are in Aug 22 15:43:09 8a in your example Aug 22 15:43:15 i didn't pull a random prefix out of nowhere Aug 22 15:43:33 you referred to it as a prefix in your own question 10 minutes ago Aug 22 15:43:46 Secondly, if this is because my machine is different from the build server, I want to push my cache up to the server. But I see now that my sstate cache no longer i divided into two-hex letter dirs, so the structure don't match up with the servers Aug 22 15:44:07 local sstate_dir is always broken up into subdirs Aug 22 15:44:32 not mine apparently Aug 22 15:44:36 it's possible you have sstate archives both at the root of SSTATE_DIR and symlinked into the prefix dirs depending on what yocto version you're using (i.e. old) Aug 22 15:45:04 in which case you should add ;downloadfilename=PATH to your SSTATE_MIRRORS, to make it download it directly inot the subdir paths rather than putting it at toplevel and symlinking it into those Aug 22 15:45:15 in either case a script to resolve the links to move them out of topdir would be trivial to write Aug 22 15:45:21 to correct the current sstate_dir Aug 22 15:45:29 s/inot/into/ Aug 22 15:45:35 I'm running on krogoth (yeah, got this right this time), so it's new enough Aug 22 15:45:58 by default the fetcher fetches into DL_DIR, and we force DL_DIR to SSTATE_DIR when fetching sstate. if the downloaded path doesnt' match expectations, it symlinks to resolve the mismatch Aug 22 15:46:31 so 1. resolve the existing symlinks to fix your current sstate_dir, 2. add downloadfilename=PATH to fix future downloaded sstate archive paths Aug 22 15:46:37 i think that should do it Aug 22 15:47:29 I misunderstood your question, i thought your sstate mirror wasn't organized by prefix subdir, but it sounds like the problem is local, not remote Aug 22 15:47:33 in which case this should fix it Aug 22 15:47:52 kergoth, sorry just noticed your message :-) .. Yes, bitbake -e | grep confirms that PREFERRED_VERSION_mesa = "10.6.3" Aug 22 15:48:04 yes, I'm retrying now :D Aug 22 15:48:14 darwish: next step is to make sure 10.6.3 is even available. run bitbake -s | grep mesa. is the 10.6.3 version listed? Aug 22 15:48:23 preferred version to a version that isn't buildable won't do a thing Aug 22 15:51:09 is it possible to set a compile flag depending on the build machine? Aug 22 15:52:54 kergoth, Aug 22 15:52:54 mesa 2:10.6.3-r0 Aug 22 15:52:54 mesa-demos :8.2.0-r0 Aug 22 15:53:03 that's interesting .. Aug 22 15:53:10 git isnt even listed.. Aug 22 15:53:47 hmm Aug 22 15:54:36 maybe because I force removed it earlier from the file system .. let's sanitize my environment Aug 22 15:55:24 kergoth, "Multiple versions of mesa are due to be built (/home/darwish/projects/gen3/salvator-x/r-car-platform/build/../poky/meta/recipes-graphics/mesa/mesa_git.bb /home/darwish/projects/gen3/salvator-x/r-car-platform/build/../poky/meta/recipes-graphics/mesa/mesa_10.6.3.bb). Only one version of a given PN should be built in any given build. You likely need to set PREFERRED_VERSION_mesa to select the correct version or don't depend on multiple versions." Aug 22 15:55:35 Which recipe is supposed to provide 'ar' (for the host machine) in $PATH? Aug 22 15:55:51 Or is that part of the host machine's requirements? Aug 22 15:55:57 for the host machine? nothing. your build server is expected to have a functional toolchain Aug 22 15:56:01 ar is part of binutils Aug 22 15:56:34 that said, if you're hitting an issue where it wants an unprefixed 'ar' but your machine has ${BUILD_PREFIX}ar, a fix for that hit the list recently, an expansion problem in sanity.bbclass Aug 22 15:59:54 my machine had /usr/bin/ar.single -> x86_64-linux-gnu-ar for reasons I don't understand; dpkg -L binutils clearly listed /usr/bin/ar Aug 22 15:59:55 by default the yocto core-image provides connman as network manager, which is ok but the connman is has the problem that if one connection has successfull an network connection it doesn't try to connect other interfaces to the internet Aug 22 16:00:03 reinstalling the package fixed the issue. Thanks, Debian, I guess? Aug 22 16:00:14 huh, werid Aug 22 16:00:30 That did uncover that 'bc' isn't using the cross-ar as it should, though. Aug 22 16:00:32 It seems like a default behaviour of connman, so I thinking about to switch to 'networkmanager' Aug 22 16:00:39 Did someone tried this? Aug 22 16:01:36 I just compiling networkmanager, but there is huge load of dependencys (e.g. libx11-native) which is crazy Aug 22 16:02:39 maybe see if there's a way to fix connman's behavior? Aug 22 16:02:40 * kergoth shrugs Aug 22 16:03:07 gah, need to rebase shallow git support again due to the chdir fixups Aug 22 16:03:42 kergoth: downloadfilename worked, thanks! Aug 22 16:03:47 np Aug 22 16:04:16 kergoth: well I talked to the developer of connman and he says that this is not easy to change this behaviour Aug 22 16:04:33 ah Aug 22 16:04:35 what about systemd-networkd Aug 22 16:09:26 Hey everyone, I have a small question... I was looking at this commit https://git.congatec.com/yocto/meta-openembedded/commit/45688928778ea5b3a6ed6d8ca614247772846b8f where ruby-native was added as a dependency for ruby in the bb, why is this required even though it was already in ruby's .inc file at the time? https://git.congatec.com/yocto/meta-openembedded/blob/45688928778ea5b3a6ed6d8ca614247772846b8f/meta-ruby/recipes-devtools/r Aug 22 16:12:45 sorry if it's a dumb question, it seemed to fix my problem so I'm wondering why it was necessary.. Aug 22 16:13:36 good question, ask JaMa Aug 22 16:13:54 it was 3 years ago… Aug 22 16:14:50 for now I'm stuck on danny so, haha, been looking at 3-year old stuff Aug 22 16:26:41 Has anyone tried populate_sdk on machine with MULTILIBS configured? Unfortunately I need multilib:lib32 to build one recipe (but don't need it in the SDK). The sdk corei7-64-oe-linux sysroot is polluted with lib32 stuff. Aug 22 16:34:36 hello Aug 22 16:34:44 a question regarding toaster Aug 22 16:35:09 is it possible to set PREMIRRORS variable in toaster ? Aug 22 16:35:26 i tried to add the variable in the toaster web interface but it failed Aug 22 16:35:34 spaces are not supported :( Aug 22 16:35:46 and the variable contains several ones Aug 22 16:36:41 kergoth, Any recommended path I should follow? Why bitbake -s is not showing mesa_git, while bitbake itself complains of conflicts between mesa_git and mesa_10.6.3 Aug 22 16:39:24 * darwish is getting desperate about this .. maybe it's a bug within bitbake? will possibly post to the mailing list :-( Aug 22 16:50:44 iskander: sounds like we went a bit too far when validating variable values in Toaster :/ Could you open a bug in bugzilla.yoctoproject.org? Aug 22 16:52:37 belen: will do, how hard is it to fix ? Aug 22 16:54:12 iskander: thanks! shouldn't be too hard, although we are tight on resources of late. We will try in any case Aug 22 16:55:47 i'm a programmer and would like to fix it soon, therefore, i would appreciate any pointers Aug 22 16:57:42 iskander: excellent, although I've just tried and have been able to add a variable with spaces in the value. Could you tell me how you are trying to add it, and the Yocto Project version you are using? Aug 22 16:59:13 i use yocto 2.1 krogoth Aug 22 16:59:22 i created a new project Aug 22 16:59:35 and went to 'BitBake Variables' page Aug 22 16:59:47 there is a filed to add a new variable Aug 22 16:59:51 field* Aug 22 17:00:04 name = PREMIRRORS Aug 22 17:00:41 value = "git://.*/* http:///yocto-cache http://.*/* and so on" Aug 22 17:01:02 i tried to add it and it failed Aug 22 17:01:29 i copied PREMIRRORS from my local.conf Aug 22 17:01:36 so it should be fine Aug 22 17:01:47 it works without toaster Aug 22 17:02:09 another question Aug 22 17:02:17 PREMIRRORS contains newlines too Aug 22 17:02:42 iskander: thanks. I think I can see the problem now. I get an error saying "too many values to unpack" :/ Aug 22 17:02:43 but toaster's value field allows only oneline Aug 22 17:02:55 belen: yeah, that's the error i get Aug 22 17:03:55 i would like to use PREMIRRORS to save network bandwidth and speed things up Aug 22 17:05:36 iskander: sure. There is just something upsetting Toaster in that value for some reason. Definitely a bug. Aug 22 17:13:10 Hello. Aug 22 17:14:30 I have a question pertaining to Yocto, Mono, and MySQL, and was wondering if this might be the right place to ask? Aug 22 17:15:45 belen, i submitted the bug report Aug 22 17:15:55 iskander: thanks for that Aug 22 17:16:25 if you have some patches to test then i'm ready Aug 22 17:31:40 Heitomos: ask away, and see if you get a response Aug 22 17:33:28 Well, I am working on a Yocto board, using a C# Mono Program, that's attempting to connect to the local MariaDB instance. From the Terminal, I can type 'mysql' and then my username and password, and get in to edit the database. However, when attempting to, from inside of mono, connection with a MySqlConnection object, it cannot connect to the host. I've tried turning on MySql error logging, but that didn't help, and I've otherwise s Aug 22 17:36:41 So I was wondering if anyone knew how to do that particular brand of connecting, or if I'm just missing something horrendously obvious? Aug 22 17:55:12 hmm, think I'll add a reset-layers sub-command to bitbake-layers which resets BBLAYERS to the value from the template by looking at templateconf.cfg Aug 22 17:55:39 (or falling back to just keeping the core layer, if templateconf.cfg is missing or invalid) Aug 22 17:56:24 Hmm, I can't decide whether I want to try to get bitbake-layers to stop parsing bitbake.conf. it'd be nice, since it'd be harder to get into a situation where its commands are unusable Aug 22 17:56:50 which in turn would make it easier to keep small focused granular commands rather than putting everything into one to avoid a possibly broken intermediate state Aug 22 17:56:53 hmm Aug 22 18:16:24 Does yocto have any tools for uploading local sstate cache and/or DL to one of its mirror? Aug 22 18:16:37 Or is NFS the only approach for this? Aug 22 18:16:54 nfs or rsync, as you've been told repeatedly Aug 22 18:18:02 Someone here (and I don't remember who) claimed NFS to be iffy. Perhaps I misunderstood the statement. I suppose it's not you then, kergoth Aug 22 18:19:46 in the eyes of many, NFS is always iffy, for just about everything Aug 22 18:20:17 I'm writing the rsync scripts now, but ran into a little problem with file permission and time-syncing. A setgid directory and rsync is not the easiest approach apparently. Aug 22 18:23:12 Planned on using rsync -rtlv. The -t is required for synchronizing the timestamps on file (preventing multiple uploads), yet it fails when trying to set timestamps on the dirs. And then colleague 2 comes along overwriting all the timestamps, and thus your rsync might behave differently the next time. Aug 22 18:23:46 So I think it's needed to look at this syncing thing with a little more logic, so perhaps I'll write a py script for it Aug 22 18:25:48 Does anyone know much about using Mono and MySQL on Yocto? Aug 22 18:31:27 Heitomos: there is not much to know about specific components from a Yocto PoV other than "is there a recipe?", which can be answered by the layerindex Aug 22 19:00:41 ooi, what is the demographics of yocto users? What are the most common application areas for these projects? Anyone knows? Aug 22 19:17:29 hello all ! I have a question about yocto and kernel configuration throw bbappend. Is it the right place ? Aug 22 19:25:15 sveinse, all over the world! I accidentially fell in one :D Aug 22 19:25:17 (Belgium) Aug 22 19:57:52 kergoth: fwiw my bb fork has a "bb generated" command now Aug 22 19:58:15 cool, will have to check it out Aug 22 20:26:40 hello Aug 22 20:32:04 I'm trying for the life of me to get my build to pull fresh source from git for a recipe I added. this works $ bitback -c fetch pcmx, but $ bitbake pcmx fails. Says do_unpack finished. error function failed Fetcher. reference is not a tree. I've been doing this between builds. $ bitbake -c cleanall pcmx; bitbake -c fetch pcmx; bitbake pcmx. When it looked like it was not pullng fresh, I went into the Aug 22 20:32:10 build/tmp/..../pcmx/.../git and deleted all the subdirs pulled from git. Still i get the error when I try to build pcmx. My only known solution is rm -rf from the root dir and pull the project from git and do process from start. Surely there is a better way. Any tips? Aug 22 20:51:21 davis: cleanall should work just fine. what makes you say it 'looked like it was not pulling fresh'? did you heck the do_fetch task log? Aug 22 20:56:04 kergoth: no. Aug 22 20:56:58 fwiw, i rm -rf at top level. I'm doing a fresh build now. its just that its odd. Sometimes I can work with buidl, do work, build repeat. Other times no. Aug 22 20:59:04 i may have some issues with tree. Even when I do builds fresh, i'll get problems where it can not pull a repot. break. I just keep repeating after a break and it eventually works. Aug 22 20:59:24 it might due to network. Aug 22 21:29:57 RP: if build_mirror_data fails, runfetchcmd will raise FetchError, which will result in mirror checking. using a different mirror won't fix the problem, however, if the problem was in tarring up what we've already fetched. long term I think we should split up mirror tarball construction into a separate method, possibly a separate task, but in the meantime I'm thinking when the fetch core calls build_mirror_data, it should catch FetchError and raise it Aug 22 21:29:57 as something else, i.e. SystemExit/bb.fatal, to prevent mirror checking from proceeding. Aug 22 21:30:05 bluelightning: ^ Aug 22 21:30:06 thoughts? Aug 22 21:30:24 though rp is probably gone, he should see it later Aug 22 21:47:12 kergoth: I have to admit some of that error path is a still bit of a mystery to me, but what you're suggesting sounds sensible Aug 22 22:09:05 hello! I'm using yocto since few days now and trying to customize my kernel config. I do not manage to get fragment working in my kernel bbappend file. I'm on the fido branch and I have seen the same ind of issue browsing on google (particularly: https://lists.yoctoproject.org/pipermail/yocto/2015-June/025119.html). Any known issue on this subject ? Aug 22 22:16:42 joel__: welcome to yocto world !. how are you adding the fragment to metadata Aug 22 22:17:21 is it via SRC_URI? if yes then you can specify like SRC_URI_append = " file://your_config" Aug 22 22:17:33 in you .bbappend for kernel recipe Aug 22 22:18:05 if you are using linux-yocto then there is a linux-yocto workflow available as well. Aug 22 22:20:36 hello khem. thanks for reply. yes it is exactly what i'm doing, SRC_URI_append in my bbappend. the fragment is retrieved using bb menuconfig and diffconfig. At this moment, just to check my bbappend is read by bb, I have put the full kernel configuration in SRC_URI : SRC_URI_append = " file://defconfig". It works. But if I only put a fragment, it is not applied. Aug 22 22:21:21 i'm not using linux-yocto Aug 22 22:23:26 is it possible that something is wrong in the recipe linux-foo.bb I'm trying to patch with my linux-foo.bbapend ? something which prevent bb to apply fragment ? Aug 22 22:38:52 joel__: most kernel recipes don't support fragments at all. Aug 22 22:39:05 only specific ones do, off the top of my head linux-yocto and linux-qoriq Aug 22 22:40:28 most of the ones that don't can have support for them added, often via bbappend, but it depends on the recipe and how it sets up the config Aug 22 22:42:45 hello kergoth. ok thanks for the info, that's good point to know :-) Yes I have used linux-yocto before and I have allready done fragment with it, it was working. Do you know what I can look for to get support of fragments ? Aug 22 22:45:01 basically you have to add kern-tools-native to DEPENDS, then make sure merge_config.sh -m "${B}/.config" is run after defconfig is copied to .config Aug 22 22:45:17 the details vary, as there's no standard for when/where defconfig is copied to .config Aug 22 22:46:27 doing so from the recipe itself is usually easier than via bbappend, unless it relies on kernel.bbclass to copy defconfig to .config, otherwise you can end up with hacks like https://github.com/MentorEmbedded/meta-mentor/blob/46e62a22a61a996b04e5a0192476e241ab89cca5/meta-mel/fsl-arm/recipes-kernel/linux/linux-ls1_3.12.bbappend#L1-L18 Aug 22 22:47:24 see also the busybox recipe and linux-yocto in oe-core, they both use merge_config.sh Aug 22 22:47:36 erm, busybox does, anyway, i think linux-yocto uses its own tooling Aug 22 22:48:15 ok looks great particularly this link Aug 22 22:48:33 I was also reading http://www.yoctoproject.org/docs/1.8/kernel-dev/kernel-dev.html#using-an-in-tree-defconfig-file. is it interesting for my issue ? Aug 22 22:49:26 those instructions only apply for linux-yocto, and isn't about using fragments, it's about how you use a defconfig from inside the source tree rather than from the metadata Aug 22 22:51:20 ok Aug 22 22:51:20 if the recipe is letting kernel.bbclass copy defconfig from WORKDIR to .config, then adding fragments is easier, you can just DEPENDS += "kern-tools-native"; do_configure_prepend () { cp -f ${WORKDIR}/defconfig .config; merge_config.sh -m .config ${@' '.join(s for s in src_patches(d, True) if s.endswith('cfg'))}; } Aug 22 22:51:22 (untested) Aug 22 22:51:49 endswith('.cfg') is probably better than endswith('cfg'), though Aug 22 22:52:03 yes quite the same line that the one written in the bbappend you have just show me the link Aug 22 22:52:20 yes agreed .cfg is probably better Aug 22 22:52:31 the ${@} there just grabs the .cfg filenames from SRC_URI Aug 22 22:54:27 yes Aug 22 22:54:44 worth noting that linux-yocto sets ARCH for the kernel when calling merge_configs: ARCH=${ARCH} merge_config.sh -O ${B} ${config_flags} ${configs} Aug 22 22:54:59 not sure if that's needed or not, but it's a possibility Aug 22 22:55:37 (turns out it does use merge_configs, just calls scc --configs to get the fragments Aug 22 22:55:39 ) Aug 22 22:57:20 ok Aug 22 22:57:30 all those info sounds good for me :-) Aug 22 22:57:52 should get you going the right direction, anyway :) Aug 22 22:58:02 :) Aug 22 22:59:00 thanks you very much for your help Aug 22 23:03:35 np **** ENDING LOGGING AT Tue Aug 23 02:59:58 2016