**** BEGIN LOGGING AT Fri Aug 21 02:59:58 2015 Aug 21 03:04:35 there we go, added rootfs_check_host_user_contaminated to image.bbclass Aug 21 06:59:14 hi guys Aug 21 07:53:11 can anyone help me to understand this please: http://pastebin.com/hdFS1ucr ? i tried almost everything Aug 21 08:03:23 drou: I can only think you've got bad linking going on and rpm is picking out that bad path and adding it as a dependency Aug 21 08:03:51 drou: you absolutely shouldn't have that full path showing up anywhere near packages Aug 21 08:05:41 bluelightning: bad linking in the shared lib produced ? not sure to understand Aug 21 08:06:05 Hi guys! I have a geode board and everything looking fine, but I have some issue with python3. http://pastebin.com/Ui0Rwmfn This code breaks on python3.4, but works fine with python2 where should I request more info or/and open issue? Aug 21 08:06:50 on desktop version of python3.4 everything is fine of course Aug 21 08:08:14 drou: yes, in the library that you are linking that's going into the package Aug 21 08:08:39 bluelightning: so probably something wrong with the do_compile in the recipe Aug 21 08:08:47 drou: that would be my guess yes Aug 21 08:09:21 pohmelie: breaks how ? Aug 21 08:10:57 bluelightning: I have a set of 'TARGET_CC_ARCH_append_pn- = " -fPIC -pie " ' placed inside a packagegroup.bb file. Is there a better way to do it ? Aug 21 08:11:26 milan_: inside a packagegroup? I don't see how that could work Aug 21 08:11:54 milan_: you'd need to do that from the configuration level, probably a .inc file you can "require" from your custom distro config Aug 21 08:13:41 bluelightning: http://pastebin.com/cvc3TKQV Aug 21 08:14:10 bluelightning: actually i try to make a shared library based on a library in which the Makefile only provides a static one Aug 21 08:14:43 pohmelie: hmm, I don't know what would cause that Aug 21 08:15:10 drou: I remember the discussion Aug 21 08:16:21 bluelightning: This is a very specific requirement for a particular image and customer. So will it be possible to isolate it the required way when used from distro config? Aug 21 08:17:17 we dont want to add an extra layer here Aug 21 08:17:49 bluelightning: should I raise issue on bugtracker? Or make some tests? Aug 21 08:17:53 milan_: this cannot be done from the image, you will need to keep the distro config (or at least, the TMPDIR produced from the build with this config enabled) separate for this particular customr Aug 21 08:18:00 pohmelie: yes please file a bug Aug 21 08:46:27 bluelightning: i guess there is no way to know which shared libraries are needed just by reading the previous static library ? Aug 21 08:52:37 Hello Aug 21 08:52:59 How can I use bitbake menuconfig with gcc ? Aug 21 08:56:18 looks like it didn't help :( Aug 21 09:06:33 Minipada, what do you mean with gcc? Aug 21 09:07:46 I would like to edit gcc options Aug 21 09:08:23 Minipada, which options? options outside of the kernel config? Aug 21 09:09:40 You mean, gcc options are chosen in the kernel ? Aug 21 09:09:56 Minipada, no, bitbake has its own gcc options Aug 21 09:10:33 menuconfig is used to create the kernel, but everything else is going to use bitbake gcc options to compile Aug 21 09:11:46 i wasn't aware that gcc has a menuconfig interface. probably it just has none, and everything has to passed in through options. Aug 21 09:12:21 menuconfig is a linux kernel target Aug 21 09:12:30 Oh ok then, if I want to edit floating points unit and tune for CPU, I would have to modify recipes ? Aug 21 09:12:56 Minipada, usually you can change the bsp options Aug 21 09:14:22 Minipada: now that sounds way more like tuning the machine configuration. Aug 21 09:15:11 Ok, because I used to do that with "make menuconfig" with crosstoolng. OK, now I see Aug 21 09:15:15 Minipada: have a look at the conf/machine directories in the layers that come with poky to see how this works. Aug 21 09:16:09 Minipada: ah. then its probably crosstoolng providing such an interface. but in openembedded, you usually do not tinker the tollchain directly, but have it derived from the required machine configuration. Aug 21 09:20:17 there are also ways to apply it in local.conf Aug 21 09:20:40 redengin: .. which is hacky and in fact only good for quick tinkering and testing. Aug 21 09:21:13 yeah, but with Minipada's understanding, it may be better to test it first :) Aug 21 09:21:52 oh, not necessarily. i find deriving a machine that only changes one or two config flags rather easy, actually. Aug 21 09:22:33 until you commit breakable changes to the build server Aug 21 09:23:13 well, testing locally before commiting is not really something that is applicable only here. i take that as granted. Aug 21 09:25:25 my experience has been that changing upper layers doesn't always make bitbake do the right thing once a previous build exists Aug 21 09:26:24 which again stems from the incorrect use upon an existing build. nothing that necessarily comes from the buildserver<->local difference, or would advocate using local.conf Aug 21 09:26:53 in short: do proper local testing before commiting. Aug 21 09:27:17 my experience has been that local.conf causes bitbake to correct itself moreso than editing upper layers Aug 21 09:28:22 but I've not used it in awhile, my practices may be obselete Aug 21 09:28:27 so you tinker up something in local.conf, do a non-clean build, call it tested, then throw it in a machine file and commit after that? Aug 21 09:28:46 sorry, but thats doesn't sound like a very recommendable practise. Aug 21 09:30:16 from a process standpoint, I want a test on a new build, to run equally well as a clean build, editing local.conf seemed to give expedient results Aug 21 09:30:52 I can back out a local.conf faster than a full bsp change Aug 21 09:31:06 i mean, whatever works for you. but its nothing that i would hand out as a best practise. Aug 21 09:32:06 LetoThe2nd, are you suggesting that yocto will only "work" on clean builds? Aug 21 09:32:38 redengin: i am suggesting that i personally only will call it tested if a clean build works as expected. Aug 21 09:33:59 I agree, but I had really hoped all of bitbake's dependency management would actually payoff Aug 21 09:47:14 redengin: I think mostly it will; that's not really the issue; the issue is local.conf is only meant to have "local" settings in it; the idea is you put the majority of configuration into layers as appropriate and then setting up that configuration again on a different build machine is a trivial exercise Aug 21 09:58:32 morning all: i have a little doubt. i have a running yocto on edison... i add repo.opkg.net but i don't find nginx. i found it on openembedded git but... i don't understand.... Aug 21 09:58:47 i havee to create a new image ? Aug 21 09:58:57 or i can compile/install it within edison yocto Aug 21 09:59:02 ? Aug 21 10:12:28 int: this is a question for the Intel maker forums really Aug 21 10:12:45 int: but if I had to guess I would say the feeds you're adding aren't compatible Aug 21 10:13:04 there *are* feeds out there that are compatible though, search the maker forums, it's been discussed at length Aug 21 10:13:24 bluelightning: talking about feeds at lunch time is.... weir Aug 21 10:13:27 d Aug 21 10:13:34 ok... and without feed i have to recreate the image for every software i want? Aug 21 10:14:34 int: you'd have to write a recipe and build it, yes Aug 21 10:14:45 fuckme :\ Aug 21 10:14:56 like slackware 2003 :P Aug 21 10:15:18 12:14 < int> fuckme :\ .... erm no, negative. Aug 21 10:15:22 thanks bluelightning Aug 21 10:15:31 LetoThe2nd :P Aug 21 10:26:04 Hi, I'm trying to install pip from a core-image-minimal, but getting a python erro ImportError: No module named pkgutil Aug 21 10:26:05 Aug 21 10:26:08 any ideas? Aug 21 10:26:28 pkgutil is supposed to be part of python itself Aug 21 10:27:12 ljose: "python itself" still can be splitted into a lot of subpackages Aug 21 10:27:48 ljose: have a look at the deploy packages directory and see if you're maybe just missing somthing. Aug 21 10:29:43 ljose: plus, there is a python-pip recipe available which should hopefully have correct dependencies etc. Aug 21 10:29:59 ljose: http://git.yoctoproject.org/cgit/cgit.cgi/poky/tree/meta/recipes-devtools/python/python3-pip_6.1.1.bb Aug 21 10:30:44 I'm using Python2 there pip is not include in Python Aug 21 10:31:14 but I see there is a python-pkgutil rpm in deploy packages Aug 21 10:34:39 is there a YUM like command to install the RPMs? Aug 21 10:35:11 if you didn't explicitly include it, there isn't even rpm Aug 21 10:35:44 mind, oe/poky do *not* automatically provide a package feed that can be put onto some distrubution server or such. Aug 21 10:36:28 I included rpm, but was looking for something that can install the dependencies from a repository automatically Aug 21 10:36:56 see, you're implicitly assuming that there even is something like a "repository" Aug 21 10:37:06 the image created one Aug 21 10:37:29 is it really a repository, or just a directory with a bunch of rpms inside? : Aug 21 10:37:45 tmp/deploy/rpm/all/repodata/ Aug 21 10:38:36 ah didn't know it creates something for rpm there. Aug 21 10:38:55 I guess it is used to install the packages when the imagen is created Aug 21 10:39:17 possibly. Aug 21 10:39:48 still, the straightforward way should be to just rebuild your image with IMAGE_INSTALL and friends adjusted to what you need. Aug 21 10:40:12 yes I will do that Aug 21 10:40:43 yes, it does indeed produce feeds Aug 21 10:41:05 for the purpose of constructing the image Aug 21 10:41:23 you can also build other recipes then "bitbake package-index" to update the index in the feeds Aug 21 10:46:52 ljose: if you use rpm packages and enabled tehe package-management image feature, then you'll have smart in the image Aug 21 10:47:09 and your deploy/rpm directory is a rpm feed Aug 21 10:56:57 rburton: Thanks that makes sense Aug 21 11:00:23 as bluelightning said, when an image is created the feed is updated, but if you build a package and then want to use it in the feed you'll need to bitbake package-index to update the feed metadata Aug 21 11:40:21 bluelightning: Should it be possible to set TARGET_CC_ARCH using any OVERRIDES value ? e.g. TARGET_CC_ARCH_append_foo where foo is listed in DISTROOVERRIDES or DISTROFEATURES ? Aug 21 11:52:39 bluelightning: finaly found the issue, it was something in the recipe that depends on nmea. Aug 21 12:29:07 milan_: it would have to be in OVERRIDES (perhaps via DISTROOVERRIDES); but yes Aug 21 12:29:43 drou: ok, but any idea how it ended up with the full path in there? Aug 21 12:29:55 drou: I'm just wondering if we need to add some kind of QA check Aug 21 13:28:24 hello all Aug 21 13:48:13 bluelightning, RP: how to find a path to certain binary from the sysroot? Aug 21 13:48:34 i'd like to run rpm from there but don't know how to correctly get the path of the binary Aug 21 13:50:13 is thr binary in the PATH? then it's 'which' Aug 21 13:50:51 or are you talking about compiling or something? gcc can be asked where certaining libraries and other components are.. (don't remember the exact arguments off hand) Aug 21 13:51:08 i'd like to run that from inside a python function in bitbake Aug 21 13:51:49 or prolly i should use fakeroot for running it Aug 21 13:51:54 if you can do it on the shell you can do it in python.. figure out the shell symantics.. then translate to python or something ele Aug 21 14:06:55 fray: i need to run the rpm binary from inside python and i'd like to know the correct way to determine it's path (from inside the same python function) Aug 21 14:09:15 I thought the sysroot path was already in the python env if you are running as a python task in a recipe. Aug 21 14:09:37 but maybe I'm just imagining that . Aug 21 14:10:01 ya, it's available in bitbake var store.. Aug 21 14:10:57 yeah thanks, i think STAGING_BINDIR_NATIVE is what i'm looking for Aug 21 15:49:53 i have built master with meta-security-smack included in my layer stack (MACHINE=qemuarm) Aug 21 15:50:13 anyone know how to actually enable it? Doesn't seem to be there after booting qemuarm machine Aug 21 15:50:43 must be missing some steps or configuration somewhere Aug 21 15:51:10 challinan: https://github.com/01org/meta-intel-iot-security/tree/master/meta-security-smack#i-adding-the-security-smack-layer-to-your-build Aug 21 15:51:18 aka the README.md Aug 21 15:51:25 * challinan looks Aug 21 15:52:06 of course it also only appends linux-yocto, but that should be okay for you sinc eyou're building qemuarm Aug 21 15:52:35 indeed, my config did seem to patch my kernel Aug 21 15:53:24 Unrelated, but here we are, finally created a ROOTFS_POSTPROCESS_COMMAND function analagous to the host-user-contaminated package QA check: https://github.com/kergoth/poky/commit/ce8e6b0 Aug 21 15:54:10 hmm, maybe it didn't patch my kernel. Well, I have enough to proceed. Thanks kergoth Aug 21 15:55:45 i think it's more configuration than anything Aug 21 15:55:46 bb show -r virtual/kernel SRC_URI | grep -i smack Aug 21 15:55:51 from a quick look at the bbappends Aug 21 15:56:22 yeah :( the smack layer is patching an earlier kernel version Aug 21 15:57:44 should be straightforward to copy one of the existing appends to the new name, i would think Aug 21 15:57:49 at least assuming the configuration hasn't changed Aug 21 15:57:56 or pin the kernel version ? Aug 21 15:58:33 yeah, true Aug 21 15:58:41 * kergoth gets caffeine and breakfast Aug 21 15:59:38 RP: thoughts on my host user contamination checks? to add one to image.bbclass for ROOTFS_POSTPROCESS_COMMAND, i can't define the variables which call os.getuid() / os.getpid() in that context, even with :=, since it's parsed for the task with pseudo already running -- need it in a global config file or INHERIT'd class. I figured bitbake.conf, but I'm not a big fan of adding more variables there Aug 21 16:15:44 kergoth: where are those? I'm clearly missing something :/ Aug 21 16:16:01 kergoth: not having a good day/week :( Aug 21 16:17:11 kergoth: immediately expanded variable somewhere like image.bbclass comes to mind but I'm not 100% of the problem you're solving... Aug 21 16:18:16 RP: https://github.com/kergoth/poky/commit/ce8e6b06e3e225b1fb58bbaac95380ccaf3564a9 - if i define the variables it needs in insane.bbclass, the values will be 0/0, not the real user running bitbake, since insane.bbclass is parsed by the task under pseudo, hence https://github.com/kergoth/poky/commit/cc81b6ba29cc8a7ab313591225e12eb1a2b1c02a Aug 21 16:18:47 just wondering if you have an off-the-cuff better suggestion Aug 21 16:22:26 RP: :( Aug 21 16:23:03 kergoth: executing something with PSEUDO_UNLOAD=1 is the only other thing I can think of Aug 21 16:23:12 (to get the real uid) Aug 21 16:24:26 kergoth: Like you, I'm not keen on bitbake.conf but I can't see many other ways to do this Aug 21 16:25:14 Hmm, that's true. could do that in the rootfs function, but would still want to use the variables in insane.bbclass, so as to avoid calling processes at each path callback. might try that Aug 21 16:25:32 thanks Aug 21 16:27:49 interestingly, the check found something right off the bat, one of the opkg feed files is owned by the user running bitbake Aug 21 16:27:50 weird Aug 21 16:27:55 just one of them, though Aug 21 17:05:42 Hello Aug 21 17:06:13 Is there a way to use private repositories when writing recipes ? Aug 21 17:06:33 I read about SRC_URI, but without success Aug 21 17:07:12 when do RCONFLILCTS_${PN} = "foo (<= 2.3)" checks get run? Aug 21 17:07:25 doesn't prevent a do_compile failure... Aug 21 17:07:27 Minipada: you mean github private repositories? Aug 21 17:07:38 Minipada: it's no different than any other repo, you just need to deal with authentication Aug 21 17:07:50 bitbake won't prompt for username/password, so you have to deal with that externally Aug 21 17:08:05 if you get a git clone from teh url at a prompt to work without prompting you for authentication, then it'll work with bitbake Aug 21 17:08:28 e.g. add your ssh key to your agent, and set ;protocol=ssh on the url or use a git url replacement to get it to use ssh to clone Aug 21 17:08:50 or set ;protocol=https and put your auth in ~/.netrc and chmod 0600 that Aug 21 17:09:47 Yes, for github. Ok I'm gonna try it Aug 21 17:10:04 i've used both of those methods successfully in the past Aug 21 17:10:10 actually use them on a daily basis Aug 21 17:10:52 might be possible to get git to prompt using an x11 app or something, but never messed with that Aug 21 17:11:13 Ok thanks :) Aug 21 17:38:44 I have a yocto build from someone else that builds on a USB image (exFAT) and loads as a loop mount. I need to convert to an SSD rootFS. After much searching, I have not been able to find the config that sets the image style. Can soemone point me in the right direction please. Aug 21 18:03:55 so i found an obvious but stupid mistake in my pseudo patches, which is that I'd commented out the code that let it allow fsync operations due to performance testing, and forgotten to remove that. Still checking for other stuff that might also be busted. Aug 21 18:03:59 At least one buildhistory test seems to show files showing up with the wrong UIDs. Aug 21 18:26:29 seebs: not related, but you may find https://github.com/kergoth/poky/commit/2b4e7f831f7c2c9892d0403ebd1ade9a8f498d40 and https://github.com/kergoth/poky/commit/ed1225af91d1f1b32df5fbd0d5b42a9c3f677a71 of interest Aug 21 18:27:14 Oh, neat! Aug 21 18:27:19 That would actually be sorta helpful. Aug 21 18:27:52 nice to spot them as early as possible, rather than having to poke after the fact Aug 21 18:27:54 ugh, seriously? if dropbox can't connect to its servers, it'll just keep trying rapidly and consuming >75% cpu, unless you pause syncing Aug 21 18:28:42 .... neat Aug 21 18:30:11 trying out this mac app TripMode, which when enabled only allows network access to your specified apps, adn blocks everything else, and remembers the enabled/disabled state per-network-interface / wifi IP, so you don't waste data on your phone or consume bandwith at the coffee shop Aug 21 18:30:15 and watched dropbox crap itself Aug 21 18:30:17 heh Aug 21 18:30:23 Oh neat. Aug 21 18:31:37 ... okay that's not-good Aug 21 18:31:56 so my test on core-image-minimal, only visible change appears to be something changing from group shutdown to group root. Aug 21 18:32:09 huh Aug 21 18:32:10 on core-image-base I was seeing much larger changes. That should not be happening, I think. Aug 21 18:32:55 I'm sort of suspecting that something is not archiving extended attributes when it now needs to to work. Aug 21 19:23:19 hmm, need to poke at ostree one of these days Aug 21 19:25:00 someone should really shut up these libpvr2d warnings from meta-ti Aug 21 19:25:45 kergoth: which ones? rdeps? Aug 21 19:25:54 yeah, file-rdeps Aug 21 19:26:08 i'm assuming its due to the use of PRIVATE_LIBS, so it can't map it to the package Aug 21 19:26:17 haven't looked into it, though Aug 21 19:29:06 well, I found part of the problem, anyway. Not all of it, but enough to explain at least some leakage. Aug 21 19:29:55 Idle question, does anyone know of places in bitbake that attempt to interact with pseudo databases, or know how sstate interacts with xattrs in the filesystem? I'm thinking I should be reviewing any code that might think it knows how pseudo databases are stored, because something seems to be Acting Up. Aug 21 19:30:54 richard's probably your best bet there Aug 21 19:31:03 i don't know of anything personally, but that doens't necessarily mean much Aug 21 19:32:02 So the one I found is: If you have xattr support enabled (which you have to for the thing which stores attributes in xattrs), a SET_XATTR operation will create a dummy link in the file database... with invalid data. This seems to be actually harmless in normal use, since we never look at that, and just use the xattr values anyway. Aug 21 19:32:19 But if something were, say, copying the files and dropping the extended attributes, then relying on the database, things would go wrong. Aug 21 21:02:33 wow. bb.fetch.build_mirroruris ends up trying to create DL_DIR. Aug 21 21:02:34 that's just sad Aug 21 21:02:42 I haven't tried to fetch a damn thing Aug 21 21:03:53 hmm may be the code needs to move out ? Aug 21 21:16:44 yeah, the localpath method of the local fetcher creates dl_dir Aug 21 21:16:44 * kergoth rolls eyes Aug 21 21:40:43 Hmm, in revamping my sstate distro fallback bits, should i rip out obvious local fallbacks from SSTATE_MIRRORS and apply that to the standard mirrors in SSTATE_MIRRORS, which would only handle specific cases, or use a separate variable to hold the distro fallback info.. Aug 21 22:26:06 khem, ping Aug 21 22:36:02 rburton, for nodistro mtd-utils fails after the xattr change Aug 21 22:37:12 urgh, really? Aug 21 22:37:56 yep Aug 21 22:38:24 I'm not up-to date..do I have to adjust something else? Aug 21 22:38:54 how does it fail? Aug 21 22:39:05 I've reverted 24fde4d and it builds Aug 21 22:39:07 mom Aug 21 22:39:07 i was reliably informed there were no extra build deps by enabling xattrs Aug 21 22:41:03 | LD ftl_check Aug 21 22:41:03 | CC mkfs.jffs2.o Aug 21 22:41:03 | mkfs.jffs2.c:70:21: fatal error: sys/acl.h: No such file or directory Aug 21 22:41:03 | #include Aug 21 22:41:17 note: nodistro Aug 21 22:41:34 distro shouldn't make a difference Aug 21 22:42:38 btw mtd-utils 1.5.2 is out Aug 21 22:43:48 native or host? Aug 21 22:43:56 erm, native or target? Aug 21 22:45:13 seems target Aug 21 22:45:22 my target sysroot has sys/acl.h in Aug 21 22:45:24 btw btw, your comment here? Aug 21 22:45:26 http://patchwork.openembedded.org/patch/99971/ Aug 21 22:45:54 about distro featyre I mean Aug 21 22:46:17 couldn't decide if it was worth it considering the build dep impact was nil Aug 21 22:47:42 hm the build deps bit was a lie Aug 21 22:47:50 o, sorry it is -native Aug 21 22:47:55 ant_home: remove your revert and bitbake acl first Aug 21 22:49:06 ant_home: fancy adding the acl dep (and maybe even making it respect the distro feature?) Aug 21 22:49:11 argh..no, it is for target..bah Aug 21 22:49:54 ok, let's try as you say Aug 21 22:53:05 yes, needs acl Aug 21 22:54:16 in the deps Aug 21 22:54:33 patch please :) Aug 21 22:56:25 hmm acl is a distro feature, isn't? Aug 21 22:56:44 a quick grep said not a default one at least Aug 21 22:57:13 oh you're right Aug 21 22:59:17 we should probably turn that on in poky then Aug 21 22:59:38 khem, worth adding acl to default-distrovars? Aug 21 23:00:53 rburton, let's bother RP and khem... Aug 21 23:01:25 ant_home: you send the patch for mtdutils and i'll file a bug for distro features Aug 21 23:01:29 fwiw this is the single recipe failing for core-image-base / nodistro Aug 21 23:02:27 ok Aug 21 23:02:39 good, the autobuilder does builds of nodsitro every time we do a build on it Aug 21 23:16:51 oh my...google.com :/ Aug 21 23:24:03 Allow less secure apps: ON Aug 21 23:24:12 it resets itself..darn Aug 21 23:45:19 gn Aug 22 02:06:16 I had an idea for a bitbake feature earlier. I'm not convinced it's a great idea, but it'd definitely fill a need for something i've had to do on quite a few occasions. I'm thinking of a filter function flag on a variable. sort of a postprocessing function. not before, not after. not prepend, not append, but when the variable is expanded, its value is passed into any postprocessing functions to modify the value Aug 22 02:06:53 i've had to do that quite often, e.g. set up anonymous python or an event handler to do a getvar, modify/alter/add/remove/whatever, then setvar, but when you do that, it's bound to a particular point in time, and it forces early expansion of the variable Aug 22 02:07:16 whereas the proposed function would simply alter the value at expansion time Aug 22 02:57:04 hmm, sanity.bbclass checking of SSTATE_MIRRORS is just broken. last i checked, there's no requirement that the replacement be absolute. the original url isn't absolute, there's no reason you couldn't alter part of it Aug 22 02:58:01 in fact, i'm doing so Aug 22 02:58:15 hmm **** ENDING LOGGING AT Sat Aug 22 02:59:58 2015