**** BEGIN LOGGING AT Wed Oct 09 02:59:58 2013 Oct 09 06:42:58 good morning everyone i have a question: my build fails because i have no space left on /, opkg trys to copy the kernel-vmlinux to my /tmp, is that correct or shoud it be copied to a different location and i configured s.th. wrong ? Oct 09 06:43:25 oh and my build-root-dir is on a seperat partition Oct 09 06:43:34 it has space left Oct 09 07:47:12 good morning Oct 09 08:12:09 good morning everybody Oct 09 08:12:34 I'm having a problem with flex. Oct 09 08:13:44 I've built an sdk image and when building a package directly on the board I'm getting: Oct 09 08:13:49 flex: fatal internal error, exec of /opt/yocto_builder/build/tmp/sysroots/x86_64-linux/usr/bin/m4 failed Oct 09 08:14:48 /opt/yocto_builder/build/tmp/sysroots/x86_64-linux is clearly a path from my Yocto build host, so it shouldn't go inside a target binary. Oct 09 08:17:00 I think the problem is in the M4 variable here, but I don't fully understand what the recipe is doing: http://cgit.openembedded.org/openembedded-core/tree/meta/recipes-devtools/flex/flex.inc Oct 09 08:20:46 mmm... something has already been done, but doesn't fix the problem completely: https://github.com/openembedded/oe-core/commit/7bdb617f2f0e246feb4dc32931fdb87258fd1207 Oct 09 09:09:22 morning all Oct 09 09:09:32 hi bluelightning Oct 09 09:09:39 hi RP Oct 09 09:12:10 RP: Regarding the discussion we had the other day about a way to prevent variables to be affected by the OVERRIDES mechanism, would something like this be ok with you: http://pastebin.com/cG6nGBeg Oct 09 10:03:26 Saur: I'm not thrilled by it :/ Oct 09 10:06:19 RP: Well, to me it seems like a lot better solution than having to append _foo_foo to each variable that I want to protect from foo being in OVERRIDES... The flag name "no_overrides" may not be the best, though... Oct 09 10:09:17 Saur: Its a better solution for your particular problem, I just worry about adding things like this into the core Oct 09 10:10:29 RP: Hmm, ok... I can understand that. Oct 09 11:24:31 heh.. Gentoo a well Oct 09 11:24:33 Linux systems which have / and /usr on separate file systems but do not Oct 09 11:24:33 use an initramfs will not be supported starting on 01-Nov-2013. Oct 09 11:39:45 Has anyone else had issues with dropbear? I'm using the latest recipe in poky Oct 09 11:40:04 's git and I get a constant oops if the connection goes up and down Oct 09 12:51:04 pev: no problem here. Though I'm using a stable local network Oct 09 12:57:12 is 1.5 closed for patches now? Oct 09 13:03:14 Net147: is there an issue that you've found? Oct 09 13:04:23 bluelightning: I have a backport for Qt painting regression fix Oct 09 13:08:21 Net147: I think it's probably too late for that, but send the patch and we can get it in for 1.5.1 Oct 09 13:09:50 will dora support http://lwn.net/Articles/569832/rss Oct 09 13:09:58 GNUMake 4.0 #ftw! Oct 09 13:10:08 Crofton|work: will we upgrade to it moments before release? no :) Oct 09 13:10:32 I really want to instrument RP blood pressure and graph it :) Oct 09 13:10:32 well, s/moments/a few days/ Oct 09 13:11:03 "New features include the integration of the Guile extension language" Oct 09 13:11:08 rofl Oct 09 13:11:18 I was just being amused by that Oct 09 13:11:23 I feel a bit unwell... Oct 09 13:11:26 lets hope that does not become popular .. Oct 09 13:11:29 bluelightning: hi. Could you kind have a quick look at the flex recipe? Thanks https://www.yoctoproject.org/irc/%23yocto.2013-10-09.log.html#t2013-10-09T08:12:34 Oct 09 13:11:46 *kindly Oct 09 13:12:22 fles is trying to use a build path on the hardware? Oct 09 13:12:49 Crofton|work: yes. If i grep the flex binary it matches that host path Oct 09 13:13:11 and it actually tries to use it when running in the target Oct 09 13:13:13 panda84kde: looks like hrw fixed this a while ago, remind me what version of the build system are you using? Oct 09 13:13:30 obviously you need to put a copy of m4 there :) Oct 09 13:14:34 Crofton|work: hrw's fix actually suggested that as a workaround believe it or not :) Oct 09 13:14:54 sometimes, you need to get the job done Oct 09 13:15:13 bluelightning: master from a couple of weeks ago, but flex recipe is aligned to this latest commit: http://git.yoctoproject.org/cgit.cgi/poky/commit/meta/recipes-devtools/flex?id=990deb3b894806b512f061759e11e9aa84b2d98a Oct 09 13:15:20 oh wait Oct 09 13:15:27 any word on whether or not the new make-4.0 improves on the whole make-3.82 fiasco? Oct 09 13:15:27 the fix was basically reversed Oct 09 13:16:14 Crofton|work: I "did the job", but I though it would be good to actually fix the problem ;) Oct 09 13:16:26 right Oct 09 13:16:46 what is scary is when fixes like that get recorded on wikis Oct 09 13:18:34 by "fixes" you mean "create the dir a link to m4"? Oct 09 13:23:52 bluelightning: okay, done Oct 09 13:25:33 Crofton|work: there are much more difficult things affecting my blood pressure than that. Its a simple "no" ;-) Oct 09 13:26:47 but think about the guile supprt Oct 09 13:27:03 * bluelightning tries not to think about it Oct 09 13:27:33 Crofton|work: we're working to remove it from core, don't worry! Oct 09 13:30:31 I wonder if anyone got around to documenting the DISK_SIGNATURE variable... Oct 09 13:31:03 Net147: I don't think so :( Oct 09 13:31:22 bluelightning: easter egg then =P Oct 09 13:32:22 Net147: since you implemented it it would be great if you could send some raw material to Scott Rifenbark to add to the docs; it won't make 1.5 but at least it'll be in the next version Oct 09 13:33:24 bluelightning: I think it was described reasonably clearly in the commit message for 4382a419b4c90312d22aa55ff535b45bcf704716 Oct 09 13:33:53 bluelightning: and he was CC-ed on it if I remember correctly Oct 09 13:34:09 Net147: ok, fair enough Oct 09 13:35:18 I've added it to my list, will follow up with Scott during the 1.6 cycle Oct 09 13:36:19 certainly useful though for creating bootable images that work regardless of storage configuration Oct 09 13:37:19 though I have been using NFS boot for development since I got tired of writing new image all the time Oct 09 13:42:06 is http://www.yoctoproject.org/docs/current/mega-manual/mega-manual.html up to date for 1.5? Oct 09 13:42:35 Net147: that's the stable documentation; since the 1.5 stable release hasn't happened yet that is still 1.4 Oct 09 13:42:45 bluelightning: so http://www.yoctoproject.org/docs/latest/mega-manual/mega-manual.html then Oct 09 13:43:07 Net147: right, replace current with latest or 1.5 Oct 09 13:43:42 bluelightning: is Yocto still tested with Atom hardware? Oct 09 13:44:47 bluelightning: given that Intel seems to have discontinued all their Atom motherboards Oct 09 13:45:08 Net147: I believe we test the new genericx86 BSP with atom-based platforms, and we certainly test the specific atom-based BSPs with their respective hardware Oct 09 13:48:11 I'm not in the same division that deals with the atom boards you refer to so I can't really comment on that, but we definitely still support atom as a platform; we still have many BSPs for atom-based systems and no doubt we will continue to add more Oct 09 13:57:23 Is it possible to only recompile a specific kernel module and package the rpm? Oct 09 14:03:31 How to add 'libeigen' to yocto? Oct 09 14:05:19 libeigen recipe can be found at https://github.com/kartikmohta/meta-km/tree/master/recipes-extended/eigen Oct 09 14:05:33 did you search layers.openembedded.org Oct 09 14:07:04 no, I found it. will try adding it. Oct 09 14:07:16 in layers.openembedded.org Oct 09 14:10:19 lol, I first read that as lawyers.openembedded :S :D Oct 09 14:29:56 after adding libeigen to meta-oe does it include in my image or I should add libeigen to my Image? Oct 09 14:32:22 zaif: you need to add it to your image Oct 09 14:47:30 bluelightning: Thank You and done. Oct 09 14:47:43 how do I keep all my recipes in yocto updated? Oct 09 14:48:05 zaif_: depends, how do you define updated? Oct 09 14:50:27 libeigen was added to meta-oe some days back(i think), some recipes gets update in future, so how do I maintain updated recipes in my yocto? Oct 09 14:51:00 i cloned yocto 2 months back and was working on that Oct 09 14:51:04 zaif_: you'd have to bring in those updates manually Oct 09 14:52:07 ok. Got it. So, when if there's any error on a recipe, then update it automatically. Oct 09 14:52:19 update it manually* Oct 09 14:52:39 well, it's not always going to be a case of error occurs -> update recipe Oct 09 14:53:40 ok Oct 09 14:53:55 heh, usual case is error occurs -> check your own changes Oct 09 14:55:45 or when something changes in oe-core and you have to adapt your recipe Oct 09 15:54:26 khem`: you around? Oct 09 15:55:12 bluelightning: thanks for reporting the flex bug. Oct 09 16:00:41 panda84kde: no worries Oct 09 16:15:46 Hi all, can I use glob jockes in SRC_URI, like file://*.h ? Oct 09 16:15:52 *jockers Oct 09 16:19:24 globs should work in file uris, yes. we generally don't advise it, so it might be better to put the in a dir and add the dir to src_uri, but it should work, regardless Oct 09 16:20:02 kergoth, all my files are in the "files" dir Oct 09 16:20:14 how to instruct bitbake to stage all these files? Oct 09 16:21:41 cp: cannot stat `/home/drasko/yocto/meta-drasko/recipes-test/test/files/*.h': No such file or directory Oct 09 16:21:54 if the glob isn't working, then do what i just suggested Oct 09 16:22:02 put them in a subdir under files instead, and add that subdir Oct 09 16:36:56 Hey all. Oct 09 16:36:58 I've got a build server (buildbot) setup to build a nightly / release build. The release will trigger all the machines to build based on a manifest file Oct 09 16:37:05 I've got a build server (buildbot) setup to build a nightly / release build. The release will trigger all the machines to build based on a manifest file Oct 09 16:37:18 For the nightly, I was thinking of checking the commit messages to know which machine to build. Is there a better way to do this? I'd rather not rely on the developer checking in his/her code with the key "IMX53-BUILD" or "IMX28-BUILD". Oct 09 16:37:25 (sorry about the double post there) Oct 09 16:48:11 is there a better place for me to ask build-type questions? Oct 09 16:48:12 :) Oct 09 17:27:29 khem`: meet laur1 Oct 09 17:28:02 khem`: laur1 has been working tirelessly to reproduce (and try and create a good reproducer) of the g++ issue Oct 09 17:28:08 OK Oct 09 17:28:41 khem`: as I just told sgw_, I already tried an older kernel Oct 09 17:28:43 laur1: take it away Oct 09 17:29:02 laur1: yes Oct 09 17:29:25 laur1: symptoms are that corrupted obj is generaed Oct 09 17:30:23 khem`: yes, but I ran cc1plus manually in both cases (256MB and ulimit'ed to 128) and the object is the same Oct 09 17:30:41 khem`: I compared the preprocessed file and the assembly file, the same Oct 09 17:31:06 khem`: I'm talking about one file here: board.cpp Oct 09 17:31:38 khem`: this is where the linking fails Oct 09 17:33:44 khem`: the thing is that, when using -frepo, the file should be re-compiled several times in order to generate the .rpo file with all the template instantiations Oct 09 17:34:32 khem`: the thing is, in 256MB, the re-compilation is done only once... that's the mistery I couldn't figure out Oct 09 17:47:14 ok so asm file is exact same Oct 09 17:47:28 which says to me that gcc is out of picture for first pass Oct 09 17:50:31 khem`: btw, I did another dirty test: I used the 4.7 compiler but changed only the cc1plus from 4.8 (I also had to rename the gcc include directory to 4.8.1, etc). Got it reproduced... Oct 09 17:52:43 so you took 1.4 rootfs completely Oct 09 17:52:47 or how did you do it Oct 09 17:55:37 anyone ever work with a recipe where the application/library does not install under /usr? Oct 09 17:55:55 i.e. /opt/package_name-ver/ Oct 09 17:56:58 khem`: I've been poking around this too and its something to do with the .rpo files afaict, the "C" (Chosen) bit doesn't seem to stick Oct 09 17:57:32 khem`: well, first I tried to set PREFERRED_VERSION_gcc to 4.7 but it was a pain... so I compiled with 4.8 (even for target) and put 4.7 compiler manually. Gave it a test, and it compiled just fine Oct 09 17:57:38 khem`: Am I right in thinking that get_identifier("XXXX") should return the same pointer throughout the life of a process? Oct 09 17:57:48 khem`: that I replaced cc1plus, bang! Oct 09 17:57:59 khem`: or if not, what resets the identifiers? Oct 09 17:58:45 RP: i think it will change since frepo is like comdat section folding Oct 09 17:58:59 laur1: OK Oct 09 17:59:14 laur1: pass -Wl,--no-demangle to linking step Oct 09 17:59:22 and see if it workarounds the problem Oct 09 18:02:41 khem`: I'm going to suggest it doesn't but will let laur1 reply for sure since my test environment is a bit messed up now... Oct 09 18:03:02 denix, hey! I'm trying to use meta-ti to build a panda image and getting some LICENSE field not set errors as well as a failure to parse the cloud9 gfx image. Reading the README and such now, but wondering if this is a known issue? Oct 09 18:06:53 RP: I am thinking dmangler might be confused with "C" symbols Oct 09 18:08:02 khem`: didn't work... Oct 09 18:09:28 I have a bold theory, it might be the garbage collector kicking in and removing a string with data we need... Oct 09 18:11:06 RP: sounds plausible but why on earth does it kick in when it has 256MB of RAM compared to 128MB? Oct 09 18:11:30 dvhart: yes, it's in README :) you need BBMASK, but I'm removing it soon... Oct 09 18:11:33 RP: it would've been more plausible if it happened for 128MB... Oct 09 18:12:34 laur1: Just thinking out loud. I'm going to put some more prints in and see... Oct 09 18:14:03 RP: ok Oct 09 18:14:47 hmm, ggc is called at the right point to break this... Oct 09 18:14:56 (according to the prints) Oct 09 18:15:20 RP: lets disable gc --param ggc-min-heapsize= Oct 09 18:15:27 denix, the README mentioned BBASK to avoid dep on systemd, but what I saw was missing LICENSE and failed parsing. Oct 09 18:15:59 but yes, ignoring those recipes does get things building. Thx. Oct 09 18:19:11 khem`: how do I make it pass that to the recompile ? Oct 09 18:19:19 dvhart: the description is slightly misleading these days, since systemd is long since merged into oe-core :) Oct 09 18:19:51 denix, thanks. Are you adding license data to the image recipes? Oct 09 18:23:40 dvhart: those image recipes used to pick up license data from another layer by including a base "systemd" image... Oct 09 18:24:12 denix, got it. Thanks :) Oct 09 18:24:18 dvhart: anyway, that part got restructured and no longer relevant, so I need to clean that up... Oct 09 18:28:38 laur1, khem`: I have a strong lead. In toplev.c, if I comment out the "ggc_protect_identifiers = false;" line, my failing compile works Oct 09 18:29:05 That is turn protects entries in stringpool from ggc Oct 09 18:29:45 s/is/in/ Oct 09 18:30:04 and my get_identities() call has a matching pointer afterwards :) Oct 09 18:32:22 hmmm seems likely Oct 09 18:34:46 RP: that use to be set to true IIRC Oct 09 18:36:01 RP: I did a git blame on the file and, apparently, it was false since 2008... Oct 09 18:36:09 RP: but, it's a good finding Oct 09 18:38:24 khem`: it starts off true but gets set to false after compiles finish Oct 09 18:38:37 khem`: in this case we don't want that since there is data in memory we need Oct 09 18:39:11 laur1: I think this is a really odd bug which only happens on memory constrained systems. Who compiles in 256MB ram these days? Oct 09 18:39:18 * mr_science puts a check mark in the bluelightning-owes-me-a-beer column Oct 09 18:41:58 people doing native builds Oct 09 18:43:16 RP: which of Oct 09 18:44:02 for frepo it goes into multile compiles Oct 09 18:44:17 does it share the memory across do_compile invocation ? Oct 09 18:44:35 khem`: Right, but we need the CHOSEN bit of lang tree's bit 6 Oct 09 18:44:52 khem`: We load the .repo file, compile, save the .repo file Oct 09 18:45:05 khem`: we need the data in memory from when we loaded when we save Oct 09 18:45:17 correct hmm Oct 09 18:56:13 khem`, sorry to disagree with you over gold Oct 09 18:56:38 but I'd like ANgstrom to stay close to the released OE-* stuff Oct 09 18:59:50 hmm ... there has to be a better way to do this - I have a set of packages that install to /opt/${pkgname}/{bin,include,lib}, and one of them has expat as a requirement, and it fails on configure Oct 09 19:00:28 doesn't pick up expat, and from what I see, it's because /usr/lib/pkgconfig is not in PKG_CONFIG_PATH Oct 09 19:01:38 I'm appending /usr/lib/pkgconfig in the recipes, and that appears to fix it, but feels like a kludge Oct 09 19:04:25 I'm assuming that is has to do with libdir being set to /opt/${pkgname}/lib, and so bitbake is only putting /opt/${pkgname}/lib/pkgconfig and /opt/${pkgname}/share/pkgconfig in PKG_CONFIG_PATH Oct 09 20:23:44 trying to understand how DEPENDS works ... package B has DEPENDS = "A" where A is a library package. What specifically happens when you bitbake B, in regards to package A? Oct 09 20:24:05 does all A get dumped out into the sysroot? Oct 09 20:24:31 or A-dev? Oct 09 20:25:40 both packages are built and the -dev one will stage headers in sysroot Oct 09 20:25:48 typically Oct 09 20:26:05 what if the headers are not based in /usr/include? Oct 09 20:26:06 DEPENDS is referring to a recipe Oct 09 20:26:18 not to a package Oct 09 20:26:22 like /opt/A/include Oct 09 20:27:03 it means you are using an external toolchain / sdk Oct 09 20:27:44 we have per-machine sysroot Oct 09 20:28:50 I don't believe so ... the package is one of our custom libraries. unfortunately, it's designed to be installed outside of /usr/. :( Oct 09 20:29:51 designed == hard coded Oct 09 20:30:24 please inspect the structure of the sysroot Oct 09 20:30:38 wmcdevel: you probably need to play with FILES_{PN}-dev so that your custom location for .h files is taken into -dev package Oct 09 20:31:10 there is a default value set in meta/conf/bitbake.conf, but if you have files outside the default, you can add them in your recipe. Oct 09 20:31:16 ndec: did that ... and verified that the -dev package has them where I'm expecting them to be Oct 09 20:31:38 you can create /opt in your sysroot if it is the point Oct 09 20:31:38 hmm. so, what is the problem in fact? Oct 09 20:32:46 ndec: trying to bitbake B, and none of the headers/lib/pkg-config info from A-dev is found. Oct 09 20:33:05 are there installed in sysroot? Oct 09 20:33:26 ndec: not seeing them when I bitbake B -c devshell Oct 09 20:34:00 ant_home: is creating /opt something I should do manually, or should that be done in a .bbclass somewhere? Oct 09 20:34:30 and you sure, they are in A-dev? even if they are in /opt, that shouldn't matter, i believe. Oct 09 20:34:45 ndec: positive Oct 09 20:35:00 JaMa: this evening I did pull and incurred in Your version of make 3.82 is broken. Please revert to 3.81 or install a patched version. Oct 09 20:35:10 ndec: working with debian packages, and dumped the contents of each one that recipe A generated Oct 09 20:35:12 JaMa: Gentoo has this make since June Oct 09 20:35:33 I hacked sanity.bbclass for the moment Oct 09 20:36:04 must be smthg recent, have you seen it? Oct 09 20:36:49 wmcdevel: well, that's unexpected (at least to me). Oct 09 20:37:24 ndec: that's what I was saying ... at least I'm not the only one Oct 09 20:38:01 yeah... hopefully someone can explain... Oct 09 20:40:47 khem`: http://git.yoctoproject.org/cgit.cgi/poky-contrib/commit/?h=rpurdie/t2&id=18ef9fd54e97cce7975d91f6af73e65ea2c01b21 Oct 09 20:41:08 is there a way to tell bitbake to dump the -dev package manually? Oct 09 20:46:43 RP: nice tracking that down! Oct 09 20:46:50 wmcdevel: is best worded here: Oct 09 20:46:52 http://www.yoctoproject.org/docs/1.5/ref-manual/ref-manual.html#var-DEPENDS Oct 09 20:48:52 ant: thanks. going to try and debug the do_populate_sysroot for the depenency and see what is happening Oct 09 20:49:38 I have a feeling that somewhere in that task the files I'm expecting to see dumped out are not being found/dumped for some reason Oct 09 20:54:22 RP: do you remember offhand any recent commit enhancing the make 3.82 sanity check? It's a first on Gentoo this evening Oct 09 20:54:59 ant_home: not recently Oct 09 20:55:07 well, strange then Oct 09 20:55:25 I have this make since June Oct 09 20:57:39 RP: http://paste.debian.net/54579/ Oct 09 20:58:23 quick hack found on the web, probably defeating the purpose Oct 09 20:58:31 ant_home: there are several make bugs Oct 09 20:58:42 ant_home: so its possible that gentoo is missing a patch Oct 09 20:58:48 but hey hasn't gentoo got make 4 yet? Oct 09 20:59:05 my point is, what has changed in oe-core? Oct 09 21:00:01 I'm trying to find my old rev but it was pulled just few days ago Oct 09 21:00:01 ant_home: I'm using 3.82-r4 and haven't seen any issues with it Oct 09 21:00:06 oh Oct 09 21:00:55 same here since months Oct 09 21:01:19 Fri Oct 5 11:16:23 2012 >>> sys-devel/make-3.82-r4 Oct 09 21:01:24 I've cleaned all, even build/conf Oct 09 21:01:32 [ebuild R ] sys-devel/make-3.82-r4 USE="nls -static" 0 kB Oct 09 21:03:20 ant_home: yes, probably does defeat one of the tests :/ Oct 09 21:04:24 some distro defaults have changed? Oct 09 21:04:32 cheking... Oct 09 21:08:55 ndec: apparently sysroot_stage is not written to work with anything other than /usr/{,bin,include,lib,sbin}, so I think therein lies the source of my quandry Oct 09 21:09:29 wmcdevel: ouch. Oct 09 21:10:45 so, staging in sysroot, doesn't really use -dev at all in fact. Oct 09 21:11:16 ndec: not so much that ... it's more like, it never looks for those additional directories to begin with Oct 09 21:11:35 it's like there is some unwritten rule that all software is to be installed under /usr/ Oct 09 21:12:15 granted, that's normally where things go, but still, the ability to install to other locations should be honored Oct 09 21:12:38 googling for more info to see if someone else has stumbled upon this before Oct 09 21:17:37 RP: hm.. now I've removed the hack and it doesn't complain ... ? Oct 09 21:18:17 ah.. Oct 09 21:18:24 I did remove sanity.info Oct 09 21:18:27 It's there an EXTRA_IMAGE_FEATURE to configure bitbake to produce a file with all the patches that were applied to each recipe? Oct 09 21:18:49 RP: strange, my sanity.info is 3-liner Oct 09 21:18:51 ANITY_VERSION 1 Oct 09 21:18:51 TMPDIR /oe/oe-core/build/tmp-eglibc Oct 09 21:18:51 SSTATE_DIR /oe/oe-core/build/sstate-cache Oct 09 21:19:04 +S Oct 09 21:19:27 removing tht file triggers the make 3.82 check Oct 09 21:19:36 * ant_home confused Oct 09 21:20:12 mystery solved at least Oct 09 21:24:42 JaMa, pls try to reproduce Oct 09 21:32:34 ndec: got it ... added a sysroot_preprocess function to stage the additional directories. now to try and build the dependent package ... Oct 09 21:33:15 hmm. ok . that's good to know. Oct 09 21:33:52 that did it. good grief what a pain in the posterior. lol Oct 09 21:34:02 lol Oct 09 21:34:52 ant_home: I did many builds from scratch in last weeks, is there something special to reproduce it? Oct 09 21:35:04 the nice thing is, I added the function in a .bbclass that all the recipes use for the software suite, so -boom- all set. thought it was a pkg-config issue at first, but I guess not. Oct 09 21:35:22 JaMa: yes, just remove build/conf/sanity_info Oct 09 21:36:50 wmcdevel: in case you're thinking about next step, pls read http://lists.openembedded.org/pipermail/openembedded-core/2013-October/084896.html Oct 09 21:37:12 this for kernel headers, though Oct 09 21:37:55 the risk is having the dir wiped or overwritten Oct 09 21:41:28 ant: hopefully some time in the near future, the source tree will be set up to use the normal /usr/ pathing, so this whole go around will wind up being a moot point then Oct 09 21:42:16 everything that uses that scheme seems to work seamlessly, so we should have no problems once that change is done Oct 09 21:42:46 wmcdevel: well, I'm about bothering after release about a place to stage headers to be shared by different targets of the same arch Oct 09 21:44:00 ant_home: but why is 'staging' not simply reusing the -dev packages? Oct 09 21:44:28 without seeing the recipe I can only guess ;) Oct 09 21:44:35 the -dev packages are supposed to have everything need to build against the package, no? Oct 09 21:45:09 ant: the final image(s) that I'm building will not have any of the -dev packages installed, so should not be an issue there either, I imagine Oct 09 21:45:23 ant_home: my question was more generic. why isn't 'stating' just uncompressing -dev package into sysroot? Oct 09 21:47:52 ndec: that would make the most sense to me as well, and was my understanding of what it was doing in the first place Oct 09 21:47:57 most is decided in do_install. is it default? Oct 09 21:48:06 then the packaging Oct 09 21:49:05 in this specific case, i guess do_install() was not the default since wmcdevel has .h files in /opt/xxx Oct 09 21:49:44 so iirc, wmcdevel in his recipe had to do do_install(), and also set FILES_{PN}-dev to include /opt/xxx. Oct 09 21:50:02 so that generates 'good' packages. e.g. the -dev package has the .h from /opt. Oct 09 21:50:26 however since the -dev package is not used for 'staging' into sysroot, the .h files don't get into sysroot. Oct 09 21:50:40 you have to inspect packages-split and deploy-ipks i.e. Oct 09 21:50:42 there must be a reason why staging does not use -dev package. Oct 09 21:50:43 ndec: actually, no do_install ... made sure that FILES_{PN}-dev added the headers, .so, and pkg-config file Oct 09 21:51:08 ant_home: the packages are good. they contain all the .h files. Oct 09 21:51:57 ant_home: build/conf/sanity_info wasn't in my clean builds Oct 09 21:53:06 ndec: I based my recipes on others I have seen in the OE-core and meta-OE layers ... the only particular difference is the non-standard installation prefix Oct 09 21:53:10 ndec: check i.e. workdir of apmd recipe Oct 09 21:54:38 JaMa: so you don't have that file? Oct 09 21:55:53 no I mean that I've tested builds without this file multiple times in last weeks Oct 09 21:56:10 ah, ok, I think I've removed it as well in the past Oct 09 21:56:38 but since today the lack of it stoips the build Oct 09 21:57:41 usually I don't touch bblayers and local.conf. neither sanity_info Oct 09 22:05:44 heading out ... thanks for the help guys Oct 09 22:05:51 yw **** ENDING LOGGING AT Thu Oct 10 02:59:58 2013