**** BEGIN LOGGING AT Fri Mar 10 03:00:04 2017 Mar 10 07:02:11 if we need to add some install to image, is it by doing bitbake or do we need first to add the package into image recipe ? Mar 10 07:04:59 ranchu: the latter. bitbake $PACKAGE will do exactly what it says: buld the package and then do... nothing with it. Mar 10 07:11:00 LetoThe2nd - Thanks! Mar 10 07:15:48 Hey, any way i could get yocto with the 3.2.0 kernel? Mar 10 07:16:37 maka_: http://git.yoctoproject.org/cgit.cgi/poky/plain/meta-skeleton/recipes-kernel/linux/linux-yocto-custom.bb Mar 10 07:17:20 I tried something like that but then it gives me something like, 4.1 is the lowest version Mar 10 07:18:46 maka_: "i tried something and it gave me something" is about the second most useless error description ever. the most useless one being "it doesn't work" Mar 10 07:19:32 Yea, sorry for that, i'm trying to reproduce the error, wait a sec Mar 10 07:20:32 "NOTE: preferred version 3.2% of linux-yocto not available" Mar 10 07:22:37 maka_: probably becuase there is no maintained 3.2 by the yocto project anymore. so you'll have to rip out the inheritance from linux-yocto and replace by completely custom Mar 10 07:23:03 Hmm i'll look into that then, thanks ^^ Mar 10 07:23:45 1 other thing, i have a kernel module that works on that 3.2 kernel. Would it be very difficult to get that to work on the 4.10 kernel? Mar 10 07:23:47 maka_: its an old arm recipe, but the idea is like that: https://github.com/linux4sam/meta-atmel/blob/42587a26a3bb71d49f2dd580854b135a76b75716/recipes-kernel/linux/linux-yocto-custom_3.6.9.bb Mar 10 07:24:04 maka_: porting stuff from 3.2 to 4.10 is probably painful. Mar 10 07:24:40 Is it even possbile when i only have 1 .ko file? Mar 10 07:25:09 no. and i would not place any bets on a binary ko file working properly with a custom kernel Mar 10 07:25:56 for all sanity the kernel should refuse to load it anyways. we have module versioning and checksums in place to prevent that for good reasons. Mar 10 07:26:05 So it probably wouldn't work on a yocto image if i made it with the 3.2 kernel? Mar 10 07:26:25 yes. Mar 10 07:26:46 LetoThe2nd - if I just do bitbake does it create a binary .ipk file ? Mar 10 07:27:05 ranchu: if your package management is set to ipk: yes. otherwise: no. Mar 10 07:27:31 LetoThe2nd: Well, thanks for the info once again ^^ Mar 10 07:27:49 Is there a way to manually install the ipk into filesystem (without adding image_install in image recipe) ? Mar 10 07:28:05 I mean install in the development host Mar 10 07:28:18 ranchu: if you package management in target enabled, yes. otherwise.... not really. Mar 10 07:28:24 ranchu: in the dev host?!? Mar 10 07:28:37 maka_: hrhr Mar 10 07:28:42 yes - I just wander how to do it manually Mar 10 07:28:53 LetoThe2nd: hrhr? Mar 10 07:29:08 maka_: nothing. just ignore it. Mar 10 07:29:11 I know it is not the correct way, I am just trying to understand the steps from ipk to filesystem Mar 10 07:30:00 ranchu: basically: fakerrot into the image and use the package managers native incarnation to place stuff there. Mar 10 07:30:06 when we do bitnake it actually install the ipk into filesystem right ? Mar 10 07:30:43 so I try to understand the exacst command which trsansfer from ipk to filesystem in the development environment Mar 10 07:31:17 ah , fakeroot / Mar 10 07:31:55 chroot or fakeroot ? Mar 10 07:35:10 ranchu: look at the different stages a package goes through, basically from bitbake -c listtasks. then dig deeper. Mar 10 07:35:58 ranchu: i *think* it is fakeroot, but don't place any bets on it please. it rather sure it is not chroot. many package managers also do not need that because they internall can select a root point. Mar 10 08:34:38 Mar 10 08:38:28 kanavin: so close :) Mar 10 08:39:09 RP: yes :) Mar 10 08:39:19 RP: but why is it still crunching? Mar 10 08:39:30 the new autobuilder seemed snappier Mar 10 08:39:41 kanavin: the new AB is snappier Mar 10 08:40:17 kanavin: often depends whether it was a full build or whether valid sstate was available Mar 10 08:41:24 RP: is it okay to drop DISTRO_VERSION from /etc/issue? it's again causing trouble with multilib - if base-files and lib64-base-files have different dates in /etc/issue, they can't be both installed Mar 10 08:42:01 I can also add the version, but strip the date suffix from it Mar 10 08:42:58 kanavin: hmm, good question Mar 10 08:43:07 kanavin: I don't think we can entirely drop it Mar 10 08:44:15 "we default to cc not gcc" - why :/ Mar 10 08:44:35 They also mean "we only use cc" Mar 10 08:44:56 RP: I think it should be okay to replace '2.2-20170310" with just "2.2-snapshot" - if base-files is coming from sstate, the date would be inaccurate anyway Mar 10 08:45:16 RP: and it's only for login banner, really Mar 10 08:45:26 kanavin: remember that other distros can also set other things here :/ Mar 10 08:47:32 RP: gm. Sorry to bother you in the night but I only build in the freetime Mar 10 08:47:52 ant_work: its ok, I just am a little pushed for time at the moment :( Mar 10 08:47:58 as I said there are still issues with accessory utils not found in native sysroot Mar 10 08:48:51 I found two atm: one is tunctl/qemu-helper-native, the other is ncurses-native Mar 10 08:49:11 this is with nodistro fwiw Mar 10 08:50:45 I see in the list you were answering about tunctl right during my tests :) Let see how the discussion ends (Re: [OE-core] About scripts/oe-find-native-sysroot) Mar 10 08:58:23 RP: how about generating /etc/issue at base-files package install time? i.e. postinst Mar 10 09:02:42 kanavin: that could work, we'd have to document why Mar 10 09:03:05 ant_work: I'd assumed you were asking in relation to that Mar 10 09:03:22 RP: do you remember if do_prepare_recipe_sysroot is supposed to clean target directories before populating them? Mar 10 09:03:42 RP: it's certainly easier to check DISTRO_VERSION for "snapshot-" and drop the digits if so Mar 10 09:03:43 JaMa: it can't as that would remove e.g. quilt-native added from do_patch Mar 10 09:04:39 JaMa: it will remove anything which it noticed has had its checksum change Mar 10 09:04:50 RP: with PATH whitelist I need to add xxd, so first I've added xxdi-native and then vim-native to provide original xxd and was surprised that 2nd build (with vim-native) failed in do_prepare_recipe_sysroot, because usr/bin/xxd already existed there (from previous xxdi-native dependency) Mar 10 09:05:56 JaMa: its not 100% perfect, this is what I meant in the original commit and discussion when I said it was "recipe specific", not "task specific" Mar 10 09:06:08 so the problem is IMHO, that it should remove also the dependencies which are no longer in DEPENDS even if they haven't changed the checksum Mar 10 09:06:33 JaMa: its not that simple as you can't know if there is some other task from the recipe running in parallel for which this is a [depends] Mar 10 09:07:07 yeah, I didn't expect it to be simple, but fair enough Mar 10 09:07:38 JaMa: I added the checksum piece as a best effort to resolve it, I can't figure out any other safe way to do it Mar 10 09:08:00 I was just assuming that do_prepare_recipe_sysroot is the first (and only) function which populates the sysroot Mar 10 09:08:49 JaMa: sadly not :( Mar 10 09:09:09 JaMa: e.g. you need quilt-native for do_patch Mar 10 09:09:47 if this is the only exception, cannot we have shared sysroot just for quilt-native again? :) Mar 10 09:10:24 or any other PATCHTOOL configured Mar 10 09:10:40 JaMa: its only the start of the problems Mar 10 09:10:44 ok Mar 10 09:11:08 JaMa: see the number of tasks which set [depends] Mar 10 09:11:15 FWIW: relatively big build with latest version of PATH patch showed only 2 issues (this xxd and surprisingly another case of missing gettext inherit) Mar 10 09:11:53 world build still running it will take 12-20 more hours NOTE: Running task 12028 of 36415 Mar 10 09:11:57 JaMa: I got a greenish build on the autobuilder with my last version. cdrtools-native is bust, as is ovmf and sefltest is also broken (10 failures) Mar 10 09:12:32 and that's not counting the silently ignored issues in the build Mar 10 09:14:55 JaMa: no :/ Mar 10 09:16:41 JaMa: can only go so far though. I'm hoping to at least compare some buildhistory data Mar 10 09:33:23 JaMa: I have seen another issue since RSS was introduced. My TMPDIR is in tmpfs. The default nr_inodes was 2m and this is not enough anymore if you don't inherit rm_work. Mar 10 09:34:42 so I went out of inodes while still having half of ram free Mar 10 09:36:41 ant_work: yes I've seen it too, except I was using 6M and it still wasn't enough Mar 10 09:38:07 ant_work: but in my case it looked almost as some inodes leaking somewhere, because to finish at least some recipes, I was deleting old WORKDIRS based on number of inodes used (instead of disk space which is usually the issue) and in the end the sum of inodes reported by du -i was much lower than used inodes in df -i Mar 10 09:40:13 hm...in my case it stopped apaprently at 96% of usage (du -i) Mar 10 09:43:02 hard-links need inodes :/ Mar 10 09:47:50 ant_work: where does the additional inode come from? hard-linked files share the same inode - that's the whole point, isn't it? Mar 10 09:48:16 Hmm, perhaps not. Mar 10 09:48:32 Now I need to update my mental model :-/ Mar 10 09:51:12 No, I just made a mistake in my test. "ls -i" shows that the inode is indeed the same after "ln foo bar". Mar 10 09:51:59 Perhaps it is the increased number of directories which causes tmpfs to run out of inodes? Mar 10 09:52:20 So it's not exactly the hard-linking, but rather the directory trees containing the hard-links. Mar 10 10:05:15 pohly, yes, directory entries Mar 10 10:05:19 pohly: I've tried your rm_work change (the one which returns rm_work_all task) but it doesn't seem to be enough (still cannot fit 2 chromium builds in one tmpfs) Mar 10 10:05:39 pohly: in next world build I'm trying with whole rm_work improvement reverted to see if it will fit or not Mar 10 10:07:04 pohly: its the fact that even if the file has the same content, it needs its own directory entry Mar 10 10:07:34 pohly: so increased number of directories and directory entries Mar 10 10:08:07 JaMa: thanks for testing. I probably should submit the tentative patch for OE-core, because it fixes a regression, even if the end result still doesn't work as well as it should. Mar 10 10:14:20 rburton1: watch out for some of these patches adding True back to getVar Mar 10 10:15:13 heh Mar 10 10:20:24 rburton1: I still can't reproduce that test failure. However, I'm seeing some weird breakages like this on master-next: https://gist.github.com/bartosh/ce78c953c67c5905a51b2ac98b80f259 Mar 10 10:21:44 ed2: These are from my PATH whitelist changes. I'd love help in fixing them Mar 10 10:21:59 ed2: I have cdrtools and ovmf fixes locally Mar 10 10:25:35 RP: can you point me out to the "guilty" commits that I'd understand what's the reason of this? Mar 10 10:25:44 RP: can you push your changes somewhere? Mar 10 10:26:25 ed2: I just updated -next with my fixes Mar 10 10:26:36 RP: thanks Mar 10 10:27:04 ed2: http://git.yoctoproject.org/cgit.cgi/poky/commit/?h=master-next&id=26e750e1a0b404a229447d071b328b8e68f83dc7 is the change which likely breaks things Mar 10 10:27:28 ed2: how that loses bitbake from PATH, I don't know Mar 10 10:28:30 ed2: https://autobuilder.yocto.io/builders/nightly-oe-selftest/builds/211/steps/Running%20oe-selftest/logs/stdio is full set of failures from selftest (before I merged the last two changes) Mar 10 10:34:12 RP: WARNING: quilt-0.65-r0 do_package_qa: QA Issue: /usr/lib/quilt/ptest/quilt/scripts/dependency-graph contained in package quilt-ptest requires /data/poky-master/tmp/hosttools/perl, but no providers found in RDEPENDS_quilt-ptest? [file-rdeps] Mar 10 10:34:30 it found a perl :) Mar 10 10:34:31 RP: isn't PATH = "${TMPDIR}/hosttools" the reason of losing bitbake? Mar 10 10:40:54 RP: commenting PATH = "${TMPDIR}/hosttools" out in meta/conf/bitbake.conf made oe-selftest -r wic.Wic.test_exclude_path passed Mar 10 10:51:51 ed2: hmm, probably yes Mar 10 10:52:17 ed2: I'd guess adding bitbake to hosttools would fix it too? Mar 10 10:52:33 rburton1: yes, need to fix that too :/ Mar 10 10:53:04 RP: just a matter of PERL=perl in the quilt recipe, or rewrite things when generating the depends Mar 10 10:53:20 rburton1: I like PERL=perl Mar 10 10:54:05 rburton1: do you mean this? http://git.yoctoproject.org/cgit/cgit.cgi/poky/commit/?h=master-next&id=26e750e1a0b404a229447d071b328b8e68f83dc7 Mar 10 10:55:20 kanavin: yes Mar 10 10:56:07 RP: will test quilt fix once the current run has finished Mar 10 10:56:11 rburton1: then probably not - what happens is that gpgme's python module build system called gpgme-config (from the source tree) to find out what the linking flags should be, and it gives -L${libdir} Mar 10 10:56:28 which is of course -L/usr/lib --> host contamination Mar 10 10:57:01 it was late, I was annoyed at Mark, so sorry for a bit half-baked patch Mar 10 10:57:34 fix/resend? Mar 10 11:11:45 hi all, I have a file under mylater/recipes-core/image/myimage which a recipe to generate an image - is it possible to add a post() hook which - once the image is generated perform some actions ? Mar 10 11:14:20 rburton1: have you seen this failure before? https://autobuilder.yoctoproject.org/main/builders/nightly-ppc/builds/1100/steps/Running%20Sanity%20Tests/logs/stdio Mar 10 11:14:36 rburton1: it's connman image test failing, because it's not running when the test executes Mar 10 11:15:12 rburton1: I'm tempted to declare it a racing issue unrelated to dnf (couldn't reproduce locally), but I'm not 100% sure - maybe 99% :) Mar 10 11:18:25 pedr0_: what actions do you need to perform? Mar 10 11:26:48 creating an tar file - I've got an answer I guess Mar 10 11:27:07 but the reality is that I am still too inexperienced with OE to start with sth like that Mar 10 11:28:00 for example, I've inherited some code from an external provider and they start the build in this manner : MACHINE=xxx bitbake recipe-in-my-own-layer Mar 10 11:28:13 is it possible to embed the MACHINE type in the image recipe itself Mar 10 11:28:16 ? Mar 10 11:29:00 pedr0_: why not just set it in local.conf? Mar 10 11:30:17 well .. I would love to checkout the repo and start a build without having to fiddle with anything - moreover I have different image recipes which uses different kind of machines Mar 10 11:30:17 hello Mar 10 11:30:34 did anybody manage to install php version 7 on yocto? Mar 10 11:30:38 did anybody manage to install php version 7 on yocto? Mar 10 11:30:52 ? Mar 10 11:30:53 in my mind - you would just run bitbake - no need to do anything else more than that Mar 10 11:30:59 I have a dream ... :-) Mar 10 11:31:14 pedr0_: no, you just have not read the yocto quick start document :) Mar 10 11:31:28 I swear I did Mar 10 11:31:36 what did I miss ? Mar 10 11:32:14 pedr0_: https://www.yoctoproject.org/docs/current/yocto-project-qs/yocto-project-qs.html#building-an-image-for-hardware - 2. Configure the build Mar 10 11:32:32 pedr0_: in short: look at conf/local.conf Mar 10 11:33:10 isn't there any way to get around that ? Mar 10 11:33:48 it may be better to write an builder script which writes on the local/conf then Mar 10 11:40:06 hey Mar 10 11:40:09 what about php 7 Mar 10 11:40:11 on yocto Mar 10 11:40:21 anybody managed to get it to work? i cannot find any documentation Mar 10 11:42:15 VictorBatman: we have read you. 3 times now. Mar 10 11:43:17 pedr0_: whats the point getting around what? Mar 10 11:47:19 RP: are the selftest failures on the most recent build caused by the PATH filtering? Do they need triaging? Mar 10 11:47:43 ovmf build fails for qemux86 atm? Mar 10 11:50:01 To minimize the number of actions to perform - not a big deal though Mar 10 11:58:23 pedr0_: if you need to tweak local.conf more than once, you're doing something wrong Mar 10 12:00:11 joshuagl: they're caused by path filtering afaict. Although I can't reproduce them all. Probably no need to triage Mar 10 12:01:13 again - I'd love to tweak it zero times Mar 10 12:01:19 RP: ack, thanks Mar 10 12:01:31 I'd love to tweak it zero times Mar 10 12:06:47 pohly: i guess you're the right person to ask about ovmf, i'm trying to run wic selftest and it's failing with some efi tests, can ovmf be built for qemux86 at all? Mar 10 12:11:56 rburton1: EXTRA_OECONF_append_class-target = "--with-perl=perl" seemed to help (I used append to avoid a world rebuild although I forgot -native was seperate) Mar 10 12:21:59 anyone tried zapcc already? Mar 10 12:25:34 mborzecki: eduardas_m added those wic selftests recently, I myself haven't run them. Mar 10 12:26:06 mborzecki: what are the failures? Mar 10 12:26:48 Nope, not eduardas_m - I was looking for Ed Bartosh, who doesn't seem to be online at the moment. Mar 10 12:26:52 ovmf build fails with undefined reference to `ReadUnaligned64' (also WriteUnaligned64 and such) Mar 10 12:27:45 mborzecki: I just built it here FWIW and it does build on the autobuilders Mar 10 12:28:22 qemux86? Mar 10 12:28:32 mborzecki: qemux86-64 Mar 10 12:29:21 that's what AB builds too, right? Mar 10 12:29:28 mborzecki: could be I guess Mar 10 12:29:46 mborzecki: I thought I was using qemux86 locally but clearly not sorry Mar 10 12:30:27 hm, i'll try rebuilding with qemux86-64, maybe it will go away :) Mar 10 12:37:35 rburton1: patch for quilt in -next Mar 10 13:33:21 tried qemux86-64, ovmf do_compile still fails https://paste.fedoraproject.org/paste/BznPJOGnDre4UTcknoYUXF5M1UNdIGYhyRLivL9gydE=/raw Mar 10 13:33:44 do I need to have efi in MACHINE_FEATURES? Mar 10 13:45:13 hi guys, I'm trying to add a new file to the /etc/profile.d for my image using a bbappend to qtbase recipe but the do_rootfs fails as there are some missing packages (qt3-examples in my case which is introduced by packagegroup-qt5-demos) Mar 10 13:46:05 I'm not doing anything else but a do_install_append where I install my file to /etc/profile.d and then use FILES_${PN} += to package it Mar 10 13:46:37 any idea on why is this happening is much appreciated Mar 10 13:47:28 packages-split/qt3d-examples is empty Mar 10 14:01:01 I'm trying to compile the fsl-image-multimedia-full using morty version in an ubuntu 14.04 Mar 10 14:01:19 and in the python compilation I get this error arm-fslc-linux-gnueabi-gcc: error: unrecognized command line option '-fp-model' Mar 10 14:24:02 caiortp: Hi, can you show all log file? Mar 10 14:31:47 berton, I already cleanall and restart the build , after finish I will see if there's some error and put in the pastebin, tks Mar 10 14:32:42 caiortp: ah ok. If you this error again let me know Mar 10 14:35:47 my kernel (linux-xlnx 4.9) seems to be missing dma-mapping. I don't see the .h file in /usr/include/linux Mar 10 14:38:04 berton, tks Mar 10 16:45:25 How do you prevent the "auto provides" based on the .so that a recipe installs ? Mar 10 16:45:51 I have a recipe that installs libc6.so somewhere on the rootfs, but that lib isn't used by anything else than the recipe Mar 10 16:46:05 Unfortunately, yocto detects the .so and declares every recipe to be dependant on it Mar 10 16:47:14 (even though I don't install the .so in a regular lib directory like /usr/lib or /lib) Mar 10 18:09:43 halstead: abs are idle, master-next it ready for the test round of testing when you're done, thanks! Mar 10 18:21:31 Hi guys, quick question can I build an RTEMS image using yocto? Mar 10 18:28:46 rtems: good question, I think with some effort you can Mar 10 18:30:23 rtems: zephyr uses yocto project to build their SDKs Mar 10 18:31:44 and I know gcc toolchains support RTEMS targets so wouldn't be that hard Mar 10 18:36:03 Cool! I'm completely new to yocto (and building o Mar 10 18:36:28 os'es generally) so I was trying to find some instructions online but couldn't come across any so i thought maybe it isn't doable Mar 10 18:36:35 but that's good to hear Mar 10 18:38:27 rtems: sure, see https://github.com/zephyrproject-rtos/meta-zephyr-sdk Mar 10 18:38:59 if you need help feel free to post your questions here Mar 10 18:41:45 khem: thank you! just to be clear what you have linked is the setup for building a zephyr os right? And I can use it as a baseline for building an rtems image? Mar 10 18:42:13 rtems: actually, no, Mar 10 18:42:31 rtems: you will generate an SDK for the rtos Mar 10 18:42:43 and then use the SDK for rtos Mar 10 18:43:02 building a rtos using OE framework itself is another thing Mar 10 18:43:20 once we could build uclinux but that was long time ago Mar 10 19:01:21 khem: Ahh I see, ok so im going to have to do some research/reading into zephyr and try to implement it, I'm pretty lost here Mar 10 19:03:29 yes Mar 10 19:03:49 I am sure if you follow what zephyr did thats will be a pretty good template Mar 10 19:10:51 Cool, thanks for pointing me in the direction! Mar 10 19:13:10 One last question, does zephyr support ppc architecture? I don't see it on the supported boards page Mar 10 19:25:02 khem: you know if anyone has used a non-linux TARGET_OS recently? i.e. bare metal toolchains, or rtos as you were saying Mar 10 19:25:14 i know we used to do it from time to time, i.e. we had an avr toolchain build working at one point Mar 10 19:25:21 wondering if it's still tested Mar 10 21:22:23 kergoth: yes zephyr uses baremetal/newlib targetted toolchains build out of OE Mar 10 21:25:10 * nerdboy spreads the "luv" to u-boot **** ENDING LOGGING AT Sat Mar 11 03:00:01 2017