**** BEGIN LOGGING AT Mon Oct 03 02:59:58 2016 Oct 03 07:38:23 good morning Oct 03 10:00:15 what special sauce do I need to give a recipe to set a systemd unit to be enabled by default? Oct 03 10:00:31 i.e first boot Oct 03 10:01:59 CTtpollard: inherit systemd, SYSTEMD_AUTO_ENABLE = "enable" Oct 03 10:02:50 erbo, thanks I'll give it a crack Oct 03 10:04:40 CTtpollard: you should be able to check using e.g. rpm -lp --scripts tmp/foo/bar.rpm instead of having to build the complete image Oct 03 10:04:54 er, rpm -qp --scripts Oct 03 10:22:40 erbo: still disabled at first boot, this is a user service if that makes any difference to what is needed Oct 03 10:22:57 if I manually enable, and reboot then it's fine Oct 03 10:25:41 I've seen the linking to multi users wants done manually in do_install for system units, but never for default user for user units/ Oct 03 10:26:26 CTtpollard: Hmm, I don't think there's any support in systemd.bbclass for system user sessions.. But symlinking manually would of course to the trick Oct 03 10:27:13 Yeh, it's just weird that I've got many recipes that don't do the linking for user services, but they work fine Oct 03 10:27:33 maybe something else is launching them Oct 03 10:28:06 probably Oct 03 10:57:53 Hi. I'm not sure if this is a problem with yocto but my image doesn't have /lib/ld-linux.so.3 Oct 03 10:58:14 But one of my binary is depending on it Oct 03 10:58:32 root@apalis-imx6:~# ldd rustup-init libdl.so.2 => /lib/libdl.so.2 (0x76a88000) librt.so.1 => /lib/librt.so.1 (0x76a71000) libpthread.so.0 => /lib/libpthread.so.0 (0x76a49000) libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x76a30000) libc.so.6 => /lib/libc.so.6 (0x768f5000) ld-linux.so.3 => not found libm.so.6 => /lib/libm.so.6 (0x7687a000) /lib/ld-linux-armhf.so.3 (0x76a9b000) Oct 03 10:58:45 ld-linux.so.3 => not found Oct 03 11:16:59 ls /lib/ld-linux* => /lib/ld-linux-armhf.so.3 Oct 03 11:18:01 so none here either Oct 03 12:46:36 Hello. I have just built a Yocto image that uses systemd instead of sysvinit. Does anyone have experience using systemd on embedded devices? What is the best way to make good use of it? Oct 03 12:47:10 Can I significantly speed up bootup with it or something similar? Oct 03 12:47:22 Are there other tangible benefits? Oct 03 12:52:25 ernstp: it means your image was compiled with softfloat but your other binary with hardfloat Oct 03 12:55:47 eduardas_m: sounds like you need to read lennarts primer on systemd Oct 03 13:16:02 is it not the "deploy" target of a linux-yocto image, whose responsability it is to get bzImage into deploy/images/ ? Oct 03 13:29:46 hm, I have "... is tainted from a forced run" staying around for ages - those should disappear after some time, right ? Oct 03 13:30:57 just asked for a "-c deploy -f" on my kernel, and it gets rebuilt from scratch, which may be linked to the ".do_unpack is tainted from a forced run" situation Oct 03 13:30:59 rburton, you mean this blog post? http://0pointer.de/blog/projects/systemd.html Oct 03 13:31:07 or some other document? Oct 03 13:33:13 btw, is there any timeframe decided for the next poky release ? Oct 03 13:33:38 eduardas_m: systemd for admins on https://www.freedesktop.org/wiki/Software/systemd/ Oct 03 13:34:02 yann: https://wiki.yoctoproject.org/wiki/Yocto_2.2_Schedule Oct 03 13:34:19 rburton: thx Oct 03 13:37:25 rburton, thank you Oct 03 13:41:30 so the best way to get rid of "... is tainted from a forced run" is to clean a recipe completely, right ? Oct 03 13:42:59 would there be a link between those taints on a kernel recipe, and the fact that "-c deploy" would not repopulate deploy/images ? Oct 03 14:23:38 after adding some packages to image dd of sdcard image does not create proper partitions on SD card Oct 03 14:23:47 What can be the problem? Oct 03 14:33:20 eduardas_m: you shouldn't touch sdcard image after its creation, there is a class doing such task properly Oct 03 14:34:07 mckoan, really, there is a bitbake task to write to sd card? Oct 03 14:34:54 eduardas_m: which machine are you using? Oct 03 14:35:45 mckoan, Variscite DART6UL Oct 03 14:35:48 MACHINE ??= 'imx6ul-var-dart' Oct 03 14:36:08 it is not in FSL community BSP, it is from a layer you add to it Oct 03 14:37:26 eduardas_m: meta-fsl-arm/classes/image_types_fsl.bbclass Oct 03 14:40:07 mckoan, what am I supposed to be looking for here? also you have not answered: is there a specific bitbake target to write image to sdcard? Oct 03 14:49:54 If a recipe sets IMAGE_INSTALL, and I include it, I apparently cannot do IMAGE_INSTALL_machine1 += " packages..."? I need to use IMAGE_INSTALL_append_machine1 = "" for this? Oct 03 14:51:46 yes Oct 03 16:19:46 hi i.MX folks, I have a question: Oct 03 16:20:29 is it possible to generate using yocto the update.sb and update_ivt.sb files? Oct 03 16:20:50 I have to flash via mfgtool 1.6.2.055 an i.MX28 board, and the flash fails really many times due to kernel crashes Oct 03 16:21:24 so since I did update the kernel to 4.4, I'm wondering about recreating the above files to make it work Oct 03 16:22:17 otavio, ^^ :) Oct 03 16:27:27 LocutusOfBorg: the blob can be generated for U-Boot and then you don't need the imx-bootlets Oct 03 16:28:04 how can I generate it? Oct 03 16:31:56 all what I have is one file and this line Booting update firmware. Oct 03 16:32:32 and an MfgTool that works on 25% of my tries Oct 03 16:37:16 I did bitbake imx-bootlets, but nothing is generated, Oct 03 16:37:26 and I'm pretty sure I have to add some initramfs kernels Oct 03 16:38:20 last time I used my kernel zImage, and I didn't got an updater, but rather a kernel run that wasn't receiving mfgtool commands Oct 03 16:38:33 I suspect I need some initramfs and some tools that receives stuff from usb cable Oct 03 16:45:06 LocutusOfBorg: the mfgtool support for imx28 is not complete Oct 03 16:45:27 LocutusOfBorg: this needs development; it should not be hard but needs some development for sure Oct 03 16:46:50 LocutusOfBorg: If possible, I would use mxsboot and skip mfgtool completely. Using mainline U-Boot it is easy to install the boxes using TFTP or similar Oct 03 16:48:08 well, that would mean change the production system Oct 03 16:48:12 but I can consider this Oct 03 17:08:08 on linux-yocto, .scc are supposed to be applied *after* defconfig, right ? Oct 03 17:14:15 I have USB_HID=m set in the vendor-shipped defconfig, and I'm overriding it to "y" in a scc/cfg snippet. This works as expected with a 4.1 kernel, but on a 4.4 kernel it ends up being "m". What could cause that ? Oct 03 17:23:09 d'oh, parameter subtly changed name :( Oct 03 17:31:07 Did toolchain installs used to be owned by the installing user? (in, say, 1.6.3 or so?) Oct 03 17:31:40 * fishey1 is dealing with moving some folks from an old poky fork to the current release & running into toolchain issues due to how it is used Oct 03 17:32:54 similarly: is there a correct way to handle a setup where people want to build a few different packages using the toolchain, and have some of these install headers & libraries that the other ones depend on? Perhaps by using the extracted rootfs as the sysroot? Oct 03 23:54:30 fishey1: there should be not much operational changes for Oct 03 23:54:32 sdk Oct 03 23:54:46 fishey1: can you describe your issues in detail Oct 03 23:55:18 usually, it does not include the applications Oct 03 23:55:50 development libs and headers but it does include the same for the dependencies Oct 04 00:08:14 why is e.g. read_subpackage_metadata() a bitbake-style python function? what advantage does that give over a regular python function? just that overrides can be applied to it (though that isn't done in poky)? Oct 04 00:08:37 insane.bbclass and a bunch of package*.bbclass call it with bb.build.exec_func("read_subpackage_metadata", d) Oct 04 00:09:02 there is bitbake context and then there is oe context Oct 04 00:09:22 insane class lives in OE config namespace Oct 04 00:10:10 hmm... could you elaborate? they all call it the same way, and all callers as well as the definition is in meta/classes/. Oct 04 00:16:51 yes but its sort of oe library Oct 04 00:16:59 where this funciton is Oct 04 00:17:17 and you want to insert it into bitbake context Oct 04 00:17:27 or say into runqueue Oct 04 00:18:28 does bb.build.exec_func() run the function async? i had completely missed that in that case. Oct 04 00:18:52 its sort of way to call python function from shell Oct 04 00:19:53 so it's to make it general and callable from both python and shell functions? Oct 04 00:21:49 and nah, not async, afaics... Oct 04 00:24:02 right Oct 04 00:24:09 async is not the usecase here Oct 04 00:26:24 i might have some vague idea at least. having it be some standalone unit for execution rather than just a helper function. Oct 04 00:26:47 nothing seems to use it like that, but i think i kinda-sorta see the thinking, maybe... Oct 04 00:28:21 its used by all different package classes Oct 04 00:28:30 so its more of a library call here Oct 04 00:29:37 would making it a regular python function have any operational disadvantages? (i'm not going to change it. i'm just curious.) Oct 04 00:30:15 Ulfalizer: it was written before the "def" syntax even existed Oct 04 00:30:19 :) Oct 04 00:30:30 ok, that's an explanation i can wrap my head around at least :P Oct 04 00:31:34 for the future, i'd suggest checking git history for things like that. git can actually limit the log to a function if you give it a range of line numbers or regex patterns for start/stop Oct 04 00:31:48 its called in do_package_write () task Oct 04 00:32:11 oh... didn't know you could do that. thanks for the tip. Oct 04 00:32:25 though in this case you can probably just use git log -S Oct 04 00:32:35 which finds commits which add/remove lines mentioning the following Oct 04 00:32:40 so it'd find the commits that added the calls to it Oct 04 00:32:43 oh so its about syntactic change hah Oct 04 00:32:53 khem: yeah, i've grepped for all call sites. was just curious as to why it's a bitbake-style python function rather than a regular def-style helper function. Oct 04 00:32:57 bb.build.exec_func() vs just calling the thing Oct 04 00:33:15 and i'm happy with "historical" Oct 04 00:33:18 * Ulfalizer checks the log Oct 04 00:33:20 i dont think it matters Oct 04 00:33:41 in this case i think you'd have to check oe classic to dig far enough back :) Oct 04 00:33:47 the pkgdata bits have been there a long time Oct 04 00:34:33 all bbclasses seem to use python func () Oct 04 00:34:57 yeah good old days when python sucked Oct 04 00:35:07 less :) Oct 04 00:36:13 are there any other advantages to bitbake-style python with exec_func() vs. def-style functions besides being able to apply overrides to them by the way (completely unrelated to the earlier discussion)? Oct 04 00:37:42 you can only override shell funcs Oct 04 00:38:37 exec_func handles dirs, cleandirs, lockfiles, and writes a log file. negatives, it can't return a value Oct 04 00:38:44 khem: nope, works for bitbake-style python functions too. see https://www.yoctoproject.org/docs/2.2/bitbake-user-manual/bitbake-user-manual.html#bitbake-style-python-functions. Oct 04 00:38:51 and i actually tried out that example Oct 04 00:38:53 internally, today, exec_func_python actually injects the function being called as a def style function under the hood anyway, just wraps it Oct 04 00:39:09 ah, yeah, had forgotten about variable flags Oct 04 00:40:04 so depends on the circumstances. more often than not def is probably fine, and utilities of that sort are often better placed in an oe python module rather than a class Oct 04 00:41:23 course back in the day we had no oe python package at all either Oct 04 02:29:24 kergoth: in case you feel like doing some proofreading: https://bugzilla.yoctoproject.org/show_bug.cgi?id=10364 Oct 04 02:29:25 Bug 10364: normal, Undecided, ---, srifenbark, NEW , Suggested documentation to clarify BitBake-style vs. regular Python functions and bb.build.exec_func() **** ENDING LOGGING AT Tue Oct 04 02:59:58 2016