**** BEGIN LOGGING AT Fri Aug 19 02:59:58 2016 Aug 19 03:33:49 Hello Aug 19 03:34:24 I'm extending a kernel recipe with a bbappend file that has a file://remove-kernel-option.cfg kernel defconfig fragment Aug 19 03:34:42 Problem is, bitbake applies my fragment before the recipe's defconfig file I'm appending! Aug 19 03:34:56 This leads to my kernel configuration fragment being moot Aug 19 03:34:59 Any advice? Aug 19 05:01:31 darwish: are you using linux-yocto Aug 19 05:02:46 khem, yup .. the recipe I'm appending over it inherits that class Aug 19 05:03:27 khem, weirdly .. when I used SRC_URI_append, instead of SRC_URI += "..", my configuration fragment get appended after recipe src_uri defconfig as expected Aug 19 05:03:36 dunno why though ... Aug 19 05:09:12 append is post operation so thats right thing to do Aug 19 08:35:04 hi guys, is there a way to have persistent /var/log directory? I have overriden volatiles file from inistscripts recipe to remove the link /var/log -> /var/volatile/log Aug 19 08:35:36 i also did a volatiles_remove = "log" from base-files recipe but it didn't help Aug 19 08:41:54 s65b40: maybe this discussion helps you: https://patchwork.openembedded.org/patch/89883/ Aug 19 08:42:10 s65b40: plus, you need writeable memory then of course. Aug 19 08:49:41 LetoThe2nd: thanks i'll have a look Aug 19 09:24:31 hi, i want modify sudoers file so that wheel group can have root privileges Aug 19 09:24:51 is the best way to just replace whole file or is there a better way? Aug 19 09:24:51 anybody understands how sanity.bbclass works ? Aug 19 09:25:15 it has hardwired references to gcc and apparently is run before local.conf is parsed Aug 19 09:34:30 boucman_work: sanity is completely overrated. Aug 19 09:34:55 LetoThe2nd: hehe Aug 19 09:35:37 * LetoThe2nd is absolutly sure. Aug 19 09:46:31 ok, I need a bitbake expert now :( Aug 19 09:46:46 this thing is getting more and more complicated every time... Aug 19 09:57:34 boucman_work: if things also get more compilated? Aug 19 09:59:10 LetoThe2nd: ?? Aug 19 10:00:53 boucman_work: nevermind, it was a rather bad play on words. Aug 19 10:01:46 ok Aug 19 10:01:52 (still didn't get it :P ) Aug 19 11:44:03 RP: are you the one to whom I can ask complicated bitbake questions ? Aug 19 12:16:16 I appended a recipe which gets its sources from git. Within my recipe i want to add additional files which are also fetched from a git rep Aug 19 12:16:44 boucman_work: I'm someone who might be able to answer... Aug 19 12:16:54 RP: hehe Aug 19 12:17:24 The problem is, that it always replaces the dir where i want to ADD the files. I already tried destsuffix and subdir, without success Aug 19 12:17:56 RP: I'm still trying to build yocto with no binary called gcc in my PATH (i.e using the BUILD_* variables) and i'm stumbling a bit on sanity.bbclass (it might be a bitbake problem triggered by it Aug 19 12:18:46 The basic problem is that the sanity check for gcc seem to happen before local.conf is loaded, so before yocto has any chance to provide the local compiler's name. Aug 19 12:18:57 Here is wht I have figured out so far Aug 19 12:19:27 "someone" calls command.py:runAsyncCommand Aug 19 12:20:02 this triggers cooker.updateCache Aug 19 12:20:29 which triggers bb.event.SanityCheck Aug 19 12:20:44 and sanity.bbclass has registered a handler for that event. Aug 19 12:21:04 in sanity. bbcheck we have multiple lines like the following Aug 19 12:21:18 status,res = oe.utils.getstatusoutput("${BUILD_PREFIX}gcc gcc_test.c -o gcc_test") Aug 19 12:21:36 which "works" even when BUILD_PREFIX is not set Aug 19 12:22:17 however when I have no gcc, it fails, and when I replace ${BUILD_PREFIX}gcc with ${BUILD_CC}, it fails too Aug 19 12:22:23 Aug 19 12:23:11 so, is that expected ? how can I know who triggers the async call ? any lead on how to fix that ? Aug 19 12:23:56 boucman_work: I'd very much doubt that BUILD_PREFIX isn't set at that point Aug 19 12:24:56 boucman_work: what has actually happened is the code is broken Aug 19 12:25:04 boucman_work: it should have a d.expand() around those calls Aug 19 12:25:25 boucman_work: once upon a time, bitbake expanded python functions. It doesn't anymore and it looks like nobody noticed this bug Aug 19 12:25:46 aaah Aug 19 12:25:58 d.expand around the executable line Aug 19 12:26:03 thx, i'll try that and come back to you Aug 19 12:26:14 boucman_work: sanity_data.expand() in this case Aug 19 12:26:31 yes, thx Aug 19 12:26:39 boucman_work: I'll queue a patch for that, it looks like a genuine bug where things happened to work as its often "" Aug 19 12:28:54 nobody knows? Aug 19 12:32:03 rp: as you wish, I have the infra ready to test if you want Aug 19 12:33:12 boucman_work: happy for tests! :) Aug 19 12:34:06 ok, i'm in the bug situation, ready to test whenever you want Aug 19 12:39:21 boucman_work: http://git.yoctoproject.org/cgit.cgi/poky-contrib/commit/?h=rpurdie/wip&id=385003209f18aab5e3877b91d3562007540541fe Aug 19 12:54:34 RP it works (though you didn't replace the ref to gcc, i'll do that with my patch serie) Aug 19 12:59:15 boucman_work: ok, fair enough Aug 19 15:05:43 I have various failures in glibc on master right now... is that a known bug ? Aug 19 16:43:47 Hmm, slightly irritating that bitbake-layers has to parse bitbake.conf for some of the commands that really have no need of going that far.. but tinfoil and cookerdata don't separate bblayers/layer.conf parsing from bitbake.conf parsing Aug 19 16:43:49 oh well Aug 19 16:57:17 specifically, if local.conf inherits a class in a layer that's not currently configured, we can't run bitbake-layers add-layer to bring it back :) have to add it manually Aug 19 16:57:33 is there a recent summary of what the best type of build machine is to use for OE? Aug 19 16:57:49 more cores vs. faster cores, more memory, SSDs, etc Aug 19 16:58:03 there is no 'best', it depends on your needs and finances Aug 19 16:58:33 kergoth: sure, but i guess more of the latter two is fine Aug 19 16:58:52 But just wondering if I should get 4 faster cores vs. 12 slower cores Aug 19 17:00:30 I expect the folks who manage the yocto project autobuilder probably have the best feel for such things, for where the bottlenecks are. I'd expect the latter to be better in that case, as we run a hell of a lot of tasks, and most are more i/o bound than cpu bound, but that's just off the top of my head, lacking first-hand experience with the really hefty machines Aug 19 17:00:58 maybe one of them will jump in Aug 19 17:01:31 mattsm: we tested this a while back and found that 8 faster CPUs gave us better performance than 16 slower ones, but that may have changed over time or have been related to the amount of memory per CPU we had. Aug 19 17:02:18 neverpanic: cool thanks Aug 19 17:02:21 Also, the difference of 4 vs. 12 is so large that 12 CPUs are probably going to be faster Aug 19 17:03:29 * paulg uses raid0 across two disks; I'll trade 2x the i/o for 1/2 the reliability when using a builder for yocto Aug 19 17:03:32 i expect we probably have trouble completely using 16 cores due to bottlenecks int he dependency graph. guessing we don't use the full capacity Aug 19 17:04:56 on the other hand the default tunings mean that on a 16-way you potentially have the equivalent of make -j256 running at some times Aug 19 17:05:31 I'm sure there's diminishing returns on make parallelism in most upstream buildsystems Aug 19 17:05:33 * kergoth shrugs Aug 19 17:07:42 ya, no matter what, do_rootfs will still suck. :-/ Aug 19 17:07:55 so 8 faster ones, seems like a good compromise Aug 19 17:14:04 * paulg pulls in gcc updates ; guessing that will trigger a rebuild of everything.... Aug 19 17:15:17 ntl, I've not had any issues running 16x16 in 64GB, building a fairly complex image (full X11 etc.) Aug 19 17:17:07 there is a certain threshold where bitbake will look in /proc/cpuinfo and nag you if you get too greedy IIRC. Aug 19 17:21:47 as kergoth says, there are bottlenecks for sure... Aug 19 17:21:50 Currently 1 running tasks (4314 of 9615) 44% |############################################# | Aug 19 17:21:50 0: glibc-2.24-r0 do_compile (pid 12754) Aug 19 17:29:23 Don't forget I/O bottlenecks Aug 19 17:29:40 got to build in a massive 64GB ramdisk Aug 19 17:38:51 * paulg likes the dependency info that zeddii 's new kernel tools output. Aug 19 17:40:17 hweaving: we've also noticed that the difference between SSDs and ramdisk builds isn't actually very large (contrary to our expectations, tbh) Aug 19 17:40:17 * kergoth ponders Aug 19 17:41:19 #2216 fixed by http://git.yoctoproject.org/cgit/cgit.cgi/poky/commit/?id=f03d66bcc1fde80fd8923a400d2f2edfee576068 now breaks Python's distutils.sysconfig, which parses pyconfig.h, but doesn't follow #includes Aug 19 17:42:02 neverpanic: Perhaps a combination of CPU, storage, and scheduling bottlenecks Aug 19 17:45:34 kergoth: any idea what would break if http://git.yoctoproject.org/cgit/cgit.cgi/poky/commit/?id=f03d66bcc1fde80fd8923a400d2f2edfee576068 is reverted? Aug 19 17:45:59 not offhand, should probably check #2216 Aug 19 17:46:35 it seems awefully unrelated; I'm just wondering what use case the multilib_header support solves. Aug 19 17:49:02 pretty sure that's designed to handle the case where a header varies between multilib builds, but of course includedir isn't split by multilib, libdir is, so the file would conflict / be wrong if it wasn't split, so oe_multilib_header moves/renames them with a wrapper Aug 19 17:49:12 * kergoth shrugs, haven't dealt much with it Aug 19 17:52:28 well, this cmake situation is partially resolved. the install to fully qualifed dirs is resolved and it builds to completeion. Now just need to seperate combined make and make install to seperate targets and the bitbake of a saner cmake project can resume again. Aug 19 18:35:48 hey all, I'm working with dizzy (1.7.3) and had a quick question - is there an easy way to run Xsessions as a non-root user? I'm looking at /etc/X11/XServer, which is started by the systemd service xserver-nodm Aug 19 18:36:18 should I change the exec line to run as a different user? Aug 19 18:43:10 So, what do build servers and the like usually run these days? I mean, Linux, yes. But I'm wondering more specifically about distro choices, etc. Aug 19 18:43:31 I'm sort of assuming "gentoo" is probably not my best choice for reproducible builds. Aug 19 18:45:05 start with Ubuntu and go from there, unless you have a soft spot for Fedora :) Aug 19 18:46:23 ya.. F23/F24, or latest Ubuntu Aug 19 19:03:19 I'll download one of those. I have a 14.04 DVD, but that seems old. Aug 19 19:05:50 seebs, you should use overc. :) Aug 19 19:06:04 use yocto to build yocto. Aug 19 19:06:31 and then tell me which parts annoy you so we can improve it. :) Aug 19 20:04:20 man this laptop sucks. it can't boot from a DVD unless the DVD is actually in the drive. what is this, the 1990s? Aug 19 20:04:56 I think I'd have to dig my usb dvd burner out of a closet to even have a dvd drive to boot a dvd nowadays :) Aug 19 21:28:54 kexec-tools compile failing on arm arch with master-next Aug 19 21:29:35 oops - didn't mean to type that in this window... Aug 19 21:39:57 Hmm. I'm guessing https://github.com/openembedded/openembedded-core/compare/master...kergoth:scriptutils-log-improvements is probably reasonable, and might get merged, but would also break at least some of the unit tests Aug 19 21:41:08 guess i should fix those too, and possibly postpone to post 2.2 Aug 19 21:41:26 definitely eases scripting, though **** ENDING LOGGING AT Sat Aug 20 02:59:58 2016