**** BEGIN LOGGING AT Fri Aug 30 02:59:58 2013 Aug 30 03:15:28 http://autobuilder.yoctoproject.org/pub/releases/CURRENT/machines/ has wrong sabresd directory; it is missing a /q/ ;) Aug 30 06:19:12 hi i got the below error while compiling kerenel-3.6.9 for arm arch using poky toolchain,my host system is 64-bit ubuntu 12.04 LTS,the erros is http://pastebin.com/EyBAFRvE,can you help me Aug 30 06:45:51 how to clean all built images ? Aug 30 06:46:49 -c cleanall? Aug 30 06:47:09 that requires a target Aug 30 06:47:13 or not ? Aug 30 06:49:06 qt-x: so? Aug 30 06:49:37 I want to clean all my builds Aug 30 06:50:23 I don't think I\m making any sens Aug 30 06:50:36 well, you can always rm -rf Aug 30 06:50:59 that was the alternative :D Aug 30 06:51:34 how about bitbake world -c cleanall? Aug 30 06:52:09 I'have just tried that it did not work for some recipe Aug 30 06:52:23 well, rm -rf, then. Aug 30 06:52:24 I'have just tried that. It did not work for some recipe Aug 30 06:52:31 on build ? Aug 30 06:52:56 directory Aug 30 06:53:33 well, in python: for image in ('core-image-minimal', ...): subprocess.call(['bitbake', image', '-c', 'cleanall']) or something like that Aug 30 06:54:16 you could get the images dynamically from bitbake-layers Aug 30 06:54:41 so you could write such a simple script. Aug 30 06:54:45 but I guess rm -rf is simpler. Aug 30 06:55:00 btw, I created a bugreport for this a while ago... you can vote if you wish. :D Aug 30 06:56:12 qt-x: https://bugzilla.yoctoproject.org/show_bug.cgi?id=4922 Aug 30 06:56:13 Bug 4922: enhancement, Low, Future, richard.purdie, NEW , There is no clean option for bitbake or a separate util Aug 30 06:56:29 qt-x: a cleanall option for bitbake would help you without target, right ? Aug 30 06:56:37 that is what I wanted to have in there, too. Aug 30 06:57:30 I think so. Aug 30 06:58:00 qt-x: well, comment in there with a +1 then. Aug 30 06:58:13 qt-x: they will see there is more than just one person interested. :D Aug 30 07:02:50 It' becomes a feature :D Aug 30 07:03:34 well, only if they see there is enough interest, I guess. Aug 30 07:05:07 qt-x: are you a qt developer? Aug 30 07:07:12 I play with qt when the situation requires it. Aug 30 07:08:55 lpapp: anyway if added a +1 depending on the implication of such a command they might do'it. Aug 30 07:09:16 qt-x: yes, saw, thanks. Aug 30 07:09:24 I need to leave now. Aug 30 07:09:37 have a nice day nigth Aug 30 07:21:13 hi i have used kernel-3.6.9 and configure for at91_dt_defconfig, and my host is ubuntu 64 bit,i have used the yocto toolchain for cross compilation,i got the below error while compiling kernel http://pastebin.com/JGeXL8Q4 can you please tell me that issues Aug 30 08:11:56 morning all Aug 30 08:27:44 How do people here usually ensure newly-flashed boxes have the correct time set? Do you just set up NTP, or is there some way to ensure that date is set properly when flashing? Aug 30 08:28:44 ntp is the way to go, if you are network-connected Aug 30 08:29:17 tf, Alright, that's what I'll go for, then. Aug 30 08:29:20 but you need a sutiable ntp pool Aug 30 08:29:34 particularly, if you are doing a commercial product Aug 30 08:29:40 tf, Yea I saw something mentioned about that very briefly on the wiki. Aug 30 08:29:52 tf, Hmm we sorta are. Does that mean we need to buy-in to a NTP server then? Aug 30 08:32:21 1s Aug 30 08:33:41 tf, no prob. Aug 30 08:35:03 Stygia: to use ntp, you have to request an ntp pool for your project Aug 30 08:35:39 tf, Yea, I saw a page on the wiki (and on pool.ntp.org) to indicate as much. We will do that before deployment, then. :) Aug 30 08:36:50 Stygia: http://www.pool.ntp.org/vendors/contribute.html Aug 30 08:37:35 Stygia: when you have a pool, you can then enable ntp in connman, which JustWorks Aug 30 08:37:53 tf, isn't connman a GUI tool? Aug 30 08:38:06 tf, But I'll look into it. :) We have a Friday meeting now. Talk to ya later, and thanks. Aug 30 08:38:35 Stygia: https://github.com/Guacamayo/meta-guacamayo/blob/master/meta-guacamayo/recipes-connectivity/connman/connman_1.5.bbappend Aug 30 08:38:59 connman's not well documented, but that settings file has what you need Aug 30 09:14:28 Morning all Aug 30 09:34:25 So, I've got a recipe that builds fine via bitbake. However, when I build an image that includes it I just get : Aug 30 09:34:32 | Collected errors: Aug 30 09:34:43 | * opkg_install_cmd: Cannot install package pevs-package Aug 30 09:35:20 pev: was pevs-package created? it may have been defined but empty and thus never produced in the end Aug 30 09:36:05 Ah... That to do with the contents of FILES_${PN}? That rings a bell... Aug 30 09:36:40 that variable is how you influence what goes into each package yes Aug 30 09:36:51 Righto, I'll have a dig, thanks Aug 30 09:36:59 if you want to see the contents of each package you can have a look under packages-split in the workdir for the recipe btw Aug 30 09:38:35 Yep, that'll be the problem then :-) Aug 30 09:38:56 Hm, would be nice to add a warning into the build system looking for an empty dir there as that would be easy to spot? Aug 30 09:40:06 I've been thinking about a way to warn if a package depended upon by another package wasn't produced; it would certainly be helpful Aug 30 09:40:14 just not quite sure how to implement the check Aug 30 09:42:18 find /path/name -maxdepth 0 -empty -exec echo {} is empty. \; Aug 30 09:42:45 Not sure where in the build system though - surely at the loop where it's trying to package each test before doing the packaging Aug 30 09:43:21 right, sure :) it's how to find out what depends on each package in the context of the QA checks in do_package that I'm not sure of Aug 30 09:45:12 Do you actually need to care about dependencies? Aug 30 09:47:30 well yes... a package not being produced that nothing depends on isn't a problem Aug 30 09:50:25 OK, another idiot question. My recipie builds my makefiles 'all' rule by doing "bitbake pevs-test" but what do I do to provoke the install rule to test? Is there a bitbake command to do all the packaging? Aug 30 09:52:12 pev: if it's a pure makefile and not autotools or something similar you need to define a do_install function which will run "make install" (usually oe_runmake install with some arguments to ensure it installs to ${D}) Aug 30 09:52:31 Ahhh! That's the bit I didn't know Aug 30 09:53:38 as for packaging, once stuff goes into ${D} (the "image" subdir of the workdir, if you want to examine it) packaging based on the values of FILES and PACKAGES (both of which have sane defaults, but can be extended/overridden) should be automatic Aug 30 09:58:46 OK, I added oe_runmake install DESTDIR=${D} Aug 30 09:58:53 which is used by other recipes Aug 30 10:00:19 now I get "ERROR: Function failed: split_and_strip_files" Aug 30 10:00:25 and an objcopy error Aug 30 10:01:57 ERROR: objcopy failed with exit code 1 (cmd was 'arm-amltd-linux-gnueabi-objcopy' --only-keep-debug '/workdrive/projects/pevtest/src/pevtest_build/tmp/work/armv7a-vfp-neon-amltd-linux-gnueabi/pevtest-system/1.0-r0/package/usr/bin/pevtest-server' '/workdrive/projects/pevtest/src/pevtest_build/tmp/work/armv7a-vfp-neon-amltd-linux-gnueabi/pevtest-system/1.0-r0/package/usr/bin/.debug/pevtest-server') Aug 30 10:03:37 so, in current versions we print what the actual error is from that command Aug 30 10:03:44 at least we should be Aug 30 10:04:55 you could try bitbake -c devshell pevtest and run that command manually in the shell to see what it says Aug 30 10:05:58 arm-amltd-linux-gnueabi-objcopy:/workdrive/projects/pevtest/src/pevtest_p81_build/tmp/work/armv7a-vfp-neon-amltd-linux-gnueabi/pevtest-system/1.0-r0/package/usr/bin/pevtest-server: File format not recognized Aug 30 10:06:04 from the logfile Aug 30 10:06:59 file /workdrive/projects/pevtest/src/pevtest_p81_build/tmp/work/armv7a-vfp-neon-amltd-linux-gnueabi/pevtest-system/1.0-r0/package/usr/bin/pevtest-server Aug 30 10:07:22 pevtest-server: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked (uses shared libs), for GNU/Linux 2.6.24, BuildID[sha1]=0xa942bef03d9cf29148797302bcd7c7b4bc9e7e68, not stripped Aug 30 10:09:55 Ah Aug 30 10:10:09 I see. It hasn't cross compiled for some reason... :-D Aug 30 10:11:58 Hmpf. So what does bitbake do when invoking my makefile that I'm missing...? Aug 30 10:13:31 right Aug 30 10:13:35 I suspected as much Aug 30 10:14:01 although really it should have warned about this earlier; it's really a bug that you get this error first instead Aug 30 10:14:40 so with makefiles it's hard for us to consistently be able to specify the right options to enable cross-compilation Aug 30 10:14:49 so you basically need to do this yourself Aug 30 10:15:15 often it's as easy as passing in values for variables such as CC etc. Aug 30 10:15:26 doesn't bitbake pass in a CROSS variable or overide CC or something similar? Aug 30 10:16:03 it does set those variables, but you may have to tell make to pull them in Aug 30 10:16:58 pev: e.g. http://www.yoctoproject.org/docs/current/dev-manual/dev-manual.html#usingpoky-extend-addpkg-makefile Aug 30 10:17:34 I was defining CC=gcc in the Makefile... Aug 30 10:17:49 Changing that to CC?=gcc seems to do the trick. (ish!) Aug 30 10:17:58 see http://cgit.openembedded.org/openembedded-core/tree/meta/recipes-devtools/mtd/mtd-utils_1.5.0.bb Aug 30 10:18:05 right, I suspect forcing it on the command line would do the same thing Aug 30 10:18:07 pev: ^^ this is one rich example Aug 30 10:19:08 ant_work: That looks interesting, thanks! Aug 30 10:19:17 pev: or th esimple one http://cgit.openembedded.org/meta-handheld/tree/recipes-bsp/zaurus-utils/nandlogical-klibc_1.0.0.bb Aug 30 10:19:53 pev: sorry not that one for klibc.. this one Aug 30 10:19:56 http://cgit.openembedded.org/meta-handheld/tree/recipes-bsp/zaurus-utils/nandlogical_1.0.0.bb Aug 30 10:38:39 Result! Thanks for the help guys, got that nailed and in the build finally... Now just have to mess around with creating the modified sysinit files again... Yuck! Aug 30 10:39:13 Oh no, spoke too soon Aug 30 10:39:20 still hasnt cleared the opkg error... Aug 30 10:39:40 Meh... Aug 30 10:49:32 RP: got a clue for this? https://bugzilla.yoctoproject.org/show_bug.cgi?id=5043 Aug 30 10:49:33 Bug 5043: enhancement, Low, Future, richard.purdie, NEW , Extend the layer creation documentation with own bblayers sample creation Aug 30 10:50:06 is there any good workaround? Aug 30 10:50:12 or undocumented feature? Aug 30 10:52:20 pev: I'm guessing the output package is still empty Aug 30 10:52:32 pev: you might want to take a look at what it's installing Aug 30 10:52:37 bluelightning: hey Aug 30 10:52:48 pev: you should be getting warnings if there are unpackaged files though Aug 30 10:52:52 lpapp: hi Aug 30 10:53:19 bluelightning: is there a way to run ./boot.sh (custom shell script) in do_compile before the regular configure/make? Aug 30 10:53:25 bluelightning: simply putting boot.sh in there did not work Aug 30 10:53:34 I think it was looking into the wrong folder. Aug 30 10:53:45 boot.sh will be in ..../git/ because the source is fetched from git. Aug 30 10:54:04 bluelightning: btw grep reveals an ample range of syntaxes for the vars appended to oe_runmake Aug 30 10:54:07 lpapp: is S set to point to "${WORKDIR}/git" ? Aug 30 10:54:24 bluelightning: yes Aug 30 10:54:30 ant_work: right, probably depends on who wrote it Aug 30 10:54:43 bluelightning: it's just about enclosing the vars in ' Aug 30 10:54:46 mostly Aug 30 10:54:55 lpapp: then that will be the current directory when do_configure executes Aug 30 10:55:01 bluelightning: S = "${WORKDIR}/git" Aug 30 10:55:04 inherit autotools Aug 30 10:55:13 lpapp: what is in boot.sh anyway? Aug 30 10:55:14 do_compile() { Aug 30 10:55:20 boot.sh Aug 30 10:55:20 } Aug 30 10:55:32 lpapp: ./boot.sh perhaps? Aug 30 10:55:33 bluelightning: autoreconf+custom stuff. Aug 30 10:55:58 bluelightning: http://yuba.stanford.edu/git/gitweb.cgi?p=openflow.git;a=blob;f=boot.sh;h=97921fac0a61267ae8c8bf339331f5ea757bb9cf;hb=HEAD Aug 30 10:56:08 lpapp: probably you should separate out the custom stuff and do that in do_configure leaving autotools to do the autoreconf Aug 30 10:56:40 bluelightning: it was because I was doing multiple packages. Swapping it back to a single one sorted it fine - isn't a big deal so have left it that way. Thanks for the advice! Aug 30 10:56:49 bluelightning: I am just using the upstream software. Aug 30 10:56:50 pev: np Aug 30 10:57:35 Error, TMPDIR has changed location. You need to either move it back to /home/lpapp/Projects/Yocto/poky-dylan-9.0.1/build/tmp or rebuild Aug 30 10:57:38 what? Aug 30 10:57:44 lpapp: having looked in that file just now I can't see anything of value Aug 30 10:57:45 there is no way to simply override a variable after a move? Aug 30 10:57:50 do I really need to rebuild stuff?! Aug 30 10:58:06 lpapp: just let autotools.bbclass do what is needed Aug 30 10:58:11 bluelightning: their configure depends on deb stuff Aug 30 10:58:14 like it or not. Aug 30 10:58:29 and that script generates that at the beginning... Aug 30 10:58:36 that's unfortunate Aug 30 10:59:20 bluelightning: right Aug 30 10:59:20 lpapp: yep, because upstream software hardcodes paths into its executables, not our fault Aug 30 10:59:35 that is in reply to TMPDIR? Aug 30 10:59:38 lpapp: yes Aug 30 10:59:46 :'( Aug 30 10:59:59 what is the proper command to rebuild stuff in this case? Aug 30 11:00:01 rm -rf manually? Aug 30 11:00:13 delete the entire TMPDIR or move it back Aug 30 11:01:16 if you have a sstate folder, deleting TMPDIR, isn't such a big deal. takes less than 3 mins for me to 'rebuild' everything, which in turns just uses stuff from sstate Aug 30 11:03:39 hmm, it is kinda weird I have an image name like this ... uImage--3.2.1- Aug 30 11:03:42 why double dash? Aug 30 11:04:18 * lpapp has deleted the tmp folder Aug 30 11:04:21 variable probably not set I'm guessing Aug 30 11:04:35 which is kinda weird because it used to work Aug 30 11:04:43 * lpapp guesses it is some kernel variable. Aug 30 11:06:14 LINUX_VERSION_EXTENSION ?= "-foo-${LINUX_KERNEL_TYPE}" Aug 30 11:06:16 I have this Aug 30 11:07:10 and this, KERNEL_RELEASE = "3.2.1" Aug 30 11:07:22 and this, KERNEL_IMAGETYPE = "uImage" Aug 30 11:07:27 is any of those wrong? Aug 30 11:09:06 and this, KERNEL_RELEASE = "3.2.1"LINUX_VERSION_EXTENSION ?= "-yocto-${LINUX_KERNEL_TYPE}" Aug 30 11:09:14 yocto also uses LINUX_VERSION_EXTENSION ?= "-yocto-${LINUX_KERNEL_TYPE}" Aug 30 11:10:26 I'd suggest looking at the bitbake -e output to see what the history for KERNEL_IMAGE_BASE_NAME is Aug 30 11:10:35 then you'll be able to see which variable is apparently blank Aug 30 11:11:28 is that ok without cleaning up? Aug 30 11:11:46 bitbake -e polatis-core-image | grep KERNEL_IMAGE_BASE_NAME Aug 30 11:11:48 empty output Aug 30 11:12:40 same stuff for modules fwiw, modules--3.2.1... Aug 30 11:13:20 it's from empty PE Aug 30 11:13:43 PE? Aug 30 11:14:02 package epoch Aug 30 11:14:23 the epoch of the recipe. The default value is "0". The field is used to make upgrades possible when the versioning scheme changes in some backwards incompatible way. Aug 30 11:14:26 yeah, found it. Aug 30 11:14:31 not sure why that would be needed. Aug 30 11:14:36 the yocto kernel does not seem to use that either. Aug 30 11:14:59 KERNEL_IMAGE_BASE_NAME is used by more kernel recipes then linux-yocto Aug 30 11:15:31 so? Aug 30 11:15:59 (it was not like this with denzil btw) Aug 30 11:16:07 (denzil generated single dash) Aug 30 11:17:39 grep -rn PKGE ./meta-* -> returns empty. Aug 30 11:17:59 ./meta/conf/bitbake.conf:199:PKGE ?= "${@['','${PE\x7d'][d.getVar('PE',1) > 0]}" Aug 30 11:18:22 ok, that is beyond my league, I am afraid. Aug 30 11:18:52 so what is causing regression from denzil to dylan update, and how can it be resolved? Aug 30 11:20:20 lpapp: http://cgit.openembedded.org/openembedded-core/commit/?id=3b5fe6d2cd58b5db70c3d319ffc66b1cbff1f43a Aug 30 11:20:31 I just searched the git log for PKGE to find that btw Aug 30 11:21:03 then that Aug 30 11:21:04 http://cgit.openembedded.org/openembedded-core/commit/meta/classes/kernel.bbclass?id=1392f959cb8cd50b5a4492899e54f3ed68ef56d7 Aug 30 11:21:32 git blame would have also got the answer Aug 30 11:21:43 yes, lpapp your0e rather lazy ;) Aug 30 11:22:07 I really do not know how it is "lazy" to follow the advice of not having a git repository .... Aug 30 11:22:17 and using a tarball where git is *obviously* out of scope. Aug 30 11:23:17 anyway it's not good if you get it empty Aug 30 11:23:54 well, that is the default. Aug 30 11:24:33 actually, I do not see the Yocto kernel setting that either. Aug 30 11:25:07 this change should have preserved the compatibility, i.e. single dash is the default if not specified, otherwise double dash Aug 30 11:25:14 it breaks the interface... Aug 30 11:25:39 you are running a prserver? Aug 30 11:25:45 ? Aug 30 11:26:03 bluelightning: link please ;) Aug 30 11:26:33 https://wiki.yoctoproject.org/wiki/PR_Service Aug 30 11:26:35 anyway, what is the workaround? Aug 30 11:26:43 I do not wanna see double dash Aug 30 11:26:45 that is silly. Aug 30 11:26:47 since when is filename "interface"? Aug 30 11:26:54 JaMa: since ever Aug 30 11:27:12 SRCREVs were also shortened, so it will be different either way Aug 30 11:27:27 well, you do not care about that. Aug 30 11:27:32 if you want the same then change KERNEL_IMAGE_BASE_NAME or continue to use old version Aug 30 11:27:34 usually when you archive, SRCREV is blobbed. Aug 30 11:27:46 not to mention, the yocto kernel does not change the PE Aug 30 11:27:51 well, this looks like a documentation stuff Aug 30 11:27:57 if it is not in the migration guide yet. Aug 30 11:28:02 * lpapp is checking Aug 30 11:28:07 link to latest is the same Aug 30 11:28:08 lpapp: check your local.conf Aug 30 11:28:17 PRSERV_HOST = "localhost:0" Aug 30 11:28:31 recommended also Aug 30 11:28:32 INHERIT += "buildhistory" Aug 30 11:28:32 BUILDHISTORY_COMMIT = "1" Aug 30 11:28:39 yeah, nothing in here about it: http://www.yoctoproject.org/docs/1.3/poky-ref-manual/poky-ref-manual.html#migration Aug 30 11:28:45 I will submit a bugreport for it. Aug 30 11:29:22 ant_work: let us stay on topic. ;) Aug 30 11:30:04 lpapp: "...data is obtained from the PR service..." Aug 30 11:32:10 where should one submit bugs/features for individual recipes? Aug 30 11:32:30 i've patched the mpd recipe so it optionally supports MPD's new CD playback features Aug 30 11:32:45 https://bugzilla.yoctoproject.org/show_bug.cgi?id=5089 Aug 30 11:32:46 Bug 5089: normal, Undecided, ---, scott.m.rifenbark, NEW , Missing KERNEL_IMAGE_BASE_NAME in the migration guide Aug 30 11:33:20 so the next question is: why is the yocto kernel not using this new "feature"? Aug 30 11:34:19 also, why is it silly by default with "--"? Aug 30 11:34:26 if anything, it should be -0- or so Aug 30 11:34:39 "--" looks silly, or is it just me? Aug 30 11:34:53 with -0- people will ask what's the "0" Aug 30 11:35:04 +1 Aug 30 11:35:06 PR is appended to packagename also only when != 0 Aug 30 11:35:06 the default value? Aug 30 11:35:14 s/PR/PE/ Aug 30 11:35:32 well, it should be either single dash, or -0- Aug 30 11:35:36 double dash is silly IMHO Aug 30 11:35:46 it has no any additional value to a single dash Aug 30 11:35:50 lpapp: the default value for PKGE is empty, same answer as "0" Aug 30 11:36:05 no Aug 30 11:36:08 '--' is silly Aug 30 11:36:15 it does not contribute much to '-' Aug 30 11:36:20 and -0 does? Aug 30 11:36:23 yes Aug 30 11:36:27 because that means a number Aug 30 11:36:30 not a string placeholder Aug 30 11:36:37 not an "enum" etc. Aug 30 11:36:41 it means the PE "version" Aug 30 11:36:47 which is zero (the default) Aug 30 11:37:06 so I propose to change it either to '-' or '-0-' unless I am missing something. Aug 30 11:38:25 modules--3.2.1-r2-foo-20130829152142.tgz Aug 30 11:38:28 hmm, that changed too Aug 30 11:38:32 the datum was not there before! Aug 30 11:39:02 thanks to this, I will be busy today with fixing CI issues. :( Aug 30 11:40:19 so finally fix your silli CI scripts Aug 30 11:40:33 s/li/ly/ Aug 30 11:40:37 date* Aug 30 11:41:12 pff Aug 30 11:41:29 "silly" CI script because someone decides it is cool to break stuff for many end users due to such a nuance. Aug 30 11:41:43 actually, for compatibility, I prefer '-' over '-0-' Aug 30 12:03:57 bluelightning: any reason why my generated rpm packages are empty for my own software? Aug 30 12:04:06 bluelightning: the default do_install does not put things right in there? Aug 30 12:22:36 do I need to inherit anything for the proper packaging, or is there some documentation about it, and how to avoid empty packageS? Aug 30 12:22:39 packages?* Aug 30 12:36:50 JaMa: perhaps BUGTRACKER could be added to the qt5 packages? Aug 30 12:37:09 Shall I submit such a change? Aug 30 12:45:24 oh, I need FILES for pretty much anything? Aug 30 12:45:48 but but but... the debian packaging works the way that if you do not pass anything, you will get one monolithic package. I would be more than happy with that for now. Is there such an option to turn on? Aug 30 12:57:34 I see that certain packages do not use the FILES variable, just install in the do_install... are they equivalent? Aug 30 13:04:11 which way is recommended? Aug 30 13:07:23 lpapp: maybe they just rely on the 'default' FILES valueS? Aug 30 13:08:34 ndec: which is? Aug 30 13:08:50 in meta/conf/bitbake.conf Aug 30 13:09:00 I cannot read that Aug 30 13:09:08 what do you mena? Aug 30 13:09:11 mean? Aug 30 13:10:03 right, so installation or FILES are needed Aug 30 13:10:15 both are needed Aug 30 13:10:33 well, no. Aug 30 13:10:36 I can work it without FILES Aug 30 13:10:41 packages are created by taking files from where they have been installed. Aug 30 13:10:45 actually, I even think this should be documented. Aug 30 13:10:54 you do not necessarily need FILES if you install stuff correctly. Aug 30 13:11:22 there are sane default for FILES (for all packages, including -dev, -dgb, ... too). most software would work fine with these default. Aug 30 13:11:48 yep Aug 30 13:11:50 you might to play with FILES if you have more complex packages. Aug 30 13:11:54 I think this should be documented. Aug 30 13:11:57 I will open a bugreport Aug 30 13:12:04 for what? Aug 30 13:12:05 this is not mentioned under the FILES var. Aug 30 13:12:11 to mention that you do not need it for simple stuff Aug 30 13:12:14 simple and sane Aug 30 13:12:20 you just need to install is all Aug 30 13:13:13 well, most variables have 'sane' default values, if you go this path... Aug 30 13:13:33 I am not sure about it. Aug 30 13:13:44 but when I go to FILES, I would like to read that it is not necessary for simple use cases Aug 30 13:13:46 it's not about FILES so it's not something to add to the variable glossary entry for FILES Aug 30 13:13:48 you just need to make it is installed. Aug 30 13:13:53 sure, it is. Aug 30 13:13:58 you're missing do_install Aug 30 13:13:59 I am actually filing a doc bugreport about it. Aug 30 13:14:09 no changing of FILES will help you around that Aug 30 13:14:17 sure, it will. Aug 30 13:14:32 how can I force bitbake recipe parsing cache to refresh? Aug 30 13:14:35 I'm sure it won't, not if do_install isn't defined Aug 30 13:14:43 I changed one line in machine config and bitbake uses the old one Aug 30 13:15:01 bluelightning: you know better what would help me? :D Aug 30 13:15:08 eren: rm -rf tmp ;-) Aug 30 13:15:13 err no Aug 30 13:15:30 eren: it should do that automatically, something is not right if it isn't Aug 30 13:15:48 eren: or, you changed it back to a value that you had previously and it's using the cache from the last time you set that Aug 30 13:15:50 * lpapp has always lacked ldd for the arm toolchain. Aug 30 13:16:25 bluelightning: now it parsed Aug 30 13:16:51 bluelightning: I added one line in IMAGE_FSTYPES, it did not parse the recipes and got directly from cache Aug 30 13:17:03 i removed all the lines except the one I want, it parsed the recipes Aug 30 13:21:04 bluelightning: I am not disagreeing about missing do_install Aug 30 13:21:15 but I disagree about not making a reference and a note about default to FILES Aug 30 13:21:22 reference to do_install Aug 30 13:21:30 there's only so much you can put in a glossary entry Aug 30 13:21:48 we're planning an expanded section on how to write new recipes to be written in 1.6 Aug 30 13:22:02 that's where that kind of thing belongs, IMO Aug 30 13:22:22 it belongs to there, too Aug 30 13:22:33 one is glossary, the other is example put up with glossary entries. Aug 30 13:22:43 there is also a new section coming in the 1.5 manual explaining how the build system generally operates and points to variables used during each stage of the build Aug 30 13:23:00 you can even look at it in the "latest" version of the manuals on the website Aug 30 13:23:39 I do not dare. :D Aug 30 13:25:48 bluelightning: in which manual is that new section? Aug 30 13:25:53 hi there, in plenty or recipes people use 'install' to copy files to image. i tried rsync instead and Yocto says : WARNING: QA Issue: galileo-target: Files/directories were installed but not shipped Aug 30 13:26:31 ndec: the ref manual I believe Aug 30 13:26:32 is rsync not liked by Yocto? Aug 30 13:27:20 Krz_: forgive me for asking but why would you need to use rsync in do_install? Aug 30 13:28:08 sorry, I used rsync in Makefile, to let non-Yocto users have the same as Yocto users Aug 30 13:28:29 and in do_install I just call make install-mytarget Aug 30 13:28:37 which call rsync from Makefile Aug 30 13:28:41 does that make sense? Aug 30 13:29:06 Krz_: I'm understanding a little more but still not why rsync would be useful in that particular context Aug 30 13:30:01 so basically in my source package I created whole directory tree instead of hiding instructions on how to build it inside Makefile Aug 30 13:30:12 so then non-Yocto users can just copy whole tree to the image Aug 30 13:32:41 Krz_: there's nothing inherent about do_install that would prevent rsync being used, but remember that do_install is just about installing files to the holding area in ${D} before packaging, not actually into the image Aug 30 13:33:55 And wouldn't something that used rsync to (presumably) fetch files be better placed in do_fetch or something? This statement has the caveat that I haven't been following Krz_ 's usecase. Aug 30 13:35:31 in my source package I have: *.c files to create binary and few *.sh scripts just to be copied over Aug 30 13:36:34 now when I do 'make install' it rsyncs source-package/my/tree/*.sh Aug 30 13:36:43 and rsyncs binary as well Aug 30 13:37:49 thanks to that someone without Yocto can just go to my package, do 'make install DESTDIR=/path/to/my/image' Aug 30 13:38:30 I'm still confused Aug 30 13:38:43 make install can always do that without rsync Aug 30 13:38:59 so what does rsync enable here? Aug 30 13:39:27 my tree looks like: src-package/etc/my/directory/few-scripts.sh Aug 30 13:39:37 root directory is src-package Aug 30 13:40:07 rsync just takes whole tree and copies it Aug 30 13:40:23 wheres using 'install' for that is more compilicated Aug 30 13:40:56 so I used rsync - one liner Aug 30 13:42:41 Krz_: if it's only because you're wanting to copy a full directory tree, surely cp -r would work and be much simpler? Aug 30 13:43:20 yeah, but cp does not preserve attributes Aug 30 13:43:34 it has an option to do that... Aug 30 13:43:56 -p Aug 30 13:44:18 hmm, I can try that Aug 30 13:44:29 will not cp issue the same QA WARNING? Aug 30 13:44:35 I mean, rsync will work, it just seems like overkill for this Aug 30 13:44:53 whatever technique you use, you absolutely must install to ${D} Aug 30 13:44:58 within do_install Aug 30 13:45:28 and secondarily, the value of FILES for each package must be set to match whereever you are installing things to Aug 30 13:45:39 but FILES_${PN} already includes ${sysconfdir} which is /etc Aug 30 13:45:58 so I wouldn't have thought it necessary to touch FILES just for stuff in /etc Aug 30 13:46:27 my do_install look like that: oe_runmake install-scripts DESTDIR=${D}/ Aug 30 13:47:11 then Makefile picks up $DESTDIR and rsyncs the tree (which I might switch to cp -p as well) Aug 30 13:47:41 the main problem is with Yocto QA WARNING which causes files not to be installed in image Aug 30 13:48:03 Krz_: the QA warning is not what causes files not to be installed in the image Aug 30 13:48:30 Krz_: it's warning you that you haven't done the right thing to have those files end up in a package such that they can get in the image - there's a difference :) Aug 30 13:48:45 FWIW, that line looks normal, so it must be where it is putting things under ${D} Aug 30 13:49:00 I'd suggest looking in ${D} for the recipe Aug 30 13:49:42 this will tell you where that is exactly: bitbake -e recipename | grep ^D= Aug 30 13:50:29 ok I understand it more now :) Aug 30 14:00:53 it seems that I cannot run a python function inside bbclass file Aug 30 14:01:19 simply calling "build_syslinux_cfg()" gives me parse error Aug 30 14:03:16 eren: how have you defined the function? Aug 30 14:04:51 good morning ... anyone around that could assist with a couple of BSP/Kernel-related questions? Aug 30 14:05:11 bluelightning: IMAGE_CMD_alix-hddimage() { Aug 30 14:05:23 Hi all, I'm trying to build a minimal image for MIPS with DEFAULTTUNE = "mips64-n32", and I get a failure when building busybox: mips64-poky-linux-gnun32-ld: Attempt to do relocatable link with elf32-ntradbigmips input and elf64-tradbigmips output; mips64-poky-linux-gnun32-ld: applets/applets.o: file class ELFCLASS32 incompatible with ELFCLASS64 Aug 30 14:05:48 bluelightning: I called the python function with @{@bb.build.exec_func('build_syslinux_cfg', d)} Aug 30 14:06:04 I started with a clean workspace. Am I doing something wrong? Aug 30 14:06:12 bluelightning: it returns None and build halts. Then I decided RETVAL="${@bb.build....}" Aug 30 14:06:18 however, now syslinux config is not created Aug 30 14:08:00 eren: er, but those are two different functions Aug 30 14:08:33 so should I call ${@build_syslinux_cfg(d)} ? Aug 30 14:09:27 well you wouldn't just state it like that in the recipe Aug 30 14:09:55 in a variable value, yes, but then this doesn't sound like the kind of function that should be called that way Aug 30 14:10:01 that reminds me that with new syslinux bbclass it's impossible to disable console, I should file a bug about that Aug 30 14:10:16 actually what I am trying to do is to have image creation just like raspberry. I am editing boot-directdist.bbclass Aug 30 14:11:08 eren: basically then you need to have that function called from a task function Aug 30 14:11:41 I put that into IMAGE_CMD_alix-hddimage() {}, added initrd support. However, boot-directdisk.bbclass calls bb.build.exec_func('build_syslinux_cfg') before creating the image Aug 30 14:11:41 eren: I haven't read backlog yet, but please see my comments in https://bugzilla.yoctoproject.org/show_bug.cgi?id=3944 Aug 30 14:11:42 Bug 3944: enhancement, Medium, 1.4, jason.wessel, REOPENED , syslinux.bbclass needs some work to add serial support and a real menu Aug 30 14:11:54 I have an ARM-based development board that used to be supported by OE, and is not part of the Yocto/Poky build. The vendor provides a custom kernel, which I can build outside of Poky without a problem. I'd like to incorporate it into the Poky build and make a BSP for the board, but I'm at a loss on how to get it in there. Aug 30 14:12:04 bluelightning: and yes, it is called from a task function Aug 30 14:12:16 bluelightning: so there is no way to simply call build_syslinux_cfg? Aug 30 14:12:20 wmcdevel: have you looked at the Yocto Project BSP guide and kernel manuals? Aug 30 14:12:34 eren: call in what context? Aug 30 14:13:20 bluelightning: yes. I have gotten as far as creating the BSP layer and some sample bitbake recipes, but I can't seem to figure out how to get the build to not use the linux-yocto kernel. Aug 30 14:13:31 bluelightning: IMAGE_CMD_foobar() context Aug 30 14:14:12 wmcdevel: PREFERRED_PROVIDER_virtual/kernel = "your-kernel-recipe-name" Aug 30 14:14:45 wmcdevel: that would be set in your machine's conf/machine/xyz.conf file Aug 30 14:14:53 s/would/should/ Aug 30 14:15:49 bluelightning: btw, I am working on this: http://lists.openembedded.org/pipermail/openembedded-devel/2013-August/091956.html Aug 30 14:16:16 bluelightning: ok. do I need anything in particular in my kernel recipe? for example, I've seen where Git is used, quilt, etc. I have a plain-old tarball to work with. Aug 30 14:18:09 wmcdevel: not really... but for that though I would suggest using the linux-yocto-custom recipe (under meta-skeleton/recipes-kernel/, also mentioned in the kernel manual) as a template Aug 30 14:19:01 bluelightning: ok, cool. will fiddle around with that a bit and see where I get. thanks Aug 30 14:23:35 Garibaldi|work: sounds like a question for khem perhaps Aug 30 14:25:55 running this on the host side now, LD_LIBRARY_PATH=$LD_LIBRARY_PATH:. gdb --args Aug 30 14:25:59 sorry, wrong window. Aug 30 14:28:28 bluelightning: thanks, I'll keep an eye out :-) Aug 30 14:33:32 eren: right Aug 30 14:33:46 eren: but that's a shell function if I'm not mistaken Aug 30 14:35:02 eren: you cannot call python functions from shell functions unless it's something that is appropriate to be run when the function is parsed, which is what will happen if you use ${@...} inside a shell function Aug 30 14:35:54 bluelightning: I added a new task before do_rootfs Aug 30 14:36:33 it works ok, creates syslinux.cfg but at one time ${S} is deleted Aug 30 14:37:44 bluelightning: is there any way to see the order of the tasks? Aug 30 14:38:07 -c liststasks does not give me in which order bitbake runs tasks Aug 30 14:38:28 eren: after they've executed you can look at log.task_order next to the task logs for the recipe Aug 30 14:39:00 okkie Aug 30 14:40:47 do I need a FILES_${PN} in my recipe to let files be installed into image? Aug 30 14:42:10 Krz_the default value is sane for most recipes but it depends on where files are intended to be installed to within the image Aug 30 14:42:45 Krz_: you may wish to look at the default values for FILES for each package in meta/conf/bitbake.conf Aug 30 14:44:09 bluelightning: I have custom dir under /opt to install my files - I think in that case I have to specify that in FILES_${PN}... Aug 30 14:45:54 Krz_: right, you would yes, /opt is not part of the default value Aug 30 14:46:14 bluelightning: ok, so that was my problem and reason for QA WARNING Aug 30 14:51:15 ok, makes sense Aug 30 15:13:18 Hey. I was under the impression that do_install_append's would stack, so that if I have one in my bbappend, and there is one in the core recipe.inc (ntp in this case), they would be applied in turn. However, my very simple do_install_append function that simply creates a symlink causes the whole mess to fail. What could cause this? Aug 30 15:13:56 Stygia: they will stack Aug 30 15:13:59 Stygia: what's the failure? Aug 30 15:14:18 bluelightning, Just as sec and I'll reproduce it and put in it pastebin Aug 30 15:16:07 bluelightning, The recipe and the part of the error I think is the relevant bit: http://pastebin.com/Me7xTbhF Aug 30 15:16:27 Stygia: missing ${D} in front of the paths Aug 30 15:17:00 bluelightning, Ah shit. Heh I wanted to call in sick today (I'm on like 4 hours sleep), but we're busy, so. Sorry. Aug 30 15:17:31 Stygia: nm, I make mistakes like that all of the time :) Aug 30 15:17:53 Stygia: also, use bitbake -e to confirrm what you think is happening is actually happening. you'd see both appends occurred by examining the value of do_install :) Aug 30 15:17:56 bluelightning, I just copy-pasted the command I wanted that'd already worked on the box. Aug 30 15:18:08 though bb-show makes that a littlel easier since you can't grep out a function definition trivially.. Aug 30 15:18:11 heh Aug 30 15:18:11 * kergoth yawns Aug 30 15:18:20 kergoth, I wish I could say man bitbake. Aug 30 15:18:38 bitbake -h covers the usage, but admittedly has no examples or anything :) Aug 30 15:18:46 iirc someone is working on improved bitbake docs for yocto Aug 30 15:19:11 Stygia: you can do: man ../bitbake/doc/bitbake.1 Aug 30 15:19:19 though it's pretty sparse Aug 30 15:19:27 huh, good point Aug 30 15:19:32 forgot that exised :) Aug 30 15:19:48 I almost did as well until Stygia brought it up :) Aug 30 15:19:53 Fancy. Wrote that all down, then. Aug 30 15:20:21 heh, http://explainshell.com/explain/bitbake?args=-p+-u+depexp+-vDDD Aug 30 15:20:32 apparently they're using an old man page there, doesn't understand -u :) Aug 30 15:20:43 interesting site though, http://explainshell.com/explain/iptables?args=-A+INPUT+-i+eth0+-s+ip-to-block+-j+DROP Aug 30 15:21:03 oo fancy Aug 30 15:22:11 hmm, seemingly failed on cpio though Aug 30 15:22:16 indeed. not sure how smart it is, and it can't handle pipelines yet, but i could see it being usefulu for confused users looking at snippets on occasion Aug 30 15:22:21 huh Aug 30 15:22:29 Hello, what is the difference between native and nativesdk? I have seen BBCLASSEXTEND set to this. Aug 30 15:22:32 And huh. That looks cool. Aug 30 15:23:24 johnsoga: native = something to run on the build host; nativesdk = something to go into an installable SDK Aug 30 15:24:04 moin Aug 30 15:24:50 bluelighting: is it very often that a package needs to change between the build host and the sdk? Aug 30 15:25:17 the sdk can target a different host, e.g. i686 vs x86_64, for one Aug 30 15:25:26 for two, nativesdk recipes are packaged Aug 30 15:25:39 i think those are the main differences, but someone else would know better than i Aug 30 15:25:46 kergoth: ah, ok that makes sence Aug 30 15:26:42 Hmm. In a do_install_append, I have this: ln -s ${D}/usr/share/zoneinfo/Europe/Copenhagen ${D}/etc/localtime Aug 30 15:26:47 So my next question is what is the best way to determine that a bitbake file is being run as native or nativesdk so that I can change the configuration options? Aug 30 15:26:55 It now runs without errors, but after flashing, the /etc/localtime file doesn't exist. Aug 30 15:27:20 What do I need to do to effectively symlink /usr/share/zoneinfo/Europe/Copenhagen to /etc/localtime on the produced image? Aug 30 15:27:22 Stygia: that's going to create a link pointing at a path on your buildsystem :P Aug 30 15:27:25 an absolute path Aug 30 15:27:36 Random question : Anyone done much system tuning to speed up a full image build? On my machine it's a spritely 63 minutes (using pre-downloaded sources) but was wondering if there were any tricks I'm missing that are worth doing to make speedier? Aug 30 15:27:46 ln -s ${datadir}/zoneinfo/Europe/Copenhagen ${D}${sysconfdir}/localtime Aug 30 15:27:51 I have seen things like _virtualclass-native but I have also seen _native Aug 30 15:27:53 in this case you're making abroken symlink... Aug 30 15:28:17 (thats for an imx61 image) Aug 30 15:28:26 kergoth, Hmm. Why is having ${D}{sysconfdir} fundementally different from what I did? I don't have a broken symlink, I have no symlink. Aug 30 15:28:28 johnsoga: when the recipe is native, 'class-native' is in OVERRIDES, so you can override variables or append/prepend to them Aug 30 15:28:47 Stygia: well, that line will create an existing but broken symlink Aug 30 15:28:50 so you must have something else going on Aug 30 15:29:04 kergoth, Hmm weird. Alright, does it belong in do_install_append, or somewhere else in the process? Aug 30 15:29:05 using sysconfdir and datadir is just to enusre the recipe obeys our path variables, it wonto fundamentally change whats happening Aug 30 15:29:11 the main change is i didn't use ${D} in the destination of the link Aug 30 15:29:14 kergoth, Yea that's exactly what I figured. Aug 30 15:29:15 only the path to the link Aug 30 15:29:24 * nerdboy set a DEFAULT_TIMEZONE variable in his .conf Aug 30 15:29:37 that does sound cleaner Aug 30 15:30:11 nerdboy, So you're saying I can add DEFAULT_TIMEZONE=Europe/Copenhagen to my local.conf and that'd be it? Aug 30 15:30:21 i hacked it in Aug 30 15:30:46 nerdboy, Ah. So it's not a feature in vanilla yocto? Aug 30 15:31:03 nope Aug 30 15:31:17 is it possible to modify source in the workdir and rebuild that without wiping away? Aug 30 15:31:29 i wonder if we ever got an overlay recipe to let you just drop files into a mirror filesystem that'd be captured and included in the image Aug 30 15:31:30 I do not wanna modify the source, generate tarball, ship to yocto, etc Aug 30 15:31:31 the symlink thing *could* go in pkg_postinst Aug 30 15:31:36 to use for configuration Aug 30 15:31:37 it is more steps than just modifying in the workdir Aug 30 15:31:57 nerdboy, Hmm, yea I'm sorta fearing that something removes the symlink after it's created in the do_install_append stage. Aug 30 15:32:12 lpapp: http://www.yoctoproject.org/docs/current/dev-manual/dev-manual.html#modifying-temporary-source-code Aug 30 15:32:14 possible, though unlikely Aug 30 15:32:31 But I'm building a new image with kergoth's version of the ln, I'll check my tmp/ to see it's there. Aug 30 15:32:44 Stygia: we used to symlink that in default gentoo builds but stopped a while back Aug 30 15:33:04 and ntp does do_package and everything now without having been cleaned, so the bbappend definitively catches on. Aug 30 15:33:10 nerdboy, Heh, gentoo uses yocto? Aug 30 15:33:13 switched it to a cp instead Aug 30 15:33:26 no, the symlink method of setting localtime Aug 30 15:33:35 bluelightning: thanks, is there a global way with yocto to force debug builds? Aug 30 15:33:36 nerdboy, Oh? I've always ln'd timezones, that's how I've seen it done in every example and tutorial ever. Aug 30 15:33:37 *method even Aug 30 15:33:39 Stygia: you can always examine the actual ipk contents, and the contents of ${WORKDIR}/packages-split/, to see what really went into the binary package, too Aug 30 15:33:45 bluelightning: pass -g to gcc all the time? Aug 30 15:33:53 or at least for the selected softwares? Aug 30 15:34:01 kergoth, Yea, normally I just do a find tmp/ -name whatever, though. Aug 30 15:34:03 I am getting a segfault, and would like to debug it. Aug 30 15:34:20 ipkcontents () { ar p $1 data.tar.gz | tar -tvz } # prints the files in the ipk Aug 30 15:34:22 yup, sometimes i download and unpack an ipk to make sure what's in it Aug 30 15:34:36 lpapp: set DEBUG_BUILD in the recipe I guess, haven't ever done it myself Aug 30 15:34:49 ipkinfo () { ar p $1 control.tar.gz | tar -zxO ./control } is also handy from time to time, to see the metadata Aug 30 15:35:19 lpapp: usually you can just install gdb and the debugging symbols for the package into your image and that is enough Aug 30 15:35:26 lpapp: I've definitely done the latter Aug 30 15:36:10 kergoth, Ah, hmm. Looks like tzdata sets /etc/localtime, there's a tmp/work/all-poky-linux/tzdata/2013b-r0/packages-split/tzdata/etc/localtime anyway. Aug 30 15:37:23 ahh, that'd explain it Aug 30 15:37:30 apparently i din't need to do too much hacking on that one Aug 30 15:37:59 since this is already in the tzdata recipe => poky/meta/recipes-extended/tzdata/tzdata_2013b.bb:DEFAULT_TIMEZONE ?= "Universal" Aug 30 15:38:16 hah, good to know :) Aug 30 15:38:26 Stygia: try setting DEFAULT_TIMEZONE in your image recipe Aug 30 15:38:26 that should be in an faq somewhere. how do i set my timezone default Aug 30 15:38:44 nerdboy: er that won't work Aug 30 15:38:45 bluelightning: DEBUG_BUILD -> -g3 -O0? Aug 30 15:38:48 nerdboy, Yea, exactly what I did. Aug 30 15:38:56 nerdboy, I made a bbappend to tzdata Aug 30 15:39:09 Where I did DEFAULT_TIMEZONE = 'Europe/Copenhagen' Aug 30 15:39:16 Sorta insulting it acts like "Europe" is a country. But whatever. Aug 30 15:39:21 nerdboy: if the tzdata recipe is expecting to read it, setting it in the image recipe won't change the value read Aug 30 15:40:11 lpapp: setting DEBUG_BUILD=1 will use whatever is in DEBUG_OPTIMIZATION Aug 30 15:40:14 should be able to override it Aug 30 15:40:27 you can, just not from the image recipe Aug 30 15:40:37 at least not in that way Aug 30 15:40:53 maybe my grep missed something... Aug 30 15:41:11 bluelightning, Doing it in a bbappend is logical and sane, though, right? Aug 30 15:41:13 bluelightning: what is the default? Aug 30 15:41:31 A bbappend that only contains "DEFAULT_TIMEZONE = 'Europe/Copenhagen'", then it should be overwrite the value in the main recipe? Aug 30 15:41:43 It uses ?= in the main recipe, btw... whatever that means. Aug 30 15:41:47 you could bbappend tzdata, or just set it in your local.conf or distro Aug 30 15:41:50 ?= means set only if unset Aug 30 15:41:55 yup, stuck it in local.conf Aug 30 15:41:57 kergoth: what does it mean when I see class-native vs virtclass-native? Aug 30 15:41:57 so if the config metadat set its, it wont override it Aug 30 15:42:13 the virt prefix is deprecated, and possibly not even supported anymore, not sure Aug 30 15:42:13 lpapp: bitbake -e | grep ^DEBUG_OPTIMIZATION= Aug 30 15:42:36 Stygia: try setting it in local.conf Aug 30 15:42:38 johnsoga: some natives are bbclassextended, some are real recipes. we needed an override that'd work for both, hence no 'virt', since tghe real recipe isn't "virtual" Aug 30 15:42:57 nerdboy, Hmm if it works in the bbappend I'm quite happy with that. And it seems like it did. But thanks. :) Aug 30 15:43:25 this seems like the kind of thing thats conceptually better in the config metadata, since its configuration, but either will do :) Aug 30 15:43:44 yeah, I finally did it Aug 30 15:43:48 bluelightning: but at least -g3? Aug 30 15:44:07 * nerdboy needs to clean out some cruft Aug 30 15:44:10 kergoth: k thanks, some of this doesn't have good documentation yet so I have been just looking through other bb files and ncurses still uses the virt prefix. Aug 30 15:44:16 bluelightning: DEBUG_OPTIMIZATION="-O -fno-omit-frame-pointer -g -feliminate-unused-debug-types -pipe" Aug 30 15:44:28 lpapp: if you prefer a different value just set DEBUG_OPTIMIZATION to whatever you want Aug 30 15:45:07 bluelightning: I need -O0, not -O1 Aug 30 15:45:08 lpapp: FYI, -g is specified with FULL_OPTIMIZATION (the default) as well, hence how we always have debugging symbols Aug 30 15:45:32 bluelightning: what do you mean by default? Aug 30 15:45:39 by defaul if you specify DEBUG_BUILD=1? Aug 30 15:45:40 lpapp: if you never set DEBUG_BUILD Aug 30 15:46:00 the main difference, afaik, is the "debug build" disables optimizations to avoid problems when debugging, other than that, there's no point to it anymore what with the splitting of debug symbols into -dbg Aug 30 15:46:04 heh Aug 30 15:46:38 bluelightning: so -g3 is the default for any build? Aug 30 15:46:51 lpapp: no, -g is Aug 30 15:46:58 lpapp: see the value of FULL_OPTIMIZATION Aug 30 15:47:57 bluelightning: ok, that is not good enough. Aug 30 15:48:29 feel free to do as suggested above then Aug 30 15:50:57 * nerdboy has another day of hardware shenanigans and system migration Aug 30 15:51:52 is there some way to list files belonging to a particular package? Aug 30 15:52:07 e.g. see a list of files added to the image by including sysvinit Aug 30 15:52:31 what kergoth posted for listing ipk contents Aug 30 15:54:02 buildhistory has a lot of info as well, both about the package contentsa nd image contents Aug 30 15:54:09 if you use it, of couse Aug 30 15:56:02 BCMM: yep buildhistory is useful for finding out that kind of information: http://www.yoctoproject.org/docs/current/ref-manual/ref-manual.html#maintaining-build-output-quality Aug 30 15:58:35 Hmm. /etc/localtime doesn't make it into my image at all. I've set DEFAULT_TIMEZONE, and from what I can read in tzdata's do_install (http://pastebin.com/jUQSYRUQ), it should be installed, and generate an error if it isn't installed. Since it wasn't in the FILES_${PN} array I added it in my bbappend, and although tmp/work/all-poky-linux/tzdata/2013b-r0/packages-split/tzdata/etc/localtime exists, it's not part of the image. Aug 30 15:58:38 What am I missing here? Aug 30 15:59:18 things that make you go "hmm..." Aug 30 15:59:35 It seems like the pkg_postinstall_${PN} thing removes it, but only if it's invalid, and it _should_ at least print it and tell me. Aug 30 16:00:13 And this file is in tmp/ after (tmp/work/all-poky-linux/tzdata/2013b-r0/package/usr/share/zoneinfo/Europe/Copenhagen), so it's not that it doesn't exist Aug 30 16:01:50 Stygia: anything related in log.do_rootfs? Aug 30 16:02:40 bluelightning, Nope, just a single line saying it's executing the shell function do_rootfs Aug 30 16:02:53 er Aug 30 16:03:07 there should be a *lot* more than just that line Aug 30 16:03:13 bluelightning, Well in the log.do_roofts returned by my find build/ -name log.do_rootfs Aug 30 16:03:22 bluelightning, Oh... now there is. Aug 30 16:03:24 bluelightning, Weird. Aug 30 16:03:41 bluelightning, Maybe less was just teasing me, though, and wanting me to manually press down-arrow... I dunno. Aug 30 16:07:31 bluelightning, Nope... nothing that I see that looks relevant. Aug 30 16:08:09 Stygia: is it in the rootfs subdir of the image work directory Aug 30 16:08:10 I should note, I think, that /etc/localtime is only shown as a subfolder to tzdata*/, not as part of the image... Aug 30 16:08:11 ? Aug 30 16:08:18 No. Aug 30 16:08:36 ok, then I can only assume either a postinstall script or something in do_rootfs deleted it Aug 30 16:09:16 bluelightning, Hmm yup, pkg_postinst_${PN} in the main recipe deletes it if the timezone doesn't exist. However, I did a pkg_postinst_${PN}_append to do it manually, with no luck. Aug 30 16:09:49 Stygia: you'd see the output of those echo statements in log.do_rootfs if they were being run Aug 30 16:10:26 bluelightning, And hmm, yea they aren't. Aug 30 16:11:27 bluelightning, It's here, though: tmp/work/armv7a-vfp-neon-poky-linux-gnueabi/ntp/4.2.6p5-r8.0/image/etc/localtime Aug 30 16:11:53 erm Aug 30 16:12:03 I thought it was supposed to be tzdata installing this file Aug 30 16:12:20 it is Aug 30 16:12:35 bluelightning, It is. It shows up as subdirs tzdata, and _then_ that, too. Aug 30 16:12:42 tzdata owns both /etc/localtime and /etc/timezone Aug 30 16:12:52 bluelightning, Oh wait, shit, that lists ntp. That's just a leftover from my earlier recipe. Never mind. Aug 30 16:14:28 But alright, I'm giving up for now, I'll try again tomorrow. Aug 30 16:14:30 Stygia: should look like this on the target: http://paste2.org/zD67tBmc Aug 30 16:14:37 Maybe I'll see the error with more than a few hour sof sleep. Aug 30 16:14:58 root@raspberrypi:~# file /etc/localtime => /etc/localtime: symbolic link to `/usr/share/zoneinfo/PST8PDT' Aug 30 16:15:13 bluelightning: thanks Aug 30 16:15:19 nerdboy, Yea that's exactly what I expected, but it's not part of the image Aug 30 16:15:26 that's my default out of the build Aug 30 16:15:52 from setting it in local.conf Aug 30 16:16:17 anyway, go sleep on it Aug 30 16:16:23 does yocto have a network configuration file? by default, it seems to be just doing DHCP on eth0 Aug 30 16:16:30 /etc/network/interfaces Aug 30 16:16:50 my package feed is public if you want to look later... Aug 30 16:17:12 kergoth: does that work like the one in debian? Aug 30 16:17:19 yes Aug 30 16:17:21 BCMM: connman gets pulled in by some images Aug 30 16:17:22 we use ifupdown Aug 30 16:17:26 our distros are debian based originally Aug 30 16:17:28 nerdboy: good point Aug 30 16:17:36 tends to ignore interfaces Aug 30 16:17:45 indeed Aug 30 16:18:20 kergoth: thanks. what's the correct way to automatically add my own interfaces file? Aug 30 16:18:20 but yeah, without a network "manager" interfaces is it Aug 30 16:19:02 nerdboy: i haven't heard of connman before - basically like network manager with fewer dependancies and bloat? Aug 30 16:19:29 less bloaty but even wonkier Aug 30 16:19:37 if you can believe that... Aug 30 16:20:21 nerdboy: heh. any compelling reason not to just use the interfaces file + wpa_supplicant? Aug 30 16:20:23 you could install_append init-ifupdown if you need to replace the default Aug 30 16:21:00 not at all, that's what i do in the rpi build... Aug 30 16:21:21 works fine that way Aug 30 16:22:11 nerdboy: ah, i'll probably just copy what you've done then. did you do the install_append thing? Aug 30 16:22:18 the only oddity i've seen is the first boot of an image doesn't fire up networking on its own Aug 30 16:22:49 nerdboy: ah, that sounds like a problem for me - i'm doing an initramfs, so every boot is the "first boot" Aug 30 16:22:54 ifup -a one time and then everything is nominal after that Aug 30 16:23:54 i should add that as a work-around to the configure script if its still there... Aug 30 16:24:23 BCMM: netbase includes hte deafult interfaces file, you can bbappend it and alter FILESEXTRAPATHS to get it to use yours in your layer Aug 30 16:25:01 kergoth: thanks, that sounds like a pretty clean way to do it Aug 30 16:25:28 seems like that file is owned by init-ifupdown... Aug 30 16:25:36 huh, guess it got moved Aug 30 16:26:19 looks like there's a beagleboard append in bsp Aug 30 16:26:34 meta-yocto-bsp/recipes-core/init-ifupdown/init-ifupdown-1.0/beagleboard/interfaces Aug 30 16:27:38 historically its been in netbase, which is why so many layers broke when it was updated from 5.0 to 6.0 Aug 30 16:27:46 not that it really matters :) Aug 30 16:28:07 * kergoth sighs and waits on another qt5 build Aug 30 16:28:08 nerdboy: last time i looked at your repo i didn't really understand how recipes work - why do you have your own mpd recipe? Aug 30 16:28:26 hey, i can't be the only one getting corrected around here... Aug 30 16:28:34 hehe Aug 30 16:29:11 it's an append to add some config/packaging stuff Aug 30 16:29:58 mostly it adds my own init script Aug 30 16:30:11 nerdboy: you have a .bb and a .bbappend - what's the .bb do? Aug 30 16:30:25 also sets the volatile config and some perms Aug 30 16:30:41 no .bb file for mpd Aug 30 16:30:55 what are you looking at? Aug 30 16:31:02 nerdboy: uh, the bb for mpc... sorry Aug 30 16:31:15 that one didn't exist Aug 30 16:31:26 yeah, i just misread a letter Aug 30 16:32:04 nerdboy: also i noticed the stock recipe has no init script - i guess i'll use yours Aug 30 16:32:25 i'm not using systemd Aug 30 16:32:39 just plain old sysvinit Aug 30 16:33:31 same, i think - i meant the mpd recipe doesn't seem to install any kind of init script Aug 30 16:33:56 which was the initial reason for my bbappend... Aug 30 16:34:00 oh i see, you were warning me in case i needed whatever systemd uses instead Aug 30 16:37:39 systemd is a distro feature Aug 30 16:40:16 nerdboy: i haven't got it enabled - i'm using sysvinit Aug 30 16:40:48 some packages may not have a fallback init script Aug 30 16:42:05 is there still a one-time configure script if you enable ROOTFS_READ_ONLY ? Aug 30 16:43:11 oh i see, stock MPD recipe has systemd configuration, but no sysvinit script Aug 30 16:43:20 nerdboy: sorry, i don't understand teh question Aug 30 16:43:41 sorry, not for you specifically Aug 30 16:44:17 oh ok. there's so much stuff i don't know yet about yocto that i couldn't tell whether it pertained to what i was doing Aug 30 16:44:57 nerdboy: the read o nly rootfs support has changed substantially recently, though only for sysvinit. see the read-only-rootfs image feature Aug 30 16:45:32 config files control on-boot bind-mounting of volatile paths over r/o locations, with automatic copying of the content in the r/o path into the volatile path Aug 30 16:45:36 interesting stuff. Aug 30 16:46:17 so, kinda depends on how much tmpfs you can spare Aug 30 16:47:17 well, in theory you should be able to use non-tmpfs "volatile" path, e.g. a writable partition Aug 30 16:47:20 need to try doing that one of these days Aug 30 16:49:07 get out of my head... Aug 30 17:03:55 khem: are you around — I hear you might be able to help with a problem I'm having building busybox for mips with DEFAULTTUNE = "mips64-n32" Aug 30 17:12:01 how can i modify the behaviour of a .inc from another layer? do bbappends work Aug 30 17:12:16 (sorry, spurious newline) or is there some sort of .incappend? Aug 30 17:12:39 nope Aug 30 17:12:56 bbappend should do it, as long as you append the recipe that includes the .inc Aug 30 17:17:19 nerdboy: so bbappend the recipe that "require"s it, as if the inc was part of that file? Aug 30 17:20:37 heh, i see the pr server threading improvements didn't fix the can't get PRAUTO failures Aug 30 17:21:15 yup Aug 30 17:21:54 kergoth: RP is still working on it Aug 30 17:22:44 * kergoth nods Aug 30 17:36:13 morning kergoth Aug 30 17:40:31 damn, the ruby recipe has a number of problems Aug 30 17:40:33 mornign mranostay Aug 30 17:42:37 kergoth: how are the bits baking today? Aug 30 17:42:56 meh Aug 30 17:48:36 anyone know how to print/query ruby's default library search path, ala python's sys.path? Aug 30 17:55:23 aha, there we are. ruby -e 'puts $:' Aug 30 18:02:02 hi galak Aug 30 18:02:16 mranostay: hi, how's the new job? Aug 30 18:02:41 haven't quit yet. so that has to say something :P Aug 30 18:02:50 galak: going to ELCE? Aug 30 18:02:59 I will be Aug 30 18:03:11 reminds I need to register (got travel and hotel room setup) Aug 30 18:03:14 cool Aug 30 18:04:16 the other hotel block closes today Aug 30 18:04:51 yeah, I got into The Caledonian Aug 30 18:05:39 I assume you'll be there based on asking the question Aug 30 18:05:51 yep Aug 30 18:09:52 galak: know if beckyb is going as well? Aug 30 18:14:06 * zeddii still has no approval for elce, if I do go, I'll probably have to sleep on the street Aug 30 18:16:19 zeddii: Wind River? Aug 30 18:17:22 * zeddii nods Aug 30 18:24:04 aha, confirmed that ruby-native suffers from a relocation bug Aug 30 18:31:50 zeddii: no one wants to see your presentation ;-) Aug 30 18:31:59 zeddii: they just want you to drink beer and entertain Aug 30 18:33:56 mranostay: becky's is doing ok. Aug 30 18:34:04 mranostay: as of wed lunch time Aug 30 18:35:04 msm. pretty much, you must have put the word in on me. I keep getting rejected! Aug 30 18:35:17 * zeddii would go to just drink beer though Aug 30 18:41:43 im sure you would Aug 30 18:41:46 i would too ;) Aug 30 18:45:39 labour day weekend is going to involved fixing my demo :P Aug 30 18:46:57 How much does one really learn at ELCE? Aug 30 18:47:14 is there a way to tell bitbake not to run the do_kernel_checkout task? my kernel source is in a tarball, so git checkout is unnecessary. Aug 30 18:47:59 sounds like you should look at the linux-yocto-custom recipe, or don't use linux-yocto at all Aug 30 18:48:07 cfo215: ssssssh you'll ruin it for us :) Aug 30 18:48:40 mranostay, oh, so sorry... I'll keep a lid on it. Aug 30 18:49:04 kergoth: was basing my recipe on linux-yocto-custom. do I have to use linux-yocto? Aug 30 18:49:04 cfo215: you learn some but more networking Aug 30 18:50:10 too bad I don't have a corporate sponsor. I was seriously considering it till I looked at airline tickets... Aug 30 18:51:30 I'm saving that cash for my trip to the Italian wine country... Aug 30 18:54:01 kergoth: essentially what I have in my SRC_URI is the tarball, and a poop-ton of patches from the board vendor. I can always make a git repo if that's the easiest way to get around it, but doesn't make sense to me to have to go that route Aug 30 18:56:17 wmcdevel: what problem is the checkout causing ? Aug 30 18:56:27 -custom should have just worked, with no fuss. Aug 30 18:57:01 * zeddii looks back in history Aug 30 18:57:55 zeddii: Function failed: do_kernel_checkout .... tells me to look at the log file. Essentially, it's trying to do a git clone, but there is no git repo to clone from Aug 30 18:58:17 tgz is supported, what oe-core branch are you using ? Aug 30 18:58:57 zeddii: er ... not really sure. working with poky dylan Aug 30 18:59:18 let me check. I always mess up the code names to version numbers. Aug 30 18:59:25 * zeddii wonders if that makes him evil Aug 30 19:01:21 if it is dylan, and it isn't working then something broke. tgz should work just fine, or any type of archive that isn't a git repo. Aug 30 19:01:52 do you have a reproducer recipe ?I don't need your patches, just the location. I'll cook up my own as well. Aug 30 19:02:28 um, reproducer recipe? Aug 30 19:02:33 * mranostay hunts for a printers Aug 30 19:02:40 sorry, <-- noob Aug 30 19:03:01 wmcdevel. the recipe that you are trying to build that is based on linux-yocto-custom. I'd like to see it break here, so I can debug. Aug 30 19:03:13 ahh .. gotcha Aug 30 19:03:43 just a sec ... let me see if I can pastebin it Aug 30 19:04:08 perfect. Aug 30 19:06:54 http://pastebin.com/pEsbqKYf Aug 30 19:07:56 zeddii: and just for reference sake, building on Ubuntu 12.04 x86_64 Aug 30 19:09:03 zeddii: recipe name linux-friendlyarm_2.6.38.8.bb, files stored in linux-friendlyarm-2.6.38.8 Aug 30 19:10:00 np. firing up a build. Aug 30 19:10:49 kk. gonna reload the caffeine feed then ... brb Aug 30 19:20:28 zeddii: any luck? Aug 30 19:22:11 hah. fat fingered it, was just correcting my recipe. I had started a 3.10 build with a xz. Aug 30 19:23:13 :) well, at least it isn't the compression scheme that's the problem. Aug 30 19:23:32 definitely not. Aug 30 19:24:07 * zeddii waits Aug 30 19:24:24 I really like the build system, but there is so much documentation, it's almost as bad as no documentation. wish there were more concrete snippets for doing certain tasks Aug 30 19:25:37 that's something which is continually improving. meta-skelton, etc. and labs Aug 30 19:25:43 bugger. I forgot to put the checksum Aug 30 19:36:58 i'm trying to modify a recipe with a bbappend, and want to add a custom file. how can i modify the path that the recipe looks in for file:// urls? Aug 30 19:37:35 FILESEXTRAPATHS_prepend := Aug 30 19:37:42 (or whatever variant you prefer) Aug 30 19:38:02 zeddii: thanks Aug 30 19:38:23 zeddii: i guess the _prepend means i can use my layer to override existing files? Aug 30 19:38:42 it means that whatever directories you add would just be first in the search path. Aug 30 19:39:24 with the obvious addition that FILESEXTRAPATHS is already in addition to the system / built in search dirs Aug 30 19:39:58 wmcdevel, my test on master worked. I'm checking out dylan to see if it was broken then. Aug 30 19:40:45 zeddii: ok Aug 30 19:41:23 give me a few mins, fetching linux-2.6.38.8.tar.xz Aug 30 19:42:11 zeddii: uh, what is FILESEXTRAPATHS_prepend := relative to? Aug 30 19:42:23 ok. not sure if that was on kernel.org or if I got it from the vendor's DVD Aug 30 19:42:41 wmcdevel, I found it, no worries. Aug 30 19:42:47 cool Aug 30 19:42:59 as an example: FILESEXTRAPATHS_prepend := "${THISDIR}/${PN}:" Aug 30 19:43:12 ${THISDIR} is the directory of the recipe Aug 30 19:43:18 PN is the recipe name (typically). Aug 30 19:43:33 so if it is relative to the recipe, anchor it with ${THISDIR} Aug 30 19:43:51 zeddii: if i'm in a bbappend, does THISDIR refer to the append or the original recipe? Aug 30 19:43:57 bbppend Aug 30 19:44:29 thanks, that should solve my issue Aug 30 19:45:29 THISDIR is the dir of the currently parsed file. if you forget to use :=, it'll be the recipe directory rather than the bbappend directory, since it forces it to get the *current* value then and there, rather than expanding lazily as usual Aug 30 19:46:19 * zeddii nods. kergoth has all the details Aug 30 19:46:33 I didn't point out the significance of := in the example. Aug 30 19:46:53 * zeddii is the laziest checksum updater around. fetch. fail. cut and paste. Aug 30 19:52:58 is there a way to clean the build directory of a given recipe without affecting anything else? Aug 30 19:53:06 zeddii: hey, lazy can be a good thing :) Aug 30 19:53:11 bitbake -c clean recipe Aug 30 19:53:16 thanks Aug 30 19:53:18 well, taht *removes* the build dir, the entire workdir Aug 30 19:53:23 so depend son what you want Aug 30 19:54:17 just want to remove the build directory for my kernel Aug 30 19:55:14 uh, really stupid question: what language are .bb recipes in? Aug 30 19:59:33 "bitbake", python, shell (depending on the function you are looking at). Aug 30 20:00:31 ah, i'd sort of assumed it was all a shell script Aug 30 20:01:24 a lot is. Aug 30 20:01:37 depends on who writes a recipe I suppose Aug 30 20:01:43 wmcdevel, http://pastebin.com/RWb40e1d Aug 30 20:01:48 .bb isn't a shell sript, but functions/tasks defined in it can be Aug 30 20:02:12 that recipe worked for me. I think you are missing some required elements that led to an error that looked like a git clone/fetch Aug 30 20:02:29 wmcdevel, I'll add some extra checks. Aug 30 20:03:00 zeddii: cool ... trying now Aug 30 20:03:18 you can convert it back to your local file:// variant .. but I fetched right from korg. Aug 30 20:04:22 ok. will I need an md5/sha256 checksum for all my patches as well? Aug 30 20:05:03 probably not if you switch to file://, I'm not a fetcher wizard. Aug 30 20:05:25 wmcdevel, the big missing line was S="${WORKDIR}/linux-2.6.38.8" Aug 30 20:05:27 lol ... ok Aug 30 20:05:51 assuming that vendor tgz hasn't done something crazy with the directory names Aug 30 20:05:59 * zeddii mutters about shooting them if they did Aug 30 20:06:03 ;) Aug 30 20:06:42 ha ... I'm right with ya on that one. don't even get me going on "vendor support" ... *grumble grumble grumble* Aug 30 20:09:27 ok, well it didn't bork right away on the do_kernel_checkout, so let's see if it completes the build this time Aug 30 20:14:04 still patching ... will have to combine a number of them to not have patch called 150+ times Aug 30 20:20:03 * zeddii heads out for the weekend. Aug 30 20:20:20 zeddii: thanks for the help :) Aug 30 20:20:40 no problem. I've added a check to give a decent error message for the case you hit. Aug 30 20:21:15 excellent Aug 30 20:25:53 Hmm, I'm a little concerned about _remove's semantics. If a low priority bbappend removes something, and then a higher priority bbappend appends something, the lower priority removal will win, as far as I can tell from the implementation Aug 30 20:26:06 i grabbed a fix for the kernel configuration in linux-yocto by specifying a SRC_REV_meta... it grabs the fix and a few other changes, but now the kernel configuration says 3.8.11 whereas it was 3.8.4 (and the kernel remains 3.8.4). anyone foresee problems? Aug 30 20:28:54 galak: btw that block just sold out Aug 30 20:29:08 glad I booked it when I did Aug 30 20:32:15 still patching ... "validating against known patches (icp-arm-standard-meta)" coming up on about 20 minutes, and only 27% along. is that normal? what known patches would it be validating against? Aug 30 20:35:58 how would i go about making a PACKAGECONFIG key control more than one ./configure option? Aug 30 20:36:22 BCMM: there's nothing anywhere in packageconfig that says you can only use one option Aug 30 20:36:29 it just uses strings, it doesnt' care waht they contain Aug 30 20:36:56 oh right, just strings that are concatenated to teh command? Aug 30 20:37:04 PACKAGECONFIG[foo] = "--enable-something --enable-something-else,--disable-something --disable-something-else,somedep," Aug 30 20:37:10 there are multiple examples of this in oe-core Aug 30 20:37:12 yes Aug 30 20:37:15 thanks Aug 30 21:33:44 halstead: time to update the bugzilla with 1.5.1, 1.5.2 and 1.6 M1-M5 Aug 30 21:35:37 sgw_, Shall do. Aug 30 21:37:43 sgw_, Do we need a 1.7? Aug 30 21:41:13 halstead: sure Aug 30 21:41:26 kergoth: "database table is locked"? Aug 30 21:46:30 sgw_, I think it's all set. 1.5.1 was already in there so my script failed. But I fixed it by hand. Aug 30 21:47:44 halstead: sorry I should have checked those first. Aug 30 21:48:00 sgw_, I should have too. Aug 30 21:49:30 sgw_, Let me know if you notice any problems. Aug 30 21:49:48 I'm heading into to Portland. I'll be back online in a bit. Aug 30 21:51:07 halstead: later Aug 30 21:55:40 * kergoth wonders why the omap5 evm bits in the meta-ti-glsdk repo arent' in the main meta-ati repo Aug 30 22:08:26 I just tried building current poky master branch and it failed on the subversion package...looks like version 1.7.10 is no longer available from the mirrors. Can anyone point me to what needs to be changed, I'm brand new at yocto. Aug 30 22:20:42 hmm, have a lovely reproducer for PR server issues. Add in logging and it stops reproducing Aug 30 22:47:47 cfriesen: haven't tried anything with the master branch, but if you check out the dylan release/tag, that does work Aug 30 22:51:21 RP: damned heisenbugs Aug 30 22:53:22 just in case zeddii reads through the logs later, a major thank you for the help today ... kernel recipe builds now without error. w00t! Aug 30 22:53:33 have a great weekend all Aug 30 22:58:03 I've figured out part of the problem. In my tests, my PR server process is using 4GB of VM :/ Aug 30 22:58:38 well, the part I figured out is the sqlite retry upon locked code is broken (the count to 500) Aug 30 22:59:11 The 4GB of VM usage is due to forking the server off bitbake. I suspect it would be nicer to exec() it and be done Aug 30 22:59:45 (we end up with a server waiting in memory just in case the function exits which it can't) Aug 30 23:01:45 My test case is 500 processes (48 at any one time) each making 5 calls to the PR server. Longest latency is 12 seconds Aug 30 23:04:17 wouldn't the vast majority of that 4gb be shared, since its COW anyway? Aug 30 23:04:25 have you checked smem rather than ps/top? Aug 30 23:04:32 its proportional set size is handy Aug 30 23:05:31 kergoth: the vast majority should be but its still a crazy size for the PR server considering what it does Aug 30 23:05:47 kergoth: Tried the standalone one and it also hits crazy sizes... Aug 30 23:07:36 It sounds quite unlikely that it'd dirty that much of the memory unless it's doing something crazy Aug 30 23:07:48 * kergoth shrugs Aug 30 23:07:57 python memory profiling time? :) Aug 30 23:09:40 kergoth: yes, time to get the profiler and look at this I think. Something just feels wrong... Aug 30 23:10:10 * kergoth nods Aug 30 23:10:56 i found it helpful to inject an interactive python debugger session and inspect with meliae, though that'd be of limited use in the server without listening on a socket Aug 30 23:13:29 kergoth: I probably can just send it a dump command... Aug 30 23:13:45 https://pypi.python.org/pypi/rpdb/ looks promising Aug 30 23:14:00 * kergoth nods Aug 30 23:14:44 kergoth: it does look good, yes Aug 30 23:15:03 kergoth: and I have the code to launch an interactive python shell in task context too Aug 30 23:15:25 nice Aug 30 23:15:47 anyhow, Zzzz for now... Aug 30 23:16:04 night Aug 30 23:16:31 * kergoth wonders why he never tried to get https://github.com/kergoth/bitbake/commit/27c5959 merged Aug 31 01:04:51 BCMM: fixed the network issue **** ENDING LOGGING AT Sat Aug 31 02:59:58 2013