**** BEGIN LOGGING AT Tue Aug 16 02:59:58 2016 Aug 16 03:49:41 I am struggling chasing errors like do_image_tar: Taskhash mismatch e46ba58e3a9cdc353a94548fd66fb109 verses 88d394eba0f2c542200ed8b0ad4b867a Aug 16 03:49:55 this started to happen with krogoth Aug 16 03:50:15 I see it with rpi images and also some internal images Aug 16 03:50:42 when I run bitbake-diffsigs it says basehash changed from 9bbc36cad4c9ade1a346c4b6bb6c6f75 to 138f22157ec470f60a2071db0f68a4e4 Aug 16 07:32:13 khem: common problem for me as well in various do_$tasks Aug 16 07:32:48 khem: meta-raspberrypi have pushed a *fix* for it, but I'm still seeing it occur for other tasks, such as populate sdk Aug 16 07:33:18 I'm not sure what the solution is other than forcefully removing the check routine Aug 16 08:03:35 khem, CTtpollard: stop using variables that change in between the cooker doing the calculations and the worker starting - often DATETIME and so on. if those are being used, whitelist them. Aug 16 08:04:43 yeh it seems to be rogue DATETIMES in various classes Aug 16 08:25:52 Hello. I am using the DART-6UL board from Variscite. I have problems related to using Qt5 with linuxfb plugin on a Yocto build. Are there people here working with a similar setup here? I have tried writing on the qt channel, but as far as I understand few people there have anything to do with embedded development. What would be the most appropriate place to get answers on Qt5 + Yocto related questions? Aug 16 08:33:41 Hi all. Wasn't there an option you could set in your local.conf to only keep the last image that was built? Aug 16 08:33:51 I do remember something like that but i can't find it in the docs anymore Aug 16 08:34:03 Anticom: there is rm_work, but thats a little bit dofferent Aug 16 08:34:34 eduardas_m, instead of asking why doesn't A work with B, do you know how A and B don't work together? Aug 16 08:35:04 LetoThe2nd: But there was no option to delete old images? Aug 16 08:35:30 Anticom: none that i knew of. but thats not an authoritative statement. Aug 16 08:35:59 Is rburton here? Aug 16 08:37:02 RM_OLD_IMAGE Aug 16 08:37:02 http://www.yoctoproject.org/docs/2.1/ref-manual/ref-manual.html#var-RM_OLD_IMAGE Aug 16 08:37:49 * LetoThe2nd hands rburton the FLS hat. Aug 16 08:38:27 joshuagl: thanks! Aug 16 08:40:15 np Anticom Aug 16 08:44:26 redengin: The point is I do not know what the proper way to rotate a Qt5 linuxfb application by 90 degrees on screen is and run into problems. My problem is elaborated here: http://forum.qt.io/topic/70184/weird-qcombobox-behaviour-under-qt-5-5-1-linuxfb-platform-yocto-build-for-imx6ul/2 Aug 16 08:45:15 I am not even sure whether this should be approached from the Qt framework side or from the framebuffer driver side. Aug 16 08:47:33 I am actually new to embedded linux so I wouldn't be surprised I am doing this completely wrong. Aug 16 08:47:48 isn't framebuffer kind of deprecated anyways? Aug 16 08:48:10 imx should have a decent usable drm, maybe even opengl implementation. Aug 16 08:48:28 it sounds like a Qt bug based on expectations of Ubuntu builds Aug 16 08:48:35 imx6ul does not have a graphics accelerator, OpenGL is not usable Aug 16 08:48:45 eduardas_m, have you tried other distros? Aug 16 08:49:34 only Yocto and Ubuntu 14.04 (works properly with the latter) Aug 16 08:49:53 am investigating further Aug 16 08:50:32 eduardas_m, I'd try it out on other distros to determine if its just Ubuntu niceness that makes it work Aug 16 08:51:50 redengin: OOI, What has Ubuntu (or distros for that matter) to do with Variscites Qt5 yocto build? Aug 16 08:52:29 We use Qt5 on imx, but we do use full fsl opengl thou Aug 16 08:53:11 sveinse, all the plumbing back to decide where/how to place things within the display Aug 16 08:57:24 sveinse: what is the method that you would use to rotate a Qt5 widget application by 90 degrees then? perhaps using QGraphicsScene is actually the problem? Aug 16 09:00:24 sveinse: do you use X window system? Aug 16 09:05:05 eduardas_m: No, we have a full-screen linuxfb app Aug 16 09:07:44 sveinse: thank you for sharing information, but could you tell anything about rotation? how would you go about implementing a portrait app in Qt5 under linuxfb? Aug 16 09:19:25 is it still the process for patches for /classes to be sent to the oe-core ml? Aug 16 09:22:02 eduardas_m: Not sure, unfortunately. But Qt honors some environment variables, such as QT_QPA_EVDEV_TOUCHSCREEN_PARAMETERS=rotate=90 and QT_QPA_PLATFORM=eglfs Aug 16 09:22:33 and do all patches require a relevant bug-id? Aug 16 09:24:09 CTtpollard, my experience is that you need to both suggest and offer to maintain Aug 16 09:40:08 eduardas_m: I believe newer Qt5 releases will have a soft-opengl renderer, released under GPL3 Aug 16 09:43:16 https://blog.qt.io/blog/2015/01/22/introducing-the-qt-quick-2d-renderer/ Aug 16 09:46:28 If I set FOO="1" in a.bb and inherit a, then it can be accessed with ${FOO} from b.bb, right. How do I override FOO from local.conf then? Aug 16 09:49:09 sveinse: you mean inheriting a class a.bbclass from a recipe b.bb, where a.bbclass sets FOO = "1"? Aug 16 09:49:56 if you're meant to be able to set the variable in a configuration file, it will usually be set with ?= or ??= Aug 16 09:50:08 in the bbclass that is Aug 16 09:50:47 Ulfalizer: Yes, my bad. a.bbclass in my example above, yes Aug 16 09:50:52 that will leave the variable alone if it already has a value Aug 16 09:52:23 local.conf is read first when parsing? Aug 16 09:52:59 sveinse: yep... the note i added to http://www.yoctoproject.org/docs/2.2/ref-manual/ref-manual.html#usingpoky-debugging-viewing-variable-values might clarify things Aug 16 09:53:52 first the configuration (.conf) files are parsed, and then your particular recipe (along with any bbclasses it inherits) Aug 16 09:54:15 any variables set in the configuration files will be visible by the time recipes are parsed Aug 16 09:55:25 Ulfalizer: Ah, great. The NOTE block in section 2.3.2 which you just referred to answered another of my questions Aug 16 09:56:09 neat :) Aug 16 09:56:33 The docs is great, but as a n00b you tend to get lost in all the information -- even if you have read it once. :D Aug 16 09:57:52 that note was added within the last week i think :) Aug 16 09:58:37 i've added quite a lot of stuff to the 2.2. manuals Aug 16 09:58:44 how do you determine ?= vs ??=. Am I right to assume: Use ?= and you'll know it when you need ??= ? Aug 16 09:59:59 I see PACKAGECONFIG is usually defined with ??= Aug 16 10:00:10 sveinse: ??= is when you don't care whether another assignment comes before or after the ??=. that other assignment will win either way. Aug 16 10:01:42 FOO ??= "foo" means "at the end of parsing, check if FOO was assigned a value. if not, assign 'foo' to FOO." Aug 16 10:02:06 it's a "last resort" default in a way Aug 16 10:05:06 To summarize: I put FOO ??= "defaults" into a.bbclass. From b.bb I "inherit a" and can set FOO ?= "something else". I can then also do FOO_pn-b="mylocal", right? Not FOO_pn-a? Aug 16 10:07:23 a ?= will always "win" over a ??=. if you have a FOO ?= "foo", then either that will assign "foo" to FOO, or FOO already had a value. either way, FOO will have a value at the end of parsing, and so the ??= value will never get assigned. Aug 16 10:08:28 and yeah, FOO_pn-b would be the override that applies for b.bb. it's set based on the name of the recipe, not based on the name of the bbclass. Aug 16 10:08:47 Ulfalizer: Great. Understood! Aug 16 10:08:52 you could look up the OVERRIDES variable to see the underlying mechanism behind overrides btw. i think it helps to understand how it works. Aug 16 10:09:01 the rest becomes mostly "guessable" then :) Aug 16 10:09:23 np Aug 16 10:09:27 food time Aug 16 10:09:38 Actually I did read the OVERRIDES section, and I found the docs to very confusing Aug 16 10:09:46 Let me see if I can find it and explain Aug 16 10:14:05 section 3.2.1 in BB manual under "selecting a variable". The example lists OVERRIDES = "architecture:os:machine". I don't understand the significance of that list. At first I might think that these are the TEST_* variable to chose from, but then the next example with KBRANCH= does not show OVERRIDES being set. Aug 16 10:14:33 I've read this many times, and I just can't get my head around this Aug 16 10:15:34 So either I'm slow, or perhaps this section needs some better examples Aug 16 10:20:08 sveinse: Ulfalizer: datastore-copies: cool note, I inferred this myself after some painstakingly long debug sessions Aug 16 10:21:57 hi. is it possible to run a task with bitbake, without running the tasks it depends on? Aug 16 10:22:21 i want to run deploy task, but i don't want to run the compile task, even if code has changed in the recipe Aug 16 10:23:40 this is to build a device tree without recompiling the entire kernel. so i change the dts in my recipe, run "bitbake -c unpack " and then "bitbake -c deploy ". but the last command also triggers a complete rebuild which is not necessary and takes a lot of time. Aug 16 10:31:20 ionte: bitbake -c task -f Aug 16 10:31:57 sveinse: the way it works is that FOO_someoverride will only be applied of "someoverride" appears in OVERRIDES Aug 16 10:33:10 so the current "configuration" (which recipe? which machine? etc.) is written to OVERRIDES, and that'll make those overrides apply Aug 16 10:33:31 and yeah, the manual sometimes isn't as explicit as it could be Aug 16 10:35:39 sometimes it makes things seem fancier than they really are :P Aug 16 10:36:27 Ulfalizer: So there is something like OVERRIDES="${MACHINE}" behind the scenes and then the example of KBRANCH_qemux86 will be selected? Aug 16 10:37:53 sveinse: yep Aug 16 10:38:14 bitbake -e will show you the value of OVERRIDES etc Aug 16 10:38:36 it adds some indirection with a variable called MACHINEOVERRIDES though, that's added ot OVERRIDES Aug 16 10:38:45 but that's just to confuse you ;) Aug 16 10:39:05 also, makes it easier to customize Aug 16 10:39:30 poky/meta/conf/bitbake.conf:OVERRIDES = "${TARGET_OS}:${TRANSLATED_TARGET_ARCH}:build-${BUILD_OS}:pn-${PN}:${MACHINEOVERRIDES}:${DISTROOVERRIDES}:${CLASSOVERRIDE}:forcevariable" Aug 16 10:40:09 Ulfalizer: brilliant. :D (and it took four sentences to understand what I could not read from the text). IM-very-HO you should add that to the KBRANCH example. That OVERRIDES is set with ${MACHINES} and that is what makes it select the alternative overrides. Aug 16 10:41:12 sveinse: i'm working on some other documentation atm. the documentation maintainer is very open to contributions though: https://bugzilla.yoctoproject.org/enter_bug.cgi?classification=Documentation :) Aug 16 10:41:43 Ulfalizer: all right, thanks Aug 16 10:43:01 rburton, I've been running a day or so with sstate cache on NTFS (via the fuse driver). It seems to be working perfectly. Perhaps thought you wanted to know Aug 16 10:43:37 sveinse: Neanderthal Technology Fireplace System? Aug 16 10:43:45 LetoThe2nd: Yes Aug 16 10:44:07 sveinse: i hear mammoth steak work nicely along that, too! Aug 16 10:44:11 sveinse: presumably because its a local disk and as you said, the file system is fine with the names, its windows that isn't. sstate over SMB on a NTFS file system is likely not going to work. Aug 16 10:44:26 (SCNR) Aug 16 10:44:30 No actually, NTFS in itself isn't so bad. It's windows front end which is Aug 16 10:45:13 The reason I have NTFS btw, is because I share this drive with Win machines. And unfortunately (and it's plain stupid) there are no other modern FS which share well with windows :( Aug 16 10:45:35 sveinse: i know, have ntfs extensively too. usually little hassles. but this is just stuck in my mind: http://www.onefunsite.com/images/nt.gif Aug 16 10:46:49 :D LOL. Well, I'm representing a very little Linux bubble in our company. The rest is Windows-centric, so I'm very alone at times. Aug 16 10:47:35 I'm looking forward to my presentation to the SW director that you need at least 50GB and native Linux machine to do anything useful in Yocto Aug 16 10:47:45 I wonder how i can sugar that pill :P Aug 16 10:48:47 sveinse: tell him you need a 32way build box with 128G ram Aug 16 10:49:09 sveinse: then let him negotiate you down to your actual needs. Aug 16 10:49:28 and if he just says 'yes', you're set anyways. Aug 16 10:49:50 jubr: nope, that forces the task to be run, which in turn may run other tasks. in my case, forcing "deploy" after an "unpack" also runs "compile" etc Aug 16 10:51:30 What I'm planning on is having a large build server running linux, and let it populate the sstate cache. Most devs will do app development, so they'll work on one recipe only. If they then can utilize the sstate cache, they can probably get away with running the yocto build virtually. Aug 16 10:52:02 sveinse: to be fair the disk requirements of yocto basically mean spending £50 on a hard disk Aug 16 10:52:07 sveinse: hm, sounds very much like you actually want to handle them the extensible sdk Aug 16 10:52:23 sveinse: spinning rust is far more economical and the performance gain of ssd isn't as impressive as you'd like, given sufficient ram Aug 16 10:52:27 sveinse: which basically is a shrinkwrapped sstate for app developers to work on one recipe :) Aug 16 10:52:41 rburton: ++ Aug 16 10:52:54 given enough ram, all ssds for OE stuff are moot. Aug 16 10:53:00 speaking of which i really need to raid my 2tb disks Aug 16 10:53:34 LetoThe2nd: We do SDK today. We have a product on the marked running Yocto, developed by a subcontractor. They provides us with a SDK and a core-image. We then build our app and integrate that into the core-image. Aug 16 10:54:41 sveinse: mind, i explicitly said "extensible sdk", not just "sdk" Aug 16 10:55:10 sveinse: its a rather new facility that targets the app developer case amongst some others. Aug 16 10:55:17 rburton: all yocto does in the non-sstate case to check whether a task needs to be rerun is to look for a stamp file for the task with a matching input checksum, right? Aug 16 10:55:19 LetoThe2nd: The problem is that the glue to build our apps and integrate into the core-image now is a proprietary build system built by me. I'm now moving towards making recipes of everything so that we can use bb for what its good at Aug 16 10:55:43 sveinse: see: extensible sdk, really. Aug 16 10:55:56 LetoThe2nd: I will, thanks Aug 16 10:56:00 i think one thing that makes the sstate cache harder to understand at the moment is that the non-sstate case isn't explained Aug 16 10:56:13 Ulfalizer: effectively yeah. "has this been done" Aug 16 10:56:47 ok, thanks. i'm thinking of adding a section on stamps. Aug 16 10:56:54 rburton: i've conducted a few trivial tests with building on spinning rust vs. ssd vs. tmpfs. difference: neglectible, if you have enough ram that caches can take care of everything. Aug 16 10:57:09 OOI, why is the sstate cache sorting on hash-id, rather than filename? Aug 16 10:57:34 sveinse: you mean for the two-character subdirectories? Aug 16 10:57:39 How much RAM are we talking about for decent spin-drive performance with a yocto build? Aug 16 10:57:46 probably because it gives an even spread in that case Aug 16 10:58:07 sveinse: 16gb to start with. my current machine has 64gb... Aug 16 10:58:24 Ulfalizer: Yes. Ah yes, make sense. Otherwise its like the debian repo syndrome. "lib/" is overfull :P Aug 16 10:58:25 (old machine had 16gb, and after a build meminfo said that ~12gb was disk cache) Aug 16 10:58:42 sveinse: do a clean build with empty sstate, du -hs after finish, add 20% :-) Aug 16 10:58:49 sveinse: some characters are more common than others too :) Aug 16 10:59:24 sveinse, Ulfalizer: also performance: file systems don't like having a single folder with 50k files in Aug 16 10:59:59 yeah, that's the original reason for having those two-character subdirectories at all i guess Aug 16 11:00:12 otherwise everything could have been in a single directory Aug 16 11:00:15 $ find /data/poky-master/sstate/ -type f|wc -l Aug 16 11:00:15 518759 Aug 16 11:00:27 i was out by an order of magnitude :) Aug 16 11:00:28 LetoThe2nd: du + 20% = RAM size, that what you meant? Aug 16 11:00:35 sveinse: yeah. Aug 16 11:00:49 that might be *a bit* over kill :) Aug 16 11:01:06 rburton: that 128g suggestion above was not a joke Aug 16 11:01:38 LetoThe2nd: That assumes extfs, doesn't it? Aug 16 11:01:49 yeah start with 128 and negotiate down to just 64 Aug 16 11:02:01 Different fs have different caching schemes Aug 16 11:02:23 the point is, building a core-image-minimal or such is usually not a problem with less ressources. but once you pull in qt or even chromium, stuff gets nasty when you're low on ram Aug 16 11:02:59 if you're not building qt/webkit/etc then you'll be "fine" with 16gb, but get as much as you can justify Aug 16 11:03:06 my sstate cache is at 43Gb... But I'm working on two flavors of arm, intel and qemu against the same sstate cache Aug 16 11:03:24 pah Aug 16 11:03:26 rburton: thats a good bottom line. Aug 16 11:03:34 568G /data/poky-master/sstate/ Aug 16 11:03:59 * rburton waits for kergoth to join in Aug 16 11:04:00 whoah! ok. I'm standing in line all the way back :P Aug 16 11:04:05 much more important than 16gb ram vs. 32gb or so is, having a native environment. Aug 16 11:04:28 or at least some kind of docker/etc. container. Aug 16 11:05:02 (On request from my manager) I have actually on my list to benchmark the actual penalty from doing this virtual (against a phy disk) Aug 16 11:05:16 IT HURTS SO BAD Aug 16 11:05:25 * LetoThe2nd would guess a 4x-10x penalty. Aug 16 11:05:35 Let's hope so, because it's tied to funding... Aug 16 11:06:02 sveinse: obviously you need to throw more ram to the host to give the vm enough ram to play with Aug 16 11:06:38 sveinse: show him the following equation with the numbers appropriately filled in: if (time_for_tests * your_pay_per_hour) > your_wanted_invest -> just buy. Aug 16 11:07:36 LetoThe2nd: No, I don't think anybody will be unreasonable. I just have to quantify it to justify it. Aug 16 11:08:48 Is virtual build server just as bad? Our IT department runs a datacenter, so there is a requirement that every server is virtual. Aug 16 11:09:15 sveinse: thats only natural. and for investment of size, absolutely reasonable. but if we're talking about stuff thats in the sub-1k range, or maybe even sub-300, evalutaion quickly becomes pointless if its a one-off invest Aug 16 11:09:35 sveinse: depends on the kind of 'virtual' Aug 16 11:10:45 of course, i'm always speaking from my perspective Aug 16 11:12:07 if its a xeon beast tricked out with all the ram and you can get dedicates slices then the overhead won't be too much Aug 16 11:12:29 obviously thats a different experience to running oe builds inside virtualbox on a macbook air Aug 16 11:12:31 (which i've seen) Aug 16 11:13:15 LetoThe2nd: Our current build server run under Hyper-V with a dedicated phy (SSD) drive. 16 CPUS (E5-2637), but only 8GB ram (which can be extended or course). It's performance for Yocto SDK based Qt app building is decent. Aug 16 11:13:59 I think I will propose adding a spin-drive and much more RAM Aug 16 11:14:00 yeah. still that same xeon certainly makes a difference between natively being linux and running xen or even lxc/docker with direct memory access, and some hyper-v thing that maybe even has to rely on file-based container images. Aug 16 11:14:56 To my experience, the biggest improvement in a VM is getting direct access to storage Aug 16 11:16:27 that and enough ram, yes. Aug 16 11:20:09 all this talk of xeon is good for my stock, keep at it Aug 16 11:20:30 rburton: yes, your at Intel, right? Aug 16 11:20:33 yeah Aug 16 11:21:43 rburton: other noble gases relevant for you too? Aug 16 11:22:13 LetoThe2nd: I don't think Xeon is a gas. Xenon is :P Aug 16 11:22:21 sveinse: pun intended :) Aug 16 11:22:27 :D Aug 16 11:22:33 success \o/ Aug 16 11:22:46 Intel Raon :P Aug 16 11:22:53 no wait, brand used Aug 16 11:23:05 Heium. gives you more vowels. Aug 16 11:29:43 I have to say I really like the sstate scheme. rm -rf tmp; bitbake core-image, and you're set in minutes Aug 16 11:30:11 indeedy Aug 16 11:31:25 My recipe, which use EXTERNALSRC rebuilds quite a bit thou. I suppose it's because of the external-thing, right? Aug 16 11:39:51 i try to compile postgresql on krogoth branch but i have got an error checking Python.h usability... no | checking Python.h presence... no | checking for Python.h... no | configure: error: header file is required for Python | NOTE: The following config.log files may provide further information. | NOTE: /md0/gletourneur/projects/openembedded/oe-core/build/tmp-glibc/work/i586-oe-linux/postgresql/9.4.8-r0.0/build/config.log | E Aug 16 11:39:58 any idea Aug 16 11:58:09 I'm trying to figure out why the meta-oe glog recipe no longer builds on my machine. It seems it tries to link to libgcc_s.so.1 from my host distro, because -L/usr/lib is passed to the linker when linking the tests Aug 16 11:59:55 Could someone give some hint as to how it is supposed to work? libglog.la has libdir = /usr/lib so I guess that's where the -L/usr/lib comes from? Aug 16 12:00:43 yeah Aug 16 12:01:02 well, maybe. Aug 16 12:01:11 pastebin the actual logs? Aug 16 12:01:59 How do I find out why a setscene task failed? I just get a warning but no reference to a log file or any other output that would indicate what the problem actually was. Aug 16 12:02:03 warning is WARNING: Setscene task /home/clemens/dev/poky/meta/recipes-support/db/db_6.0.30.bb:do_packagedata (/home/clemens/dev/poky/meta/recipes-support/db/db_6.0.30.bb:do_packagedata_setscene) failed with exit code '1' - real task will be run instead Aug 16 12:02:28 I'm pretty sure I broke this, but I need to be able to debug this to fix it. Aug 16 12:03:13 I'm trying to build qt 5.5.1 using dylan infrastructure. For some reason the qmake binary of the qtbase package is build for the build target, instead of the target arch. Aug 16 12:03:25 is that a bug that was overcome somewhen? Aug 16 12:04:01 building target qtcore will build a target qmake, you want to build the native form first Aug 16 12:04:56 T_UNIX: Do you have DEPENDS += "qttools-native" in your recipe? Aug 16 12:05:34 sveinse: I have "grep qtbase-native * qtbase_5.5.1.bb:DEPENDS += "qtbase-native"" Aug 16 12:06:29 I think qttools[-native] is not available in dylan release. Aug 16 12:07:25 https://www.irccloud.com/pastebin/plEEbQfD/compile%20log Aug 16 12:07:26 T_UNIX: ok, sorry, then I don't know Aug 16 12:07:35 rburton: ^ Aug 16 12:08:09 rburton: there is no qtcore target :-/ neither grep nor find could reveal one Aug 16 12:09:44 rburton: I'm not quite sure what's going on.. Either the -native qmake is packaged in the package stage of the qtbase recipe, or the qmake in qtbase is compiled with the wrong compiler :-/ Aug 16 12:13:20 I'll have a closer look at the compile and install stages of the corresponding recipes.. Aug 16 12:38:17 where would I find logfiles for a setscene task? Aug 16 13:06:54 Can someone tell me the difference of DEPENDS and RDEPENDS? I don't get it out of the refreence Aug 16 13:07:14 http://www.yoctoproject.org/docs/1.8/ref-manual/ref-manual.html#var-RDEPENDS Aug 16 13:08:10 depends is build time, rdepends is runtime Aug 16 13:08:37 ok, so if I do a only depends, Yocto will not include this package into the rootfs? Aug 16 13:09:10 yes, but if the depends results in eg your application linking to a library, that library's package is automatically added to RDEPENDS Aug 16 13:09:29 which is why you never see recipes rdepending on libc and so on Aug 16 13:10:02 So is it better to define everything twice, DEPENDS and RDEPENDS or only RDEPENDS? Aug 16 13:11:07 if the runtime depend is a library then it will be automatically added Aug 16 13:11:10 anything else depends on context Aug 16 13:11:20 a package will build-depend on make but won't need it at runtime Aug 16 13:11:39 a package may dlopen() a library so it won't get detected automatically, so you need to add it to RDEPENDS Aug 16 13:11:54 an app may call other apps, so you'll need them in RDEPENDS but not DEPENDS Aug 16 13:12:07 there's no One True Answer, you just have to reflect reality in the packaging Aug 16 13:12:26 ok Aug 16 13:12:42 So I will look through my bunch of recpies, I guess I did it wrong Aug 16 13:25:46 HyP3r: the latest version of the reference manual has more information by the way: http://www.yoctoproject.org/docs/2.2/ref-manual/ref-manual.html#var-DEPENDS Aug 16 13:26:10 i added the note because i've seen a lot of people being confused by DEPENDS vs. RDEPENDS Aug 16 13:26:41 the newer versions of the manual have a lot more stuff in general Aug 16 13:29:26 HyP3r: a good rule of thumb is to only use RDEPENDS if you're really sure what you're doing, and always use DEPENDS otherwise Aug 16 13:30:42 RDEPENDS is only needed if you have dependencies between packages that yocto can't infer by itself, e.g. if some program assumes that another program is installed because it runs it directly Aug 16 13:31:02 for library dependencies, yocto can figure it out by itself Aug 16 13:42:07 ...except e.g. in the dlopen() case i just noticed rburton mentioned :P Aug 16 13:48:35 Ulfalizer: in my case the image generation recpie points with IMAGE_INSTALl to some of my packages. E.g. foo-www. This package does not use autotools or something, it has only a mkdir -p ${D}/www and cp -Rd ${S}... ${D}... Aug 16 13:48:57 So I'm pretty sure that can't see that this foo-www package needs python, apache, mod-wsgi, django, ... Aug 16 13:49:13 So I decied to add some RDEPENDS, if I understood it correct Aug 16 13:50:30 But I gues I have to to a cleanall and recompile everything to make sure I have enough build dependencys Aug 16 13:50:34 yeah, unless there's some relevant bbclass that adds the RDEPENDS for you, and they're not straightforward library dependencies, you'll have to declare them yourself in RDEPENDS Aug 16 13:51:33 HyP3r: see http://www.yoctoproject.org/docs/2.2/ref-manual/ref-manual.html#checking-for-missing-build-time-dependencies re. missing build dependencies Aug 16 13:52:22 tldr; $ wipe-sysroot + rebuild should be a safe bet Aug 16 13:52:57 yeah this tmp dir deleting sounds good, Aug 16 13:53:10 or just run wipe-sysroot Aug 16 13:54:38 yeah I like your too long didn't read :P Aug 16 13:56:23 Another thing I'm currently handling but I'm not sure if this way is good: deamons and programs like samba, ntp, apache2, avahi do have configuration files, and I want to insert my own configuration files. I solved that by createing bbappend recpies for all those recpies (samba, apache, ...) and appened the install phase: do_install_append(). While this routine I added (copy over) my own configuration Aug 16 13:56:29 I this a good way? Aug 16 13:58:00 probably not a horrible approach at least Aug 16 13:58:28 check out the CONFFILES variable too Aug 16 13:59:06 ... not a horrible approach ... ooookay. Is there a better way? Aug 16 13:59:17 not sure Aug 16 13:59:47 well then I think its ok, for me, third week with yocto \o/ Aug 16 13:59:52 can't think of anything for "global" configuration files Aug 16 14:30:42 For anyone familiar with qemu, is it normal that the last line I see is "main-loop: WARNING: I/O thread spun for 1000 iterations" when I try to run my yocto-built image on qemuarm? Aug 16 14:30:57 Google suggests some people had freezing problems for this but with other operating systems. Aug 16 15:59:20 Is there a way to configure a recpie that way, so every build time its completely _rebuilt_? Aug 16 16:01:47 HyP3r: http://www.crashcourse.ca/wiki/index.php/BitBake_task_flags Aug 16 16:01:57 esp [nostamp] Aug 16 16:03:15 https://github.com/01org/luv-yocto/blob/master/meta/classes/kernel.bbclass#L365 Aug 16 16:03:44 ok, but what should I write do_install[nostamp] or do_compile[nostamp], or...? Aug 16 16:04:19 Depends on how early you want it to start? At do_compile? do_configure? do_unpack? Aug 16 16:05:37 so I can also do do_fetch[nostamp] so the package will be allways fetched, configured, compiled... Aug 16 16:05:44 do_unpack is quite common Aug 16 16:05:49 * jubr made a mental note of the [dirs] flag, a bit nicer than all those `mkdir -p` in do_install Aug 16 16:06:06 Because I also need that this packages I always fetched Aug 16 16:06:20 I've used nostamp there on a few problematic recipes that don't play nice with ssate Aug 16 16:06:22 CTtpollard: yep, do_unpack also ensures teh do_patch is run Aug 16 16:06:48 Is this correct? Aug 16 16:07:48 HyP3r: yes, if you need it to fetch every time Aug 16 16:07:54 yes :( Aug 16 16:08:04 so it will refetch, unpack configure etc etc Aug 16 16:08:35 Its nice that yocto has for everythiing a solution Aug 16 16:09:00 * jubr agrees Aug 16 16:16:15 only downside is there's often more than one solution, and it's not always clear which is best :) Aug 16 16:16:53 HyP3r: re: overriding config files via bbappend, recipetool's appendfile command automates that process, can save some time over doing so manually, in case you didn't know Aug 16 16:23:17 kergoth: I don't even know recpietool Aug 16 16:23:41 recipetool and devtool are extremely valuable tools, really helpful and time saving Aug 16 16:23:57 kergoth: I tried doing a directory rename like I mentioned yesterday Aug 16 16:24:23 I delete tmp but made no other changes. Running bitbake again fails trying to find the absolute path of my layer.conf file -- any ideas? Aug 16 16:24:23 CTtpollard: that rpi change was from me :) but its not fixing all cases. One of these days I will nail it Aug 16 16:24:58 rburton: the problem is that vardepexclude does not help when DATE and TIME are used inside anonymous python snippets Aug 16 16:25:05 HyP3r: recipetool appendfile meta-mylayer /etc/some-file /path/to/my/replacement/file -> writes a .bbappend tinto the meta-mylayer layer for whatever recipe provides /etc/some-file (assumign it's been built already, or you can specify the recipe with -r) Aug 16 16:25:06 rburton: or even names python functions Aug 16 16:25:24 or say I could not beat bitbake to accept it Aug 16 16:25:38 khem: anonymous python won't affect the checksums, and named functions you can use vardepsexclude on Aug 16 16:25:41 thats the nub of the issue I am seeing Aug 16 16:25:41 * kergoth yawns Aug 16 16:25:48 kergoth: sound good, I'll try to find some documentation about that Aug 16 16:26:13 kergoth: thats what I thought but it seems base hash does account for anon python Aug 16 16:26:19 nope Aug 16 16:26:26 I have an example :0 Aug 16 16:26:51 anonymous python code was always explicitly *not* included. the results of what the functions do can affect che ksums, but not their code Aug 16 16:27:20 kergoth: Tnx! recipetool looks nice, will look into that one! Aug 16 16:27:38 kergoth: Would removing build/tmp and manually updating paths in build/conf be sufficient for an overall absolute path change? Aug 16 16:27:40 including the code would be a clear bug IMO. the whole point of anonymous python is to alter metadata, and the effects of that will affect other checksums regardless, so there'd be no point including it Aug 16 16:28:07 AgentElrond: yes, altering bblayers.conf would fix yourl ayer.conf error. that + wiping tmp would do Aug 16 16:28:55 kergoth: see https://gist.github.com/kraj/63a3f0ac6cd6af31e8e57d4fd35a3059 Aug 16 16:29:27 khem: that won't work, you're putting it on the indirect thing, which doesn't work given the current implementation Aug 16 16:29:31 kergoth: thats a abridged snippet with intesting bits, now if I comment out stamp = assignment to be an empty string all works ok Aug 16 16:29:45 khem: tweak_image_name[vardepsexclude] += "DATE TIME" Aug 16 16:31:05 kergoth: Thanks! Aug 16 16:32:35 kergoth: see https://gist.github.com/kraj/010ac278ecb7b69b35403136935ad571 Aug 16 16:32:47 thats with above Aug 16 16:32:51 included Aug 16 16:33:09 hmm, sounds like a bug in the checksumming code somewhere Aug 16 16:33:15 basehash mismatches are such a pain to diagnose Aug 16 16:33:25 taskhash too, but still Aug 16 16:33:53 kergoth: crap, we're stuck with Freescale's bsp based on fido, my `recipetool --help` only lists `create` :( Aug 16 16:34:00 ah Aug 16 16:34:30 kergoth: can whatdepends -r show the package that i was looking for? in the connman in core-image-sato example, it doesn't list connman. Aug 16 16:41:37 jubr: always a good thing to move forward and upgrade Aug 16 16:43:57 khem: preaching to the choir, baby. New device needs to be up & running & into field trials first... Aug 16 16:47:14 kergoth: You were of course correct. After updating bblayers.conf and blowing away tmp, the build looks like it may finish in 20 minutes which is better than 3 hours. :P Aug 16 16:47:26 nice Aug 16 16:47:27 kergoth: The only notable exception is that the Linux kernel got recompiled for some reason. Aug 16 16:58:13 kergoth: any workarounds you can think of ? Aug 16 16:58:33 ignoring date/time in hashes in python functions Aug 16 16:59:55 absolute worst case you could add both to the basehash whitelist, that should exclude it across the board, i expect Aug 16 17:02:02 kergoth: thanks for the review btw, I tweaked the commit message you mentioned. I've posted the main multi-config patch now, about to put it into -next and give it wider testing on the AB Aug 16 17:02:04 We talk alot about sharing sstate cache. But is it possible to share download dirs as well? Aug 16 17:02:15 kergoth: biggest worry now is whether I've missed some API changes somewhere :/ Aug 16 17:02:29 Its a pain but could have been a lot worse Aug 16 17:07:42 sveinse: yes, the conf variable is DL_DIR Aug 16 17:10:43 Will a yocto build always max out memory (and swap) regardless of how much memory you throw at it? It an impressive amount of ps which is started. Aug 16 17:11:13 sveinse: depends on what you do with it and many threads you start Aug 16 17:11:28 template-heavy C++ usually needs more memory than compiling C Aug 16 17:11:47 The default, which is N jobs and N parallel make, IIRC Aug 16 17:12:28 yeah, and N depends on $num_cpus, so whether your memory is maxed out depends on the relation between the number of CPUs you have and the amount of memory Aug 16 17:12:56 sveinse: ntl: its not safe to have same DL_DIR shared in different workspaces due to the fact that they can race against each other if building in parallel Aug 16 17:13:14 better is to setup a source mirror in vicinity Aug 16 17:14:02 I just upped our build server from 8G to 16G, and it seems it is CPU bound, with the occasional page miss Aug 16 17:14:05 if you point DL_DIR to same location, it actually means you are allowing read and write operations from different workspaces Aug 16 17:14:42 I have a suspicion that if I threw 64G RAM at it, the sheer number of processes would still max out memory at some point Aug 16 17:15:02 sveinse: unless you build something like webkit/chromium, more memory is only going to be used as file cache, which in itself is an improvement Aug 16 17:15:31 sveinse: more memory is going to give you a better performance no matter what Aug 16 17:15:44 unless you have fast SSDs Aug 16 17:15:55 in which case it may be less of improvement Aug 16 17:17:16 like its useless to use distcc now a days, there once was a time when we were cpu bound and distcc would be helpful, now a days the distcc distributor node becomes a bottleneck quickly Aug 16 17:17:22 Well, I have 16 core Xeon build server, running guest on a Hyper-V VM, with direct access to a SSD. Upping the memory from 8G to 16G helped (I think). If I need more, I need to have a clear rationale for it. Not just because I want it. Server RAM is expensive Aug 16 17:18:34 webkit/chromium is plenty happy to use 72+ cores and massive amounts of memory Aug 16 17:19:06 khem: I do share DL_DIR amongst workspaces. Luckily, the build thread is singleton, so only one bitbake is running at one given time :D Aug 16 17:21:30 jubr: yes thats modus operandi unfortunately for many in industry, I am a big believer in rolling releases, and moving all testing and QA to left of commit, which means if thats successful, then your master is always releasable Aug 16 17:21:50 distros like archlinux are good examples Aug 16 17:22:25 time bases releases are things of past CI/CD is more optimal Aug 16 17:22:46 sveinse: as long as you enforce that restriction you are safe Aug 16 17:23:13 kergoth: let me try to add it to basehash whitelist Aug 16 17:26:52 Has any of you experienced problem massively parallel building uim? meta-openembedded/meta-oe/recipes-support/uim/uim_1.8.6.bb in jethro. I just had a stop on uim do_install: "/srv/builds/core-image/build/tmp/work/x86_64-linux/uim-native/1.8.6-r0/build/uim/.libs/libuim.so: file not recognized: File format not recognized". Re-run bitbake, and everyone is happy. Smells awfully like race, doesn't it? Aug 16 17:35:08 sveinse: yes its a race Aug 16 17:36:26 sveinse: try PARALLEL_MAKEINST = "" for validating the issue Aug 16 17:36:42 if this works then the package has issues with parallel build Aug 16 17:37:14 especially with make install target Aug 16 17:38:01 kergoth: whats the incantation to whitelist something from basehash something with BB_BASEHASH Aug 16 17:43:57 sveinse: yeah, a gig per core is what you should have at minimum. We're running 40 core + 64GiB on our workstations and still regularly hit top RAM Aug 16 17:50:51 hello Aug 16 17:51:37 i'm having trouble pulling from a private repot with git. I can git clone the url from a bash prompt and I'm authenticated using keys, so I dont need to login or anything. Aug 16 17:52:04 however, when I try to do it with bitbake it fails. Here is my setup and error. Aug 16 17:52:17 https://gist.github.com/netskink/3d090630ccb44d9e58bbc5754538c3f2 Aug 16 17:53:07 davis: you may use ssh protocol Aug 16 17:54:16 i have protocl=ssh in the SRC_URI string Aug 16 17:55:07 fwiw, here is what it looks like Aug 16 17:55:14 SRC_URI = "git://devsupport/tfs/Linux%20PCM%20Project%20\(Reboot%20Edition\)/_git/PCMX;protocol=ssh;branch=master" Aug 16 17:55:31 something like SRC_URI = "git://git@github.com/xxxxx/arepo.git;protocol=ssh;branch=master" Aug 16 17:55:55 those %20 worries me Aug 16 17:56:01 yeah me too Aug 16 17:56:11 also the escaped ()'s Aug 16 17:56:21 anyway try git://git@github.com Aug 16 17:56:28 and this should help Aug 16 17:56:35 i've done things like this and it worked, Aug 16 17:56:41 #SRC_URI = "git://gitlab.com/netskink/cmake-tutor.git;protocol=https;branch=bitbake" Aug 16 17:57:03 gitlab does not have git protocol, so you need http or ssh Aug 16 17:59:07 adding a git@ to my url does not work. same error. Aug 16 18:07:37 khem: The way we come around that race is usually just to rerun the build. When it completes, it populates itself into the sstate cache, so usually we don't need that build any more. But it is really annoying when that happens (esp. when working on something else) Aug 16 18:11:32 BTW, rerunning the uim recipe with all cores enables also resolves the race. It just happens from time to time Aug 16 18:25:54 so here is a simpler situation. I have my own personal ssh repot here Aug 16 18:25:59 git@gitlab.com:netskink/awesome-breakout.git Aug 16 18:26:09 i did the SRC_URI like this Aug 16 18:26:21 SRC_URI = "git://git@gitlab.com:netskink/awesome-breakout.git" Aug 16 18:26:25 if you're copying and pasting that as is, it will not work. the url based syntax uses different separators Aug 16 18:26:26 yeah, that's wrong Aug 16 18:26:30 that does not worki either Aug 16 18:26:33 : isn't a valid separator between host and path in a url Aug 16 18:26:39 it's the separator between user and password Aug 16 18:26:41 davis is it gitlab ? Aug 16 18:26:49 yes its gitlab in that case Aug 16 18:26:54 then you might need to use :22/x/y/z Aug 16 18:27:01 after url Aug 16 18:27:09 git://git@gitlab.com/netskink/awesome-breakout.git;protocol=ssh Aug 16 18:27:21 :22? interesting, that should be default for the ssh protocol, no? Aug 16 18:27:32 Is there a way to run bb up until runing a recipe (going to test parallel builds in uim)? Is do_fetch a good place to stop and then rerun using a single CPU? Aug 16 18:27:37 or git://git@gitlab.com:22/netskink/awesome-breakout.git;protocol=ssh Aug 16 18:28:16 sveinse: as suggested try to nail it at root Aug 16 18:28:25 SRC_URI = "git://git@gitlab.com:/netskink/awesome-breakout.git;protocol=ssh" that works Aug 16 18:28:57 workaround is to disable parallel install Aug 16 18:29:20 khem: Yeah, and with 5300 recipes in build, you'd want to not run those with PARALLEL_MAKEINST="", right? Aug 16 18:29:28 kergoth: looking for your help on disabling something from basehash Aug 16 18:29:40 sveinse: its a per recipe change Aug 16 18:29:49 sveinse: you just want to do it for uim Aug 16 18:29:57 which is a reasonable workaround Aug 16 18:29:58 khem: Ah, then. thanks Aug 16 18:30:09 unless you want to help fixing it Aug 16 18:30:20 hmm. so the one works with gitlab. I use that as a model for this internal Team Foundation Server and it fails. so frustrating working with a m$ server. Aug 16 18:30:37 davis: so you missed protocol setting from what was suggested earlier Aug 16 18:31:02 i added ;protocol=ssh at the end Aug 16 18:31:12 davis: yeah thats right Aug 16 18:31:16 khem, dunno if I'm that experienced enough in yocto and am builds yet. But I am taking a look at it now Aug 16 18:31:25 timestmap 14:28 works Aug 16 18:31:30 sveinse: thats the spirit :) Aug 16 18:31:50 sveinse: I am the biggest noob here and sometimes I even get lucky. Aug 16 18:31:53 khem: see bitabke.conf for the default bb basehash whitelist. iirc you might need to append to it with a ConfigParsed event handler, in the past it didn't like _append very much Aug 16 18:33:22 davis: Encouraging words... I',m what, three week into yocto. A lot of things to process. (yet I've done embedded build systems for many years, so some is familiar. Its the semantics which is different.) Aug 16 18:33:36 zeddii: trying to get salt-minion running on an oe target, since I pulled in meta-cloud-services root password is not empty (unknown). Ideas? Aug 16 18:33:40 so in gitlab url, it was git@gitlab.com:/my_user_name/project-name.git;protocl=ssh Aug 16 18:34:33 i wonder if that my_user_name is part of the url, or is an identifier for a login? I thought the public key was all that was necessary to authenticate with the repot. Aug 16 18:34:52 sveinse: i'm glad I can be encouraging. sometimes its all i feel i can do. lol Aug 16 18:36:07 davis: heh, I can help you with that if you need it to. the blind leading the blind and so on... :P Aug 16 18:36:53 fwiw, in my case ive asked the admin to change the "cell?" name and the repot name so there aren't any spaces or parens involved. Aug 16 18:37:07 also to check logs to see if its an authentication problem Aug 16 18:37:25 How do you force rebuilding a recipe which is in the sstate cache? Aug 16 18:37:55 sveinse: ive been working off and on with yocto a while now. "Embedded Linux Systems with the Yocto Project" is a good book. Aug 16 18:38:03 for rebuild, try cleanstate Aug 16 18:38:12 fischerm: you might need to add EXTRA_IMAGE_FEATURES ?= "debug-tweaks" to local.conf Aug 16 18:38:25 give me a second, i find the syntax Aug 16 18:39:22 davis: if you are in good terms with your IT lords the life can be easier Aug 16 18:39:52 davis: you can replace git://git@xx.com with git://cell@xx.com Aug 16 18:39:58 for user to be cell Aug 16 18:40:20 sveinse: bitbake -ccleansstate ;bitbake recipe Aug 16 18:40:45 recipe here is recipe name without .bb Aug 16 18:41:12 davis, I didn't get the start of you URI problems, but you know you can setup username, certs and stuff in ~/.ssh/config, right? Aug 16 18:41:31 We do that for some private repos which require login. This way we don't have to bother in SRC_URI Aug 16 18:41:52 sveinse: no i did not know that. Aug 16 18:43:31 i used this as a guide Aug 16 18:43:33 http://nerderati.com/2011/03/17/simplify-your-life-with-an-ssh-config-file/ Aug 16 18:43:57 davis, heres one "live" example: http://paste.ubuntu.com/23062464/ Aug 16 18:44:06 my clone still works. Aug 16 18:44:12 from cmd line Aug 16 18:44:51 ahh you add an id file Aug 16 18:45:29 is the entry for IdentifyFile your private key or public? Aug 16 18:46:03 This is the private, because it needs it to set up the encrypted link Aug 16 18:46:41 RP: kergoth https://bugzilla.yoctoproject.org/show_bug.cgi?id=10152 Aug 16 18:46:42 Bug 10152: normal, Undecided, ---, richard.purdie, NEW , Bitbake always gets taskhash mismatches when using DATE and TIME in python functions Aug 16 18:47:06 its a regression compared to 2.0 and older releases Aug 16 18:47:12 sveinse: ok cool, still does not work though. I'll add that to my notes though. Aug 16 18:47:20 (I had to introduce the .ssh/keys/ directory, because ssh automatically tries the identities in .ssh/ and some servers kicks you out if you try too many keys) Aug 16 18:47:35 it would prevent me from having to pass an ident as command line when dealing with aws, i bet Aug 16 18:48:27 yes. And its convenient to setup username and stuff. "ssh buildserver" is all I write to log in to it Aug 16 18:48:27 sveinse: davis if you can convince your IT overlords to expose https protocol then you can just tool the use of username/pw in ~/.netrc and it will scale into automation Aug 16 18:49:54 * sveinse will cry when his IT overlords starts closing down port 22 Aug 16 18:50:18 windows admins barely know what it is Aug 16 18:50:45 khem: will give a shop Aug 16 18:50:49 khem: thx Aug 16 18:51:03 khem: team foundation server gives an opton to use https Aug 16 18:51:31 i'll look into netrc Aug 16 18:56:25 davis: you need something like Aug 16 18:57:23 machine login password Aug 16 18:57:31 entry in ~/.netrc Aug 16 18:58:09 and then you can use "git://machine-name/dir/to/repo;protocol=https" Aug 16 18:58:35 heh, i was just doing my netrc Aug 16 18:58:36 Hmm. Been running -ccleansstate and rebuilding uim on with all available CPU without failing any single time. Perhaps a missing DEPENDS, which now is present in sysroot? Aug 16 18:58:44 yes, it works with http on bash level Aug 16 18:58:56 5 times I've runned it Aug 16 18:59:04 sveinse: thats possible Aug 16 18:59:17 wipe-sysroot and then try rebuilding uim? Aug 16 18:59:19 sveinse: rm -rf tmp/;bitbake uim Aug 16 18:59:24 will expose DEPENDS issues Aug 16 18:59:36 ok, will try Aug 16 18:59:38 wipe the whole tmp Aug 16 19:00:08 it will reuse everything from sstate on rebuild so it will be matter of few minutes to rebuild Aug 16 19:00:09 whoa! that is working Aug 16 19:00:24 no Aug 16 19:00:31 and you will also get to experience the power OE at same time :) Aug 16 19:00:54 it went further but failed. perhaps, PV="1.0.0+git${SRCPV} Aug 16 19:00:56 davis: ofcourse it works, I use it every day Aug 16 19:01:00 " is a problem. Aug 16 19:01:24 whats the error ? Aug 16 19:01:29 While I'm waiting: Our server is running Ubuntu server, so lvm is installed by default. A bad idea for yocto? Aug 16 19:01:33 let me gist it Aug 16 19:02:12 sveinse: lvm is ok. nfs has gripes Aug 16 19:03:41 https://gist.github.com/netskink/3d090630ccb44d9e58bbc5754538c3f2 Aug 16 19:03:45 still waiting for my rm -rf tmp... Aug 16 19:04:03 the top one was where i tried http Aug 16 19:04:09 i should have renamed it Aug 16 19:04:32 its renamed now to http_attempt Aug 16 19:06:54 davis: you have a multiline string which is commented out, usually that has some issues with bitbake, add a new empty commented line at the end for it to notice end of commented mutliline Aug 16 19:06:58 line 10 Aug 16 19:07:11 add a # there Aug 16 19:07:43 and SRC_URI = "http://http://devsupport:8080 seems not ok Aug 16 19:07:57 SRC_URI = "http://devsupport:8080 Aug 16 19:08:01 might be ok Aug 16 19:09:23 Nope, 3 times of rm -rf tmp and rebuilding uim. No failures Aug 16 19:14:33 sveinse: so it might be rare then Aug 16 19:14:44 may be change parallelism Aug 16 19:15:03 kergoth: I added DATE and TIME to BB_HASHBASE_WHITELIST in bitbake.conf, did not help either Aug 16 19:15:53 I need more time on this, but let me observe failing a few more times, and then I can test with PARALLEL_MAKEINST="" to see if it goes away Aug 16 19:16:36 Can I configure (in local.conf) that a recipe shall never be read from sstate cache? Aug 16 19:19:02 For CI build server, do you change the bitbake UI, or just leave it at knotty? Aug 16 19:27:44 sveinse: yes but thats discouraged but you can add something like do_compile[nostamp] = "1" to the recipe Aug 16 19:27:51 as a gross hack Aug 16 19:31:21 khem: thx Aug 16 19:31:23 worked Aug 16 19:38:44 khem: its not a regression, we just added better debugging of when there are problems Aug 16 20:37:32 RP: OK, so its catching more errors I guess, which is fine, however, I have no way to shut this Aug 16 20:37:49 RP: none of vardepexclude work Aug 16 20:39:03 RP: from an ignorant user's pov its a regression Aug 16 21:02:59 khem: vardepexclude should work Aug 16 21:04:38 RP: I tried to add tweak_image_name[vardepsexclude] += "DATE TIME" Aug 16 21:04:41 has no effect Aug 16 21:24:27 RP: there seems to no way to disable this error thats my gripe Aug 16 22:11:07 khem: well, the above should work, the question is why it doesn't Aug 16 23:34:24 I am having trouble adding a recipe to Yocto. I have added a recipe to install the azure libraries. It creates 3 shared libraries. In the azure..rpm I see the libumqtt.so.1.1 libumqtt.so.1 libamqp.so.1.1 libamqp.so.1 libumock.so.1.1 libumock.so.1 files. In the the azure-dev...rpm file I set libmqtt.so libamqp.so and libumock.so along with header files etc. Aug 16 23:36:00 A second recipe to test the installation links correctly against the libraries but when do_rootfs runs I get the message "Computing transaction...error: Can't install azure-test-1.0-r0@armv5e: no package provides libumqtt.so". Aug 16 23:36:13 Any idea what I could be doing wrong? Aug 17 00:06:04 ping Aug 17 00:16:01 comparing the log.do_package output with the libgcrypt output show the run.do_package script has nothing is 'sonames'. The libgcrypt output shows libgcrypt.so.20. I this must be a clue. Aug 17 00:16:14 I know this must be a clue. Aug 17 01:24:16 I'm trying to get started with Yocto by getting familiar with bitbake to start. I am trying to follow the hello world example from appendix a of the bitbake manual. when I run 'bitbake printHello', I get the error message: "Please use a locale setting which supports utf-8". I modified the source of bitbake-worker slightly to gather more information and it seems that sys.getfilesystemencoding() is returning Aug 17 01:24:18 'ascii' when bitbake-worker is running. When I launch /usr/bin/env python3 and call the function myself, I get 'utf-8'. Any idea what is going wrong? **** ENDING LOGGING AT Wed Aug 17 02:59:58 2016