**** BEGIN LOGGING AT Mon Sep 23 02:59:57 2019 Sep 23 07:16:49 good morning Sep 23 07:18:04 mckoan: Good morning to you Sep 23 07:27:50 good morning Sep 23 10:06:47 hello all Sep 23 10:07:18 why when I say "bitbake -c " does it build all task that depend on it? rather than only building the task itself and only building the others when needed? Sep 23 10:07:32 i.e when saying "bitbake gdb -C unpack" it also appears to compile it Sep 23 10:09:03 litb, sure your not confusing -c with -C here ? Sep 23 10:09:21 kroon, I thought -C just means "ignore the cache and force execution" Sep 23 10:10:14 + "run the default task for the target" Sep 23 10:10:18 kroon, ahh I looked it up again and see now that -C does something different indeed! Sep 23 10:10:54 thanks. I'm such a lazy dude not to look it up before asking. shame on me Sep 23 10:11:39 I think "-c -f" is what I had in mind Sep 23 10:12:29 yeah Sep 23 10:13:32 well, i think dependant tasks are also run Sep 23 10:15:11 gotta fetch sources before one can unpack them Sep 23 10:23:12 kroon, sure. Fair enough. Sep 23 10:23:33 New news from stackoverflow: How to install Yocto on Beaglebone black´s Emmc using U-Boot Sep 23 12:24:06 what's the recommended way to provide a config file for a specific package? I wanna provide a config file for systemd-networkd. should I make a completely separate package that consists only of this config file? Sep 23 12:24:26 or should I use a .bbappend on the systemd recipe and add the config file there? Maybe on the systemd-conf recipe? Sep 23 12:26:09 either works Sep 23 12:26:15 systemd-conf seems appropriate Sep 23 12:26:24 currently in my OS/dist-layer I have a "recipes-core/conffiles/*.bb" folder for various config files. i.e. a "vconsole-conf.bb". and now a "eno-network.bb" for networkd Sep 23 12:26:26 that ensures it gets pulled in Sep 23 12:27:01 maybe calling it networkd-conf.bb to keep the pattern Sep 23 12:27:43 rburton, hm, I see. i thought that maybe I would mess with yocto internals this way. i mean, nothing guarantees that systemd-conf stays. so if yocto decided to remove this package or rename it, I would be screwed Sep 23 12:29:00 it would only disappear on an upgrade to oe-core, and its trivial enough to solve Sep 23 12:29:14 you'll get a build error with a dangling bbappend so you'll know Sep 23 12:30:30 ah, i see Sep 23 13:06:24 rburton, perhaps automatically populating the value of "SRC_URI" with all files in files/ would be a compact solution and deserve a class of its own. each file name has the format "folder1-folder2-...-filename" and will be installed to "${D}/folder1/folder2/.../filename". a dash can be escaped by double-dashing Sep 23 13:10:17 litb: you can use limited globs in SRC_URI, see matchbox-desktop for an example Sep 23 13:10:44 or a directory name, see xserver-nodm-common Sep 23 13:14:48 rburton: from my understanding file://dir/* is exactly the same as file://. isn't it? https://git.yoctoproject.org/cgit.cgi/poky/tree/bitbake/lib/bb/fetch2/local.py#n58 Sep 23 13:15:23 anything with * in the path will be replaced by . Sep 23 13:20:48 that search logic is suboptimal to say the least Sep 23 13:21:41 but its unpack that does the copy Sep 23 13:21:42 # Localpath can't deal with 'dir/*' entries, so it converts them to '.', Sep 23 13:21:43 # but it must be corrected back for local files copying Sep 23 13:22:10 eg Sep 23 13:22:11 NOTE: Unpacking /home/ross/Yocto/poky/meta/recipes-sato/matchbox-desktop/files/vfolders/* to /data/poky-tmp/master/work/corei7-64-poky-linux/matchbox-desktop/2.2-r0/ Sep 23 13:34:19 dl9pf: BB_HASHSERVE = "host:port" Sep 23 13:34:54 dl9pf: The server listens on port 8686 unless you tell it otherwise, so "host:8686" Sep 23 13:36:56 JPEW: good news is we're down to two failures with master-next. I have a fix for the selftest one, the esdk one is eluding me, can't even reproduce :( Sep 23 13:38:59 RP: Is that with the "only run setscene once" patch? Sep 23 13:41:21 JPEW: yes Sep 23 13:41:36 rburton: I see... I've to check this out in more depth. By any chance, do you know/understand the difference between -H and -P in cp? Because it looks to me they are contradictory and they're used together in https://git.yoctoproject.org/cgit.cgi/poky/tree/bitbake/lib/bb/fetch2/__init__.py#n1509 Sep 23 13:42:02 lol Sep 23 13:42:37 erm no Sep 23 13:42:39 RP? Sep 23 13:43:36 https://daniel.haxx.se/blog/2019/07/22/curl-goez-parallel/ Sep 23 13:43:53 adelcast: ^ may be interesting Sep 23 13:45:09 qschulz: looks like a bug :( Sep 23 13:47:13 qschulz: https://git.yoctoproject.org/cgit.cgi/poky/commit/bitbake/lib/bb/fetch2/__init__.py?id=0c3ce684100546cf628d96c0f64a635242a10e16 Sep 23 13:50:23 RP: I'd love to talk over that solution. Not that it's necessarily wrong, I just want to understand why that solution is correct. Maybe in the engineering call tomorrow? Sep 23 13:54:23 rburton, thanks! Sep 23 14:15:56 JPEW: I'm not 100% convinced its right Sep 23 14:16:09 JPEW: it is however fairly unavoidable :( Sep 23 14:17:45 RP: yup, they should have removed the P in the commit, it works because it's parsed from left to right and then H overrides P Sep 23 14:18:20 RP: well at least from https://github.com/coreutils/coreutils/blob/master/src/cp.c, which I guess is the "correct" one? Sep 23 14:18:38 qschulz: I think so Sep 23 14:53:31 Hello Guys, someone here could give a feedback about application management inside Yocto ? We have several project with different applications, I made a partition to split OS/app binary and update designated application with opkg. And as always I'm looking for the better way to do that :) Sep 23 15:05:53 PinkSnake: We generate with Yocto an immutable OS (so no rpm, dpkg or opkg) and we use a directory (optionally mounted on NFS) to provides binaries handled with specific application. Sep 23 15:07:04 alessioigor Thx for replying! Seems also a good solution :) Sep 23 15:07:47 PinkSnake: I hope so :) Sep 23 15:21:39 I want to debug a BAD_RECOMMENDATIONS which isn't taken into account, do you have any tips? Sep 23 15:22:37 qschulz: easiest way is to go onto the target and remove the package Sep 23 15:22:49 the package manager will tell you what is causing it to be installed Sep 23 15:23:01 if it removes, then your image explicitly pulls it in Sep 23 15:26:18 it's eudev-hwdb, there is only one package RRECOMMENDS-ing it, usbutils. In our image recipe we have BAD_RECOMMENDATIONS += " eudev-hwdb ". Sep 23 15:27:24 (and BAD_RECOMMENDATIONS has it in bitbake -e) Sep 23 15:27:26 bad recommendation simply is a hint, there is a way to avoid it ever being installed.. and if it's a hard dep, it'll error Sep 23 15:28:33 I know, there is no rdepends anywhere in our layers nor in poky AFAICT, so it shouldn't behave like this. Sep 23 15:28:34 PACKAGE_EXCLUDE="eudev-hwdb" Sep 23 15:28:46 or... it gets pulled by the auto-rdepends mechanism Sep 23 15:28:56 iuf your image can't be generated it'll error.. Sep 23 15:28:58 AHAH. I forgot about that one. Let's see :D Sep 23 15:29:25 BAD_RECOMMENDATION is only a hint, it says don't install it if recommended.. but if something else says it NEEDS it, then it'll go in Sep 23 15:29:43 exclude won't even allow that Sep 23 15:33:50 fray: yup, I know. let's see what it says with PACKAGE_EXCLUDE, thanks for the hint Sep 23 15:34:12 forgot about package_exclude Sep 23 15:35:28 I know it was added precisely to deal with the case where packages came in and just wouldn't stay out Sep 23 15:42:48 RP: I've got a bunch of -stable 5.2 bumps that I've done locally here while working through strace. I didn't see a M3 announcement, so I'm not sure if it is still being built .. but I was wondering if you'd rather I sit on those patches until M3 is out, or should I just fire them out anyway ? Sep 23 15:43:42 * zeddii hates getting patches when other things are blowing up .. hence why I ask. Sep 23 15:44:59 ^Piraty, thanks, seems like libcurl has had this for a while, but its not being used by opkg.....this actually should be easy to implement Sep 23 15:55:37 zeddii: M3 is built and in QA Sep 23 16:01:43 ok. so I'll send them out, and go back to printk'ing the kernel after that. Sep 23 16:02:34 all can be debugged w/ printk.. ;) Sep 23 16:04:09 truth Sep 23 16:14:33 kergoth: rewrote the bulk of bb-whatdepends with three lines of tinfoil. for recipe in tinfoil.all_recipes(): if args.recipe in recipe.depends: print(recipe) Sep 23 16:14:44 ha, nice Sep 23 16:14:59 does that cover runtime or just build time? Sep 23 16:15:08 just build time Sep 23 16:15:12 (which is all i wanted) Sep 23 16:15:19 rdepends should be done via pkgdata Sep 23 16:29:58 debatable. there's value in being able to operate based on the defined metadata vs the actual results including shlibs after a build, especially if you're investigating why something is being *built*, which was the main purpose of the script Sep 23 16:30:21 both have a certain a mount of value, i think Sep 23 16:31:19 kergoth: hm, i guess something being built would mean chasing the actual depends tree, not the DEPENDS variable Sep 23 16:31:24 yeah fair enough Sep 23 16:32:10 there's also the question of whether task depends are included. my ideal was to track down why something was sucked in and ideally why. *this* flag on *that* task is why this is being built. but never quite got there, there are a lot of forms of dependency :) Sep 23 16:32:28 yeah Sep 23 16:32:39 -g sucks when recrdeptask is involved, for example Sep 23 16:32:42 right Sep 23 16:33:15 "oh good, *everything* depends on this, that helps.." Sep 23 16:34:15 i'll keep bb-depends around for the common case Sep 23 16:34:30 but will try rewrite whatdepends to follow the actual task depends at some point Sep 23 16:34:34 yeah, makes sense, i'm sure it'll be of use Sep 23 16:34:43 often stare at -g output wondering why something like attr is depending on py3-native for example Sep 23 16:35:22 i most often have the gumption to make such a script work again when i'm in the midst of such a diagnosis, but then i'm usually under time pressure and just want to move on, so it never gets done Sep 23 16:35:27 heh Sep 23 16:48:01 kergoth: the main problem is there is no record in the system of where a dependency came from Sep 23 16:48:06 yeah Sep 23 16:49:10 An option to filter the logging by a specific dependency name may be interesting Sep 23 16:49:34 The standard logging does normally tell you but its near impossible to read. With filtering... Sep 23 17:00:31 Is it possible to modify the SRC_URI variable in a task before do_fetch? When I print d.getVar("SRC_URI") at the end of my new task, it has the files I expect, but then the fetcher debug output doesn't indicate it's looking for all the files. Sep 23 17:13:16 Hello. Would this be the right place to ask for advice on yocto cross-compilation? Sep 23 17:16:52 cdgarren: tasks get isolated copies of the metadata. tasks can't modify the data of other tasks or it'd make the build non-deterministic due to task ordering. Sep 23 17:17:02 cdgarren: use anonymous python to do something at parse time if needed, or use a prefunc Sep 23 17:17:07 Ah, got it. Thanks Sep 23 17:17:16 i.e. do_foo[prefuncs] += "my_other_function" Sep 23 17:17:31 or use _prepend/_append, depending on your requirements Sep 23 17:18:16 So I could append to SRC_URI in a prefunc, and the fetcher would get those files? Sep 23 17:21:44 yes. it'd be better to do it at parse time, though, then bitbake -e would show the changes Sep 23 17:25:23 kergoth: Got it. That works perfectly, thanks Sep 23 17:28:56 np Sep 23 17:43:31 fray: turns out PACKAGE_EXCLUDE didn't make yocto complain Sep 23 17:43:48 maybe there is a bug in the BAD_RECOMMENDATIONS then.. Sep 23 17:44:00 We removed libsolv from PACKAGECONFIG for the target opkg in thud Sep 23 17:44:05 are there builtin wrappers i can call in a recipe for unpacking a tarball, or should i just invoke tar directly? Sep 23 17:44:06 so I put it back Sep 23 17:44:10 I'd suggest explaining what you did and how to reproduce it to the oe-core mailing list.. Sep 23 17:44:35 and then I remembered I asked for backporting a few patches from opkg 4.0 to 3.6 because it fixed an issue for me with BAD_RECOMMENDATIONS Sep 23 17:44:37 RP: accidentally changed a couple of bug milestones in Bugzilla. I've reverted the changes but you may see some notification emails Sep 23 17:45:01 but apparently, (WHICH IS VERY OBVIOUS FROM THE COMMIT LOG GMRBNLRBGLKRBGJBRG), someone needs libsolv > 7.2 to make use of it Sep 23 17:45:52 so I don't know what the fuck I did to make it work nicely back when I asked to backport the patches for opkg but it definitely isn't enough now (works fine with libsolv in PACKAGECONFIG and backported libsolv_7.3) Sep 23 17:46:07 the question is if it's supposed to work without libsolv support Sep 23 17:46:17 (BAD_RECOMMENDATIONS I mean) Sep 23 17:54:48 New news from stackoverflow: Toolchain build using yocto Sep 23 18:52:44 tgamblin: np, thanks for the headsup Sep 23 19:46:23 I'm building a custom library (own recipe) on top of poky thud for Raspberry Pi Zero. I'm getting a QA error for my library stating "Architecture did not match (x86-64, expected ARM)". I checked the produced .so file, and it's x86 instead of ARM. The library is using make, and I noticed that if I change the lib install location from /usr/lib to Sep 23 19:46:23 e.g. /usr/local/lib then the QA error disappears. My question is still, that why is the library built for x86 instead of ARM? Sep 23 19:49:31 vlo86: chances are that the Makefile just doesn't properly handle the passed CC, CFLAGS, arch and all that Sep 23 19:50:01 vlo86: pretty much one of the poster examples why hand-written Makefiles are a massive PITA when it comes to cross compiling. Sep 23 19:55:31 vlo86: what LetoThe2nd said. that qa warning is explicitly to catch build systems that hardcode 'gcc' and don't use the cross compiler. Sep 23 19:55:32 Thanks for the pointer. I will check this lead. This is a third party library and I'm just trying to include it into my image =) Sep 23 19:55:53 file a bug with the library at least, if its bare makefiles you can normally override the right variables Sep 23 19:55:57 is it open source? Sep 23 19:56:02 i'm confused about kernel module recipe/package naming.. i have a recipe that makes a kernel module foo.ko, but adding kernel-module-foo to MACHINE_ESSENTIAL_EXTRA_RDEPENDS causes bitbake to say it can't find the package. what's the deal? Sep 23 19:56:15 rburton: "if the Makefile is at least halfways properly written" Sep 23 19:56:42 rburton: yeah, it's open source: https://github.com/joan2937/pigpio/blob/master/Makefile Sep 23 19:57:59 CC = $(CROSS_PREFIX)gcc Sep 23 19:58:04 if it used ?=... Sep 23 19:58:14 rburton: thats what i meant Sep 23 19:58:24 right Sep 23 19:58:29 LetoThe2nd vs. Makefiles: 1:0 Sep 23 19:58:53 when you call make, also add CC="${CC}" AR="${AR}" etc etc Sep 23 19:59:08 i also noice no way to override LDFLAGS Sep 23 19:59:11 fork it and port to cmake, meson, autotools, $WHATEVER_SANE :P Sep 23 19:59:18 that's definitely the better option :) Sep 23 19:59:19 So what you're saying is I should patch CROSS_PREFIX definition? Sep 23 19:59:31 https://github.com/joan2937/pigpio/blob/master/CMakeLists.txt <-- hey use that instead Sep 23 19:59:49 Heh, I was already considering that for a moment. But good to know now that this Makefile is in fact the problem. Sep 23 19:59:52 LIBRARY DESTINATION ${DESTDIR}/usr/local/lib <-- argh that hardcodes locations though Sep 23 20:00:12 LetoThe2nd vs. people who write handwritten Makefiles: 2:0 Sep 23 20:01:37 i take this as an omen i should call it a day. Sep 23 20:02:56 Heh. Thanks a lot for your help! Sep 23 20:06:12 https://github.com/joan2937/pigpio/issues/211 is a bug about getting cross support actually working, Sep 23 20:06:16 note how its not closed Sep 23 20:10:04 Well would you look at that =) Sep 23 20:10:34 That was a mere statement of no surprise, not a real question :D Sep 23 20:24:26 is there a way to see the working directory during recipe task execution? Sep 23 20:43:32 mischief: bitbake X -e | grep WORKDIR= ? Sep 23 21:33:13 JPEW: Still can't reproduce the esdk failure :( Sep 23 22:53:03 RP: thanks. any clue about the kernel module question? Sep 23 23:07:25 mischief: can you paste your kernel module recipe ? Sep 23 23:07:39 are you inheriting module ? Sep 24 00:46:56 anyone know how much a platinum membership costs? Sep 24 01:18:11 at least twenty dollars Sep 24 01:24:49 lol Sep 24 01:25:00 which currency is that? Sep 24 02:02:14 radsquirrel: send email to Nicolas Dechesne he can give you more info **** ENDING LOGGING AT Tue Sep 24 02:59:59 2019