**** BEGIN LOGGING AT Wed Jan 09 02:59:57 2019 **** BEGIN LOGGING AT Wed Jan 09 04:53:11 2019 Jan 09 05:08:31 New news from stackoverflow: How to store downloads folder on our private repo in yocto Jan 09 08:56:03 Currently trying to add packagemanager (apt) to yocto build for development testing Jan 09 08:56:23 Can anyone tell me if this is possible as im finding conflicting information on the net? Jan 09 08:58:59 murray_: technically its only software, so it certainly is possible. the question rather is, how much effort you need, and what you expect to gain. Jan 09 08:59:27 the recipe certainly is there: http://layers.openembedded.org/layerindex/recipe/5539/ Jan 09 08:59:33 This is just for allowing me to add packages onto the dev kit for testing now Jan 09 08:59:56 well instead of reinventing the wheel i Jan 09 09:00:07 When we go to prod i will not need the package manager as i will ensure that everything we need is reinstalled Jan 09 09:00:23 i'd stick with opkg or such for dev purposes. its already there, no or only little work required. Jan 09 09:02:24 So can i install the same packages with apt as i could with opkg? Jan 09 09:02:36 And thanks for the link :) Jan 09 09:03:33 or smart, or whatever. just see what fits best, and not just say "apt because i already know it from my desktop". wasting time there does not help your product, probably. Jan 09 09:03:36 see: https://wiki.yoctoproject.org/wiki/TipsAndTricks/EnablingAPackageFeed Jan 09 09:04:04 its rpm centric, but you should get the idea. Jan 09 09:04:45 like, https://wiki.yoctoproject.org/wiki/TipsAndTricks/EnablingAPackageFeed#Easiest_Way is only 3 steps, essntially. Jan 09 10:20:13 RP: there are four ptest failures in the new perl, I didn't check how the old one compares admittedly Jan 09 10:39:30 New news from stackoverflow: How to bitbake copy a prebuilt static library into the sdk Jan 09 10:57:13 so when i use the command dpkg -l i would expect that it displays a listof install packages Jan 09 10:57:47 Instead im just receiveing a response "dpkg not found" Jan 09 11:01:35 murray_: did you actually deselect package_deb AND the package_management IMAGE_FEATURE? Jan 09 11:01:46 otherwise, well... dpkg is not installed. Jan 09 11:02:36 Is there a way to display all the recipes using yocto? Jan 09 11:02:42 So does this have to be done when building the image? Jan 09 11:02:53 I was given this image by the firmware guys Jan 09 11:10:01 Leto are you still here? Jan 09 11:12:18 murray_: of course it has to be done when building the image Jan 09 11:12:27 learningc: try to rephrase, please. Jan 09 11:13:05 murray_: hint: if you use tabcompletion its easy to type my full nickname, and then i'll be highlighted when you address me. Jan 09 11:14:01 learningc: all recipes you can build? bitbake-layers show-recipes Jan 09 11:14:12 LetoThe2nd: Thanks. On so i will just need to build it from scratch to add the PMK Jan 09 11:14:12 murray_: so in that case, ask the firmware guys to expand your image as needed. Jan 09 11:14:40 LetoThe2nd: Yes we have asked but the turn around time is long as theyu have other things in the way. Jan 09 11:15:07 LetoThe2nd: Hence why i am going to try get a new image together myself Jan 09 11:15:30 murray_: basically, yes. and you need some form of machine that serves the package repository. but if you jump through all those hoops, it is often WAAAAAY easier to just sum up the things you want added to your dev image, and then rebuild it including those. no need for package management then. Jan 09 11:16:36 LetoThe2nd: Its basically just a webserver, ftp and nodejs that i need to add. Jan 09 11:16:52 well for development use, python's simplehttpd is enough Jan 09 11:16:55 LetoThe2nd: So besides those i have no reason for PM Jan 09 11:17:05 murray_: then decide upon which you want, and rebuild your image with those added. done Jan 09 11:17:25 if someone ships you an image without the layers and instruction to rebuild it, complain Jan 09 11:18:04 LetoThe2nd: Cant be that hard to do. Will go have a look at the building process. Jan 09 11:18:23 murray_: if its just for your local, non-reproductible tinkering, this can be as simple as adding CORE_IMAGE_EXTRA_INSTALL = "nodejs lighttpd lftp" Jan 09 11:18:29 to local.conf Jan 09 11:19:23 once you have your build set up, of course. but its really the job of your firmware guys to get you started with the initial setup first. Jan 09 11:24:57 hi,I'm a yocto newbie trying to create a new recipe: https://pastebin.com/RmnHxuVt builds fine and the rpm is populated with the .ko file but "bitbake core-image-minimal" fails with https://pastebin.com/NaBhriAy any help would be greatly appreciated Jan 09 12:42:45 rburton, bitbake-layers show-recipes works. Thanks. :) Jan 09 13:34:26 just to be sure, something like that in a recipe is wrong: S = "${WORKDIR}/${PN}-${PV}" Jan 09 13:34:49 it needs to be ${BPN}, otherwise multilib builds will break, right? Jan 09 14:12:46 derRichard: yes. also thats teh default Jan 09 14:16:23 rburton: i'm asking because i see some recipes in meta-openembedded that use just PN and fail to build Jan 09 14:16:28 * derRichard prepares patches :) Jan 09 14:24:38 derRichard: default is WORKDIR/BP, BP is BPN-PV Jan 09 14:28:32 rburton: yes. i'd remove all lines like this one: S = "${WORKDIR}/${PN}-${PV}" Jan 09 14:47:36 Is there a way to have multiple -dbg packages, for example, in a given recipe? I've tried adding the names to the PACKAGES list and specifying FILES for those names, but the various -dbg all get lumped together. Jan 09 15:07:51 tgoodwin: preferred is single package. if you really want it split set NOAUTOPACKAGEDEBUG="1" Jan 09 15:10:52 rburton: alright, thanks. Jan 09 15:42:54 hello, my image bb has TOOLCHAIN_HOST_TASK += "nativesdk-python3 \ nativesdk-python3-mako \ nativesdk-python3-numpy \ nativesdk-python3-setuptools \ nativesdk-python3-requests \ " Jan 09 15:44:03 but then I got an error when ERROR: Nothing RPROVIDES 'nativesdk-python3-asn1crypto' (but virtual:nativesdk:/home/oe-builder/meta-oe/meta-python/recipes-devtools/python/python3-cryptography_2.1.4.bb RDEPENDS on or otherwise requires it) Jan 09 15:44:38 anyone can help? Jan 09 15:45:07 the SDK need "requests" module Jan 09 15:48:57 nobitut: Do you include+enable the meta-python layer from openembedded? Jan 09 15:50:45 yes. If I remove nativesdk-python3-requests, everything will be happy Jan 09 15:53:46 determine which recipe includes asn1crypto and ensure it has nativesdk in BBCLASSEXTEND Jan 09 16:17:34 thanks, why it didn't happen to other module ? such as numpy or mako for example Jan 09 16:24:17 because they already are available for nativesdk Jan 09 16:52:37 kroon: Are you here? Jan 09 18:43:26 kroon: How does this look for a depsig format: https://pastebin.com/z3eeNAzh Jan 09 18:46:47 JPEW, so beautiful it almost brings a tear to my eyes Jan 09 18:46:58 :-D Jan 09 18:47:20 kroon: Lol. Anything missing? I omitted the file sizes.... do you think they are useful? Jan 09 18:48:48 JPEW, I think file size could be interesting to monitor yeah Jan 09 18:49:27 JPEW, I guess the cost of adding it is negligible ? Jan 09 18:50:04 Ya, its easy Jan 09 18:52:30 so 2:nd and 3:rd column is user and group Jan 09 18:52:44 then device major/minor Jan 09 18:54:19 Yes, I was thinking about adding headers Jan 09 18:55:39 I also think I need to move the major minor to a separate section; the values aren't well defined for non-device files Jan 09 18:58:16 So, "ls" prints the filesize for regular files in place of character major/minor for device nodes Jan 09 18:59:51 and the "-> " suffix for symlinks Jan 09 19:06:49 kroon: Ya, I was trying to use 'find' only (I'm also not a big fan of the postprocessing that buildhistory does to remove dangling " ->" at the end of the lines) Jan 09 19:11:15 it parses ls output? uck, that's not kosher Jan 09 19:12:52 kergoth: It parses find output, so it's not *as* bad Jan 09 19:13:49 kergoth: https://git.yoctoproject.org/cgit/cgit.cgi/poky/tree/meta/classes/buildhistory.bbclass#n524 Jan 09 19:15:05 ah, so it's to avoid having to run a ton of readlink forks or something Jan 09 19:15:48 Ya, or trying to avoid multiple passes (like I'm proposing with depsig) Jan 09 19:16:18 that doesn't seem all that unreasonable to me, so the sed is just to remove the empty entries when they aren't symlinks.. Jan 09 19:16:24 ugly, but reasonable given the alternatives Jan 09 19:16:41 there's probably a bunch of stuff in there that would have been better written as python, but it started out as shell and I built on it... Jan 09 19:17:30 Indeed.... but you can't run a python function under fakeroot :( Jan 09 19:17:41 no, true Jan 09 19:17:44 and I believe it did parse ls output at one time :D Jan 09 19:18:15 python isn't always faster, too, given the linux tools are usually written in C. i.e. our attempts to unpack tarballs in python were horribly slow compared to just running tar Jan 09 19:18:17 but yeah Jan 09 19:18:51 Anyway, I already have to do multiple passes to calculate file checksums, so I figured a separate pass for symlinks wasn't the end of the world Jan 09 19:23:32 kroon: Oh, right I didn't include the size because it lists the size of directories also, which I'm not sure is portable Jan 09 19:26:26 JPEW, aha Jan 09 19:34:44 kroon: https://pastebin.com/T0dsQ0Tv Jan 09 19:35:15 We can drop the SIZES section if you want Jan 09 19:38:01 Alterantivly, it could be split up by file type, so list "mode owner size path" for all files, then "mode owner path -> target" for all symlinks, then "mode device path" for block devices Jan 09 19:39:04 But the checksum would still have to be on their own; it's just to slow to have find run them through shasum correctly. Jan 09 19:47:19 Is is not feasible to just do one line per file, including the checksum ? Jan 09 19:48:28 It's tricky. You can't do it (quickly) with shell code; you would have to write a python script. Jan 09 19:48:48 I thought you were doing this in Python ? Jan 09 19:49:21 you need to run it under pseudo ? Jan 09 19:49:29 Right it was originally, but you can't write a python function in bitbake that runs under fakeroot (which is required to get the correct UID/GID) Jan 09 19:49:52 We *could* potentially have a shell function that invoke a python script Jan 09 19:50:05 I haven't seen that done anywhere else Jan 09 19:53:12 i've split off and used standalone python scripts from tasks before Jan 09 19:53:34 kergoth: Ok. Where does the script usually live? Jan 09 20:22:57 JPEW, the pseudo db is also an output.. Jan 09 20:23:47 JPEW, im just wondering, is it not inevitable to calculate the hashes under pseudo ? Jan 09 20:38:02 kroon: Not sure I follow Jan 09 20:40:43 JPEW, as far as I understand, pseudo uses some sort of database where it stores file metadata, containing changes the pseudo user does that cant be recorded in the actual filesystem Jan 09 20:41:20 Yes, thats mostly accurate Jan 09 20:43:01 shouldn't the information in that db also be part of the output hash ? Jan 09 20:44:06 As far as what users/groups own what files, or the db file itself? Jan 09 20:44:18 the first Jan 09 20:44:38 Yes, it does. Thats why the checksums need to be run under pseudo (fakeroot) Jan 09 20:45:03 Err, not the checksums, but the file listings (if they include the UID/GID) Jan 09 20:46:12 yeah, ok Jan 09 20:51:21 JPEW, your original format was kind of nice, \n \n .. etc. at least it only required to big file name to be printed only once Jan 09 20:53:04 Ok, let me work on the external python script for a bit, I think I can get something that has all we want looks nice, and isn't too slow Jan 09 20:54:42 coolio Jan 09 20:57:41 I'm having trouble with pulseaudio. I believe it's because I don't have a logged in user to start the daemon, and it isn't supposed to run as system Jan 09 20:57:47 I'll continue with trying to get depsig.do_package stable inbetween rebuilds.. Jan 09 20:58:20 anyone know what I should do? Jan 09 21:04:23 JPEW: when is this running? If its python task context its already under pseudo Jan 09 21:04:38 or can be anyway Jan 09 21:05:28 JPEW: either the task is already running under pseudo in which case we care about permissions or its not under pseudo and we dont? Jan 09 21:09:19 huh, never played with tracemalloc. https://docs.python.org/3/library/tracemalloc.html Jan 09 21:09:32 JPEW: FWIW we standardise on names, not UID/GID numbers Jan 09 21:09:46 JPEW: and I think we should only consider them for fakeroot tasks Jan 09 21:10:14 nate02: we have a packageconfig to enable the run-as-system mode Jan 09 21:10:23 kergoth: I wonder how bitbake looks with that, interesting Jan 09 21:11:33 ah, cool, you can filter stuff out too — https://willnewton.name/2016/12/28/debugging-memory-leaks-in-python/ has an example of that Jan 09 21:11:38 yeah, that's what i was wondering too Jan 09 21:11:48 i remember trying to use heapy and the like and it was a pain in the ass Jan 09 21:11:52 this is way easier Jan 09 21:17:56 RP: ya, thats *way* easier.... I think I got down a rabbit trail of over complication :) Jan 09 21:18:00 rburton: I'm not seeing that option, i'm on pyro and i checked master Jan 09 21:18:39 nate02: haven't configured it for an age, but try systemd and autospawn-for-root Jan 09 21:19:47 rburton: ok, thanks, will do Jan 09 21:24:45 woa Jan 09 21:24:47 #Python news: Guido accepted PEP 572. Python now has assignment expressions. Jan 09 21:24:47 if (match := http://pattern.search (data)) is not None: Jan 09 21:24:47 print(http://mo.group (1)) Jan 09 21:24:47 filtered_data = [y for x in data if (y := f(x)) is not None] Jan 09 21:25:06 course it'll be a while till thats in a release, and even longer till we can use it in bitbake, but nice Jan 09 21:43:31 i need to get the linux kernel development files installed on my target so i did a "smart install linux-dev", but smart is giving me http://172.16.1.32/rpm/imx6ul_var_dart/kernel-dev-4.1.15-r0.imx6ul_var_dart.rpm: Invalid SHA (expected 3f502eff9d9f6922b004f0bd49ad576e67f96d1d, got 22a883714387489ac2ee00215a5b6de985260e6a) Jan 09 21:44:06 i then thought i'd try rebuilding that package, but bitbake kernel-dev doesn't work Jan 09 21:44:45 https://paste.fedoraproject.org/paste/W1EiPeI2sMJm3jhyUUvJzw Jan 09 21:49:55 does that SHA include the dtsi files? Jan 09 21:51:11 nm. updating the package-index worked Jan 09 21:55:51 hmm, bitbake seems to be locking up on me after "Initializing Tasks" on the latest master Jan 09 22:00:07 nm the nm. although smart now installs kernel-dev, rpm -ql kernel-dev only shows files being installed in /boot, none of which are headers Jan 09 22:00:11 just a few files Jan 09 22:00:15 what the heck? Jan 09 22:00:54 were there some snafu's in the kernel-dev package back in Morty? Jan 09 22:13:54 i don't understand why we use FILES_${PN} and not FILES_${BPN} Jan 09 22:14:29 because BPN would be the wrong name Jan 09 22:14:39 but some use BPN Jan 09 22:14:47 only where required Jan 09 22:14:53 e.g. Jan 09 22:14:54 meta/recipes-extended/mc/mc_4.8.21.bb:FILES_${BPN}-fish = "${libexecdir}/mc/fish" Jan 09 22:14:59 that's very wrong Jan 09 22:15:12 or Jan 09 22:15:13 meta-oe/recipes-connectivity/libmtp/libmtp_1.1.16.bb:FILES_${BPN}-common = "${nonarch_base_libdir}/udev/rules.d/*" Jan 09 22:15:19 both 100% wrong Jan 09 22:15:30 patches please Jan 09 22:15:47 i'm currently doing a patch series and got confused :) Jan 09 22:15:52 that'll result in doing nothing at all when BPN != PN Jan 09 22:16:13 so, BPN is mandatory only when setting S? Jan 09 22:16:17 no Jan 09 22:16:27 BPN is the PN but with prefixes/suffices removed Jan 09 22:16:34 it's valid for S because upstream's name doesnt' change when the multilib does Jan 09 22:16:47 the url isn't suddely lib32-bash-foo.tar.gz Jan 09 22:16:56 it's bash- regardless of recipe name Jan 09 22:16:57 (or native/nativesdk extension) Jan 09 22:17:18 if PN is lib32-bash, nativesdk-bash, or bash-native, all have BPN=bash Jan 09 22:17:33 meta-multimedia/recipes-multimedia/libdc1394/libdc1394_git.bb has: Jan 09 22:17:34 S = "${WORKDIR}/git/${PN}" Jan 09 22:17:37 Base Package Name Jan 09 22:17:40 it does not build here Jan 09 22:17:44 derRichard: yeah thats wrong too Jan 09 22:17:47 the fix is using BPN Jan 09 22:17:53 and killing the line Jan 09 22:18:01 because then it is the bitbake default Jan 09 22:18:03 no, its not Jan 09 22:18:09 yeah, sorry Jan 09 22:18:11 other recipe Jan 09 22:18:19 here it is S = "${WORKDIR}/git/${BPN}" Jan 09 22:18:24 the default is ${WORKDIR}/${BPN}-${PV} Jan 09 22:18:26 yeah, that's right Jan 09 22:18:32 or rather, ${WORKDIR}/${BP} Jan 09 22:18:49 meta-multimedia/recipes-multimedia/caps/caps_0.9.24.bb Jan 09 22:18:50 S = "${WORKDIR}/${PN}-${PV}" Jan 09 22:18:53 this line can be removed Jan 09 22:18:56 yep Jan 09 22:19:18 its almost like meta-oe is 50% unmaintained/unreviewed cruft :( Jan 09 22:19:40 in meta-oe/recipes-shells/mksh/mksh_56.bb i see: Jan 09 22:19:40 ALTERNATIVE_TARGET[sh] = "${base_bindir}/${BPN}" Jan 09 22:19:44 is this correct? Jan 09 22:19:54 everywhere else the recipe uses $PN Jan 09 22:20:03 yes, because you want the target to be mksh Jan 09 22:20:14 not lib32-mksh Jan 09 22:20:25 i'd just use mksh for clarity tbh Jan 09 22:20:42 agreed. but now it makes more sense Jan 09 22:21:04 just a few hours ago i was thinking about writing a tool to compare pkgdata for lib32-foo and foo, checking that the only differences were the multilib prefixes on package names and libdir name Jan 09 22:21:16 which would spot the mc problem in oe-core Jan 09 22:21:32 when i send patches for these recipes, will your autobuilder test them all with multilib? Jan 09 22:21:45 i'm not sure if i have the time to build test everything i touch :( Jan 09 22:21:51 unlike meta-oe, oe-core has automated multilib tests so its generally right, but some packaging quirks like that mc one won't cause errors, just files in the wrong place Jan 09 22:22:14 anyway its 22:22 here, g'night Jan 09 22:38:37 hmm, in meta/recipes-support/curl/curl_7.62.0.bb i see: Jan 09 22:38:38 PACKAGES =+ "lib${BPN}" Jan 09 22:38:42 this looks also odd Jan 09 22:38:57 for both 32 and 64 builds this would be "libcurl" Jan 09 22:49:47 Who should email in order to file a request for receving an account for openembedded-core-contrib ? Jan 09 22:54:33 Or will https://www.openembedded.org/wiki/Special:RequestAccount give me that ? Jan 09 23:06:46 kroon, that only gives you a wiki account Jan 09 23:07:07 does the layer have a README with a contribution policy? Jan 09 23:09:49 Crofton, openembedded-core-contrib is just mirroring openembedded-core I assume Jan 09 23:13:57 okay, i take this back Jan 09 23:14:02 too many variables ;D Jan 09 23:17:49 Hi, I am considering including ${DATETIME} in my DISTRO_VERSION in order for each of my builds to automatically have a unique version, which matches between what's displayed in the distro's user interface (via /etc/os-release) and the generated filename of the images. Jan 09 23:17:59 Is this a good idea or is there something else I should consider? Jan 10 02:32:26 halstead: can you look at my comments on YOCTO-13101 and see if it makes sense? **** ENDING LOGGING AT Thu Jan 10 03:00:02 2019