**** BEGIN LOGGING AT Wed Mar 31 02:59:56 2021 Mar 31 03:13:51 RP: OK, that works. It's a rebuild-the-world change though :/ Mar 31 07:41:40 good morning Mar 31 08:03:44 Hi. I'm working on compute module 4 to build the custom OS. I'm using the poky and dunfell branch. I'm not able to get the usb working on it. I have tried adding ENABLE_DWC2_HOST variable in the local.conf and adding the dtoverlay command directly to config.txt file. I was not successfull in doing so. Need some suggestions on this Mar 31 08:20:34 denix: that is a really good question, not sure. It should have figured out that something was providing it and we've not needed that anywhere else Mar 31 08:31:46 hi, is gitsm fetcher utterly broken in terms of full download cache and off line builds? even on master/dunfell? Mar 31 08:35:24 mcfrisk: good question, I know there were issues, not sure of the current status Mar 31 08:41:30 I was hoping master/dunfell would have gitsm fully cached in downlaod cache but doesn't seem to be the case, and fetcher fails if I set BB_NO_NETWORK = "1" Mar 31 08:42:12 mcfrisk: would be good to document how to break it I guess, maybe an open bug Mar 31 08:42:23 Hi, my network interface comes up very late. I switched from sysvinit to systemd, before systemd it was ok... I removed systemd-networkd (bad idea?). For me it looks like, that my udhcpc unit file starts the network. What is the 'correct' way? https://dpaste.org/O9EF Mar 31 08:51:23 RP: stupid questions, why not with gitsm (or any fetcher) just tar the full source tree after do_fetch() and cache and restore that when building off line Mar 31 08:52:17 ah sorry, do_pack/do_patch would be the way then. changes the way SRC_URI's are cached.. Mar 31 08:55:05 caching too much, e.g. multiple copies of git trees, would be better than not being able to cache or off-line-build at all Mar 31 09:01:51 mcfrisk: the system is trying to be clever about caching and updating which breaks down with gitsm. I did think the major issues had been resolved :/ Mar 31 09:04:45 RP: when did being clever ever work? Mar 31 09:06:10 at least for me, there are a lot of failures from gitsm users when I try to build without network. seems like caching doesn't work at all, or at least some commands in setting up the trees only work with the git servers are available. I guess I have to dig deeper Mar 31 10:15:02 zeddii: what logs am i missing ? see what i see at https://pastebin.com/3AFFr4SQ https://pastebin.com/fMS9dakQ - the only information I can see about which scc's are really used is in the logs I added. I did not mean to offend or anything (the kmeta stuff is really nice and I understand a huge amount of work was needed to get there), just I really did not find the info (sorry, I think the frustration thereof made it into my commit tone) Mar 31 10:44:47 Hi! What is the command to open the recipe environment in a separate terminal to be able to configure/build specific package the same way it is done by Bitbake? Mar 31 10:46:20 DanmerZ: -c devshell ? Mar 31 10:46:55 BimBamBim that's it, thanks. I just could not find it in docs. Mar 31 13:05:47 not just yocto-related, but I'm sure you know the answer: Is there a exec-like filter for the "find" program to only display results where an exec is true? Like "find . -type f -exec-like stat {} \| grep Windows \;", should only show M$ related files Mar 31 13:06:40 so anyway, is there a better suited IRC room to ask questions like that one? Mar 31 13:10:35 \quit Mar 31 13:11:24 jmiehe: exec sh -c "your oneline script..." Mar 31 13:11:29 *inline Mar 31 13:17:47 derRichard: so actually the answer to "i want something like -exec" is "just use -exec then" :facepalm: thanks! Mar 31 13:19:38 this is at least what i usually do. i exec some shell script :) Mar 31 13:39:48 need some short syntax help. how to CFLAGS_append only for target, not for -native? Mar 31 13:48:13 CFLAGS-target_append? Mar 31 13:52:12 LetoThe2nd: CFLAGS_append_class-target = " -DLETO" Mar 31 13:52:28 but I think you have TARGET_CFLAGS already? Mar 31 13:52:52 LetoThe2nd: https://git.yoctoproject.org/cgit/cgit.cgi/poky/tree/meta/conf/bitbake.conf#n573 Mar 31 13:53:47 qschulz: you lost me. I just need to pass "-DWHATEVER" to the target build of sqlite. Mar 31 13:54:17 qschulz: so you'd go for TARGET_CFLAGS_append? Mar 31 13:55:45 CFLAGS_append_class-target = " -DWHATEVER" should do the trick, don't know about TARGET_CFLAGS Mar 31 13:57:08 kthx Mar 31 13:57:54 LetoThe2nd: CFLAGS_append_class-target_pn-sqlite if you like local.conf :p Mar 31 13:59:57 muahahaha local.conf Mar 31 14:13:37 ... that moment when you realize that your build broke because you c&p-ed a constant from google, effectively replacing "_" with " " Mar 31 14:16:00 LetoThe2nd: smh, THAT close to remove you from the first responder team Mar 31 14:21:16 hrhrhr Mar 31 14:54:19 <_habnabit> hello i've been told to join because yocto is using eval in places it's not necessary Mar 31 15:01:46 difference between useradd, useradd_base and extrausers? Mar 31 15:03:06 _habnabit: okay? Mar 31 15:03:23 <_habnabit> LetoThe2nd, yeah idk this is just what i was told to do Mar 31 15:04:26 are the authors of the poky/meta/classes/siteinfo.bbclass file here? Mar 31 15:04:37 _habnabit: who told you? what is the context? Mar 31 15:05:51 LetoThe2nd: i believe _habnabit wants to discuss the usage of eval() in siteinfo.bbclass. i suggested he come here because i suspect the authors of that file (siteinfo.bbclass) are on the channel Mar 31 15:05:58 <_habnabit> yates, not a he Mar 31 15:06:13 oh, sorry Mar 31 15:06:22 _habnabit: please formulate your question :) Mar 31 15:06:41 <_habnabit> i need to go to the pharmacy now anyway but i was linked to http://paste.ubuntu.com/p/hvS7Kh45NH/ Mar 31 15:07:52 ok, but what is the question? there isn't even an eval() in that paste, at least not from grepping it. Mar 31 15:09:45 LetoThe2nd: that's true. the eval() is part of the bb.utils.better_eval(call, locs) function Mar 31 15:11:26 i think _habnabit believes using eval() is unnecessary and wants to explain how this functionality (appending to the siteinfo dictionaries) can be done, but i'll let her respond Mar 31 15:11:40 can better be done Mar 31 15:12:58 well is it only about that one usecase? is it about the implementation of https://git.yoctoproject.org/cgit/cgit.cgi/poky/tree/bitbake/lib/bb/utils.py#n421 in general? all effort is welcome, but... this is more like pulling things out of a nose bit by bit. maybe an explanatory/asking mail to the ML would be better? Mar 31 15:13:50 yes, about that function. yes, ML may be better. thanks LetoThe2nd Mar 31 15:14:09 technically this even goes to bitbake, but we're not into nitpicking, so any of yocto@ or openembedded-devel@ would be suitable as a start. Mar 31 15:14:32 good point. hopefully she sees this when she gets back Mar 31 15:16:58 *they see it. that was the requested pronoun, for the record. Mar 31 15:27:35 Does anyone have any experience running nightly builds on a cloud compute server like AWS or linode? How expensive is that? Mar 31 15:33:03 nate02: it depends (TM). if you want to do without sstate (e.g. test the whol build), then you're better off just buying the HW Mar 31 15:38:07 LetoThe2nd: which I assume would be recommended for releases, right? Mar 31 15:40:21 qschulz: there are many "nightly builds" Mar 31 15:41:21 LetoThe2nd: I meant to say for companies when they do releases, to not build from sstate-cache Mar 31 15:42:41 qschulz: it is, but we don't know if the OA wants to do releases. Mar 31 15:42:48 just "nightly builds" :) Mar 31 15:45:16 is "bb.plain()" the right function to use within a python function to "print" to the bitbake output? Mar 31 15:48:43 yates: yes Mar 31 15:49:30 yates: bb.plain, bb.error, bb.warn, bb.fatal, bb.info, bb.debug IIRC Mar 31 15:49:44 bb.warn is nice, prints by default and is colored in orange/yellow Mar 31 15:53:14 qschulz: s/bb.info/bb.note/ iirc Mar 31 15:56:23 I'm here to parade my ignorance and hope someone can point me in a better direction than what I've been able to dredge up via google. Mar 31 15:56:49 I'm looking to make sure that vc4graphics are included and bundled with the image we're building... which at this point, is native Wayland + Weston Mar 31 15:56:50 no X Mar 31 15:57:32 some sources indicate that vc4 graphics are just automagically included for raspberrypi4-64 machines... but if they aren't... then it should be a trivial MACHINE_FEATURES_append Mar 31 15:57:54 so far so good... but I'm not sure I know of way to verify that the graphics drivers are installed on the resulting image Mar 31 15:58:11 the only thing I'm finding, are people suggesting I run glxgears... which I think is x-onl Mar 31 15:58:15 x-only* Mar 31 16:00:07 chrome on wayland has been quite picky, and won't start unless we disable gpu acceleration... so I'm 99% sure it's just an issue on chrome-ozone-wayland on dunfell Mar 31 16:00:55 but trying to verify that the driver is at least _present_ seems like something that shouldn't be hard to check... I just haven't been able to figure what I could be looking for Mar 31 16:01:53 I DO have a /dev/dri/card0 and /dev/dri/card1 which seems to imply that I have a definite something in this space Mar 31 16:07:42 what is packagegroup-machine-base? Mar 31 16:07:48 Spooster: I usually use kmscube for testing GPU support, it comes with libdrm-tools IIRC Mar 31 16:08:39 @jmiehe machine is set to raspberrypi4-64 in my kas.yml Mar 31 16:08:48 i'll take a look at kmscube Mar 31 16:08:52 Spooster: if the vc4 GPU driver is built-in the kernel, you can check on the target by reading the content of /proc/config.gz and check for the option (and test with kmscube for example) Mar 31 16:09:30 Spooster: if it's built as a module, you can check with buildhistory or bitbake -g that kernel-module- exists in your image Mar 31 16:10:25 Spooster: libdrm-tests I think actually Mar 31 16:10:36 I don't seem to have a /proc/config.gz but I'll take a look to see if I can find evidence of kernel-module-vc4-kms-v3d or kernel-module-vc4-fkms-v3d Mar 31 16:11:11 Spooster: if they are built-in, you won't find them (obviously, since they are not modules :) ) Mar 31 16:12:16 Is it better practice to "inherit core-image" or to "require core-image-minimal.bb"? Mar 31 16:12:31 is there any difference? Mar 31 16:13:59 jmiehe: core-image-minimal.bb inherit core-image, so the difference is all the rest Mar 31 16:15:37 jmiehe: the notable addition seems to be packagegroup-base-extended added by default but honestly I'm reading the files right now Mar 31 16:17:01 qschulz: i actually don't see where core-image is even defined Mar 31 16:17:57 wait, it's in classes ofc, my find didnt include bbclass :D Mar 31 16:18:30 Spooster: enabling vc4 is the default with meta-raspberrypi. you have to explicitly disable it if you don't want it (i.e. DISABLE_VC4GRAPHICS) Mar 31 16:19:06 ip a Mar 31 16:19:21 … wrong terminal, sorry Mar 31 16:19:46 tlwoerner: That all lines up with what I was seeing... thanks! Still looking to figure out what chromium's issue is... but I think I'm on a good path for now Mar 31 16:22:02 Spooster: i usually enable buildhistory, which gives me a list list of installed-packages.txt, that i can look through to see what ended up in my image Mar 31 16:22:22 this seems... like something I should read more on Mar 31 16:22:46 i agree with qschulz kmscube is a good tester for accelerated graphics that is independent of windowing system Mar 31 16:23:17 glmark2 with PACKAGECONFIG_append_pn-glmark2 = " drm-gl" is also a good windowing-independent acceleration graphics tester too Mar 31 16:23:43 glmark2 is in meta-openembedded/meta-oe/recipes-benchmark (or benchmarks?) Mar 31 16:24:02 Spooster: add the following to local.conf Mar 31 16:24:09 INHERIT += "buildhistory image-buildinfo buildstats-summary" Mar 31 16:24:14 BUILDHISTORY_COMMIT = "1" Mar 31 16:25:07 then, at the end of your build, check "buildhistory/images///" Mar 31 16:25:25 how much filespace should I expect this to take? Mar 31 16:25:27 and you'll find installed-packages.txt and files-in-image.txt Mar 31 16:25:40 I'm unfortunately in a space where I have to keep things lean Mar 31 16:25:59 $ du -sh buildhistory/ Mar 31 16:26:05 70M buildhistory/ Mar 31 16:26:29 it depends on the number of images, machines, c libraries… Mar 31 16:26:33 ofc Mar 31 16:26:42 i guess you could not enable the BUILDHISTORY_COMMIT Mar 31 16:27:05 but then you lose the ability to use git diff to see what changed between builds Mar 31 16:28:27 I'll have to find some time to get this running in my CI Mar 31 16:31:05 Spooster: https://twoerner.blogspot.com/2020/12/graphics-with-openembeddedyocto-without.html Mar 31 16:34:17 qschulz: thanks! Mar 31 16:35:01 tlwoerner: tyvm Mar 31 16:36:40 Spooster: once you have chromium running you can just open a new tab and use the following URI: "chrome://gpu" Mar 31 16:36:53 yeah, that's ultimately what I'm trying to narrow in on Mar 31 16:36:58 it will show you whether gpu acceleration is enabled/working Mar 31 16:37:03 ah Mar 31 16:37:08 chromium won't open unless I pass `--disable-gpu` Mar 31 16:37:15 so naturally, a lot of things are disabled Mar 31 16:37:18 oops, lol Mar 31 16:37:21 no worries Mar 31 16:38:01 kmscube under the weston user (we're running Waylend/weston as non-root) reveals interesting issues: `Failed to export dumb buffer: Permission denies` Mar 31 16:38:03 denied* Mar 31 16:38:13 too bad chromium didn't have a drm-only backend Mar 31 16:38:13 so I think I have my smoke... time to find the fire Mar 31 16:42:34 kmscube is spitting out so much output, that I need to scroll in my terminal, but I can't seem to capture it's text by piping it to | less Mar 31 16:44:27 Spooster: 2>&1 | less, it's likely going to stderr. Mar 31 16:44:40 no... it's even stranger than that Mar 31 16:44:48 there are 4 messages going to stderr... and I can work with that Mar 31 16:45:02 but the majority of the output is just....... not going to stdout or stderr somehow Mar 31 16:45:05 if you're seeing anythign else, it may be coming from your drivers to the kernel console, check dmesg Mar 31 16:46:29 I'm in my desktop environment, running `kmscube` in a terminal in the environment Mar 31 16:46:56 I get what looks like similar output to when I run it as root... I can ssh into the device as the root user in this debug build... so I might just use that Mar 31 16:47:51 but running as the display user reveals some of my trouble... I'm just confused as to how running `kmscube` produces a page of output, but `kmscube 2>&1 | less` only prints 4 lines Mar 31 16:49:50 dmesg doesn't have what I'm looking for Mar 31 16:58:41 Spooster: the logging in kmscube itself is quite straight-forward. almost all printf() and fprintf() Mar 31 16:59:00 then why on earth is `kmscube | less` empty? Mar 31 16:59:02 maybe there's buffing going on, and the rest of the output only comes with an explicity flush or when the app terminates? Mar 31 16:59:23 it's possible the other output, other than the 4 lines, comes from libraries it links to? Mar 31 17:00:00 and maybe those libraries have some voodoo going on (like detecting the terminal type of STDOUT/STDERR and changing behaviour as a result?) Mar 31 17:00:04 (just guessing) Mar 31 17:00:11 that's my only guess... Mar 31 17:00:28 maybe try redirecting to a file and gracefully exiting the app? Mar 31 17:00:41 empty file Mar 31 17:00:53 the OpenEmbedded Happy Hour is in, roughly, and hour! See you there! :-D Mar 31 17:00:58 https://www.openembedded.org/wiki/Happy_Hours Mar 31 17:01:23 Spooster: (the OEHH thing wasn't directed at you specifically ;-) Mar 31 17:02:12 I mean... I'm not going to say no Mar 31 17:35:48 RP: with RDEPENDS and that warning, it ends up building full set of nativesdk-perl, but then doesn't package it into buildtools-tarball due to RDEPENDS. I was wondering if setting DEPENDS would remove the warning and don't even build nativesdk-perl, since it's not needed Mar 31 17:38:46 sorry, I meant PROVIDES, not DEPENDS - this is for dummy-sdk-package ^ Mar 31 17:47:10 denix: I'm just wondering why we don't see this elsewhere Mar 31 17:53:15 reminder - OE happy hour is in less than 10 minutes! Mar 31 18:13:50 We're having an issue where crt0.o (for baremetal) is being stripped and breaking linking. If we add SYSROOT_INHIBIT_STRIP = '1' that of course fixes it. It appears the symtab section is being stripped off as well as others. I'm surprised a crt*.o would be stripped ever.. Is this standard behavior in Linux? Mar 31 19:11:24 fray: hey, any news on meta-xilinx-standalone hello-world sources ? :-) Mar 31 19:56:47 The OEHH was fun! It was nice having that open in the background while I was fooling around with my builds. Trying to narrow down an issue that's slowly starting to look like a permissions issue in EGL on a 64bit raspberrypi4 image. (dunfell, but we're starting to tweak dunfell with things like the latest wayland/weston recipes from master) Mar 31 19:57:13 At the end of the day, I'm just trying to run `chromium` without --disable-gpu, and now I'm at this point. Mar 31 19:57:31 running kmscube dumps a bunch of text: https://pastebin.com/9qMEDWfP Mar 31 19:58:28 Chrome errors at some point in the bowels of EGL: https://pastebin.com/SpdhJayK Mar 31 19:59:01 these uevents seemed interesting: https://pastebin.com/Ch7Ku32w Mar 31 20:01:49 ooo... just got glmark-drm dumps out some error codes that appear to be egl specific: https://pastebin.com/p61uuZxB Mar 31 20:02:37 just got glmark-drm to dump* Mar 31 21:19:53 How do I override a machine configuration, Do I just put a new .conf file in my layer with the same name? Mar 31 21:23:43 renegade: you could, but that would require the correct order in bblayers.conf to take effect. you'd be better off creating your own machine which includes the other and sets MACHINEOVERRIDES appropriately Mar 31 21:30:14 So in my case I'm trying to make some changes for the raspberrypi4. So I would go and make custom_rpi4.conf and then inside that I would set MACHINEOVERRIDES = "raspberrypi4" Mar 31 21:30:18 Is that correct? Mar 31 21:33:54 or rather MACHINEOVERRIDES = "raspberrypi4:${MACHINE}" Mar 31 21:37:44 hi, i created a local layer to hold some backports and gave it priority 90, i stuck a really old version of thrift in there and it seems to have been preferred over the newer version in meta-oe without a PREFERRED_VERSION_thrift being set, and i don't quite know how to figure out why Mar 31 21:42:44 two questions: 1. am i correct in thinking that only packages, and not entire images, are stored in the sstate cache? and 2. is there any way to cache an entire image in the sstate cache? Mar 31 21:43:47 basically i want to end up with something like two different build nodes, which share an sstate cache but not a build directory, and if i build an image on one of them and then build the same image on the other one, it will be copied out of the sstate instead of recreated Mar 31 21:48:15 jordemort: my understanding is that task results get stored in the sstate, which can be disabled by setting task[nostamp] = "1" Mar 31 21:49:13 (not packages, nor images; and if you have a task that produces an image you should be able to stick it in there and get it dutifully tar'd up; but there may be more nuance here) Mar 31 21:55:39 opello: thanks, i'll see what i can figure out Mar 31 22:03:22 hm, i'm not seeing `nostamp` anywhere obvious in the recipe for my image, but it's definitely being recreated every time Mar 31 22:03:38 (if i remove the "build" directory between builds, that is) Mar 31 22:04:00 whereas the packages are clearly being reused Mar 31 22:06:33 https://docs.yoctoproject.org/ref-manual/migration-1.5.html#images-now-rebuild-only-on-changes-instead-of-every-time says images shouldn't rebuild unnecessarily Mar 31 22:08:42 it's true, if i don't remove the build directory between runs Mar 31 22:09:12 yeah, i don't see 100s of MB of sstate tarballs named for the image either :) Mar 31 22:10:01 i feel like i need to achieve a fuller understanding of how sstate functions Mar 31 22:12:08 i have a feeling the use of the deploy directory for that stuff plays into it somehow? Mar 31 22:12:24 I could not find any documentation / message board post on this:  I have some configurations using dhcp-server and dhcp-client.  I am trying to build on master but it looks like they got moved/ripped out... is there a replacement? Mar 31 22:13:22 jordemort: probably, i know we've used the tmp/deploy/images/ directory explicitly (and scripted the run so that the image is done first, then subsequent repackaging is done) but that always felt messy Mar 31 22:17:26 I guess I will need to change to dhcpcd Mar 31 22:37:04 khem: https://github.com/FFTW/fftw3/issues/213 Mar 31 22:42:04 i think i can pull off what i want to do with --dump-signatures for eacn of the images Mar 31 22:42:18 the output of that seems to be deterministic, with or without an empty build directory Mar 31 22:42:49 so if i do that before a build, and compare it to a previous run, i can skip the build if it's unchanged, without having to keep the build directory around Mar 31 22:51:15 rpcme: dhcpcd is the replacement client, kea the replacement server Mar 31 22:52:25 smurray: thank you I don't know why server was there originally (I didn't write it) but the client is what's needed. Mar 31 22:53:04 rpcme: there's also a dhcp client in busybox that you could try, iirc Mar 31 22:53:35 sounds good.  obv that would be smaller footprint Mar 31 22:55:00 yeah, if you're already using busybox, it's "free" if it works for your usecase Mar 31 22:56:00 also, if you happen to be using systemd, systemd-networkd might be an option Mar 31 23:24:00 Crofton|cloud: it does perhaps look like a bug in gcc9 AFAICT but its not narrowed down enough to look further into gcc9 perhaps some indirect fix in gcc10 has fixed it Mar 31 23:24:38 yeah Mar 31 23:24:54 heisenbug Mar 31 23:25:34 it has motivated me to do more ptest though Mar 31 23:25:49 rpcme: udhcpc is in busybox but won t do ipv6, you might use something like dibbler but perhaps dhcpcd will cover all bases for you Mar 31 23:27:24 khem: thank you.  I will totally consider once I get a working baseline on master Mar 31 23:27:59 if size matters, then udhcpc+dibbler is a good compromize Mar 31 23:30:07 Crofton|cloud: if this is too critical performance wise perhaps consider using older gcc that will pan out ok Mar 31 23:30:28 it is just annoying supporting older gcc in newer layers Mar 31 23:30:32 maybe use meta-arm and IIRC they provided 8.3 in dunfell time frame Mar 31 23:31:35 Crofton|cloud: yeah other way around will be painful newer gcc will induct more misery on to older code Mar 31 23:47:48 RP: I'm not sure why I don't see the warning in poky. but I can confirm that it still ends up building and packaging nativesdk-perl for buildtools-tarball, even though it's not used Mar 31 23:50:37 RP: I saw the warnings in Arago distro, initially for other SDK builds that use dummy-sdk-package, but to make those don't make anything stupid, I tried buildtools-tarball, which is unmodified from oe-core and got the same warninging and behavior Mar 31 23:52:10 RP: btw, tried adding PROVIDES in dummy-sdk-package, but that may need more work, as it then throws duplicate providers errors, obviously... and switching PREFERRED_PROVIDERS may be too much to deal with :) Apr 01 00:17:26 khem, your last change to meta-odroid is causing this: cc1: error: '-Wno-error=array-parameter': no option '-Warray-parameter' Apr 01 00:44:52 armpit: hmm is it on master. ? Apr 01 00:51:31 yes. Apr 01 00:51:57 I plan on leave that on out for hardknott as YP has a gcc 10 version Apr 01 02:08:00 armpit: I think if you change the patch to use. -Wno-error instead will fix it for both gcc10 and 11 Apr 01 02:08:27 I dont have a tree handy so will take some time for me to get there, if you have it handy try the change out Apr 01 02:08:45 change -Wno-error=array-parameter to -Wno-error **** ENDING LOGGING AT Thu Apr 01 02:59:56 2021