**** BEGIN LOGGING AT Thu Mar 30 03:00:02 2017 Mar 30 04:37:00 hmmm... how can i force the rebuild of the u-boot image? Mar 30 04:40:57 done Mar 30 05:36:38 kanavin, we are using internally a form of 'jethro' (a commit from last year), therefore not all patches in your 'fuzz' brranch apply to us Mar 30 05:38:24 so i take recipes one by one and check what applies and not. in the event something from your branch is not applicable to jethro, i will rework it Mar 30 05:39:00 and i want to have this public for obvious reasons Mar 30 05:39:13 others do not have to do it again Mar 30 05:39:16 ... Mar 30 05:59:42 hello, I am not sure if it is possible to use more than one package manager in target. Is it ? Mar 30 05:59:49 I find example in docs: http://www.yoctoproject.org/docs/2.2.1/ref-manual/ref-manual.html#var-PACKAGE_CLASSES Mar 30 06:00:14 Yet, I don't understand if it is possible or not. Mar 30 06:00:52 "The build system uses only the first argument in the list as the package manager when creating your image or SDK. However, packages will be created using any additional packaging classes you specify" ?? Mar 30 06:02:19 hello, I am not sure if it is possible to use more than one package manager in target. Is it ? Mar 30 06:02:29 "The build system uses only the first argument in the list as the package manager when creating your image or SDK. However, packages will be created using any additional packaging classes you specify" ?? Mar 30 06:02:30 from http://www.yoctoproject.org/docs/2.2.1/ref-manual/ref-manual.html#var-PACKAGE_CLASSES Mar 30 06:28:26 I am not sure if it is possible to use more than one package manager in target. Is it ? "The build system uses only the first argument in the list as the package manager when creating your image or SDK. However, packages will be created using any additional packaging classes you specify" ?? Mar 30 06:31:36 ranchu: you asked thee times now in 30 minutes. and i assume you are also the one who asked on the ML. Mar 30 06:31:51 ranchu: now tell me, what for? Mar 30 06:32:23 (besides ross' answer that already was on the ML) Mar 30 06:35:57 LetoThe2nd - Right .... it's me.... Mar 30 06:36:08 ranchu: so what was not clear about the answer? Mar 30 06:36:18 I am just not too sure about the answer Mar 30 06:36:43 Because in doc it gives examples of using more than one. Mar 30 06:37:02 But I got in ML answer that it is not recommended Mar 30 06:37:14 ranchu: yes, because the build process can generate package feeds for more than one package manager. Mar 30 06:37:33 but that what it is. package feeds. your image uses only one. Mar 30 06:38:13 multiple package managers on the same system, trying to manage the same files will inevitably clash, as their databases do not know of each other Mar 30 06:38:19 If it use just one, so what the purpose of giving more than one in the PACKAGE_CLASSES ? Mar 30 06:38:40 the purpose is to be able to generate package feeds. Mar 30 06:39:34 I am a bit confused. If we can use just one, why do we need to generate package feeds (which we cna't use probably ?) ? Mar 30 06:39:58 ranchu: who says that a package feed can be consumed by only one image. Mar 30 06:40:43 ranchu: imagine creating a package feed with bitbake world, and providing two images that use it. one ipk based, one rpm based. Mar 30 06:40:52 for example. Mar 30 06:42:34 I think I start to understand. PACKAGE_CLASSES can be used for more than one image ? Mar 30 06:43:06 its just another take at the POV. usually we are image-focused here, so we build exactly what needs to go into one image. Mar 30 06:43:43 package feeds on the other hand are distribution focused. you generate everything that anybody could need, and then pick what suits your problem Mar 30 06:43:47 OK, so usually we will only use one package manager in PACKAGE_CLASSES Mar 30 06:43:56 ranchu: in about 99.9% of cases, yes. Mar 30 06:44:19 OK, Thanks a lot. I think I finally got it :) Mar 30 06:45:31 ranchu: look at the angstrom distribution. the point is that as long as the abi is stable, the package feed can be used across multiple images. Mar 30 06:45:36 *sigh* Mar 30 06:46:18 * cornel doesn't understand much of this technology Mar 30 06:46:58 its not that complicated, its just understanding what a package feed, a distribution, and an image are. Mar 30 06:50:57 so a package feed is a kinda package repository? for example a directory with packages? Mar 30 06:53:29 this document http://www.intel.com/content/dam/www/public/us/en/documents/white-papers/package-manager-white-paper.pdf says "tmp/deply" :) Mar 30 06:53:47 cornel: a package feed is something that can (with more or less effort needed) be used as a package repository. you know, that magic server thing that all people think of as "a distribution" and think that when they have runtime package management will magically solve all their problems. Mar 30 06:53:59 not sure if it's woorth fixing it, though Mar 30 06:54:20 LetoThe2nd, thank you Mar 30 06:54:36 picture is clearer now Mar 30 06:54:55 the document is outdated anyways, with kanavin's latest smart->dnf transition Mar 30 06:56:07 cornel: basically its that runtime package management as people know it is pretty useless without an upstream package repository. and the feeds are what goes into such a repository. Mar 30 06:56:55 roger that Mar 30 06:57:18 just another reason why i think runtime package management is completely overrated. Mar 30 06:58:28 LetoThe2nd, what's you ideal way of managing packages in a system already installed? Mar 30 06:59:04 cornel: why would i want to manage packages there anyways? Mar 30 06:59:28 for example to fix a bug Mar 30 06:59:49 (always talking about an embedded device, as that is the focus here. desktop/server is a completely different beast, of courese.) Mar 30 07:00:01 you say that it's simpler to just generate an uptodate image? Mar 30 07:00:24 LetoThe2nd, right, let's say a device that manages a car Mar 30 07:00:30 and you have a fix for it Mar 30 07:01:02 we always do complete image updates. so we know that the system is in a consistent state. Mar 30 07:01:09 ok Mar 30 07:01:40 i wonder how can this be done over the air, or if it is possible at all Mar 30 07:02:02 usually the volatile data is already enough to give us headaches, why should we want to have even more fun with dependencies, possibly failing packages, etc.? Mar 30 07:02:15 i see your point Mar 30 07:02:39 cornel: well we use swupdate by sbabic. a option for OTA could be mender.io AFAIK Mar 30 07:03:13 plus, in many many applications you do not have OTA. and you do not want updates that you did not trigger manually. Mar 30 07:04:34 so imagine the fun of a service tech standing in front of the device, plugging in a thumb drive containing XYZ packages, and then kicking off the package manager. just to find out at 95% percent that the update fails because something is missing, corrupted, out of disk space. Mar 30 07:05:13 if he has only one binary blob, those things go away. theres still enough to go wrong. Mar 30 07:06:46 LetoThe2nd, for a hands on update, i agree Mar 30 07:07:09 the problem i see when doing this ota is that it can also fail, and it will no longer be just one package Mar 30 07:07:20 so you have to have a backup Mar 30 07:07:29 and this means duplicate storage space Mar 30 07:07:33 see mender.io :) Mar 30 07:07:47 i don't know what mender.io is doing Mar 30 07:07:57 i've seen some mentions of a server ,though Mar 30 07:08:30 ok: Image-based updates using a dual A/B rootfs partition layout with rollback support Mar 30 07:08:33 it does exactly what you just said. duplicate rom space and use that for OAT updates. Mar 30 07:08:42 yep Mar 30 07:08:56 but in automotive this means increased cost, right? Mar 30 07:09:06 it means increased cost everywhere. Mar 30 07:09:07 i mean significantly increased Mar 30 07:09:10 sure Mar 30 07:09:12 now that depends. Mar 30 07:09:23 but automotive tends to be a little more expensive :) Mar 30 07:09:47 but in an ideal world, yes, i woud go for this solution Mar 30 07:09:51 i don't know what the automotive guys are doing, i'm more like an industrial guy. Mar 30 07:10:28 and our customers are more like "OTA? go away. we need something that works hands on. and only hands on." Mar 30 07:10:39 my understanding is that automotive things are more expensive because the various requirements (security, integrity,etc) are stronger Mar 30 07:10:52 that's beside the usual greed and cartel practices, of course Mar 30 07:11:21 automative certainly has a lot of checks and validation in place. and yes, those do come with a price Mar 30 07:11:43 ability to do large temperature swings is a cost too. Mar 30 07:12:24 many things cost money. Mar 30 07:12:24 right Mar 30 07:13:01 I just wish I used my iphone to log into my car and set it up. Mar 30 07:13:42 can you use a computer for this? Mar 30 07:13:47 my 29K car should come with a handheld device that does that. Mar 30 07:14:00 so true Mar 30 07:14:23 what i wish is there's a way for these cars to cutoff network connection to external world Mar 30 07:14:33 yes, a computer can be used...but the engineers do not choose to expose anything useful. Mar 30 07:14:51 you would have to give me car and iphone for free so i would accept that. Mar 30 07:15:05 it's now the year 2017. And if I leave my headlights on it kills the battery. Mar 30 07:15:26 that's fucking stone age. Mar 30 07:15:33 :/ Mar 30 07:16:24 well mine does just beep. stone age technology that works. Mar 30 07:16:45 it will beep for hours whilst the battery dies. Mar 30 07:16:52 useless. Mar 30 07:17:27 it will beep the moment i pull out the key and open the door. i cannot leave the car without noticing it beeping. so no hours involved ;-) Mar 30 07:17:41 it can't alert your smartphone. Mar 30 07:17:48 which smartphone? Mar 30 07:17:56 :) Mar 30 07:18:09 the one in your pocket. Mar 30 07:18:29 Snert: would you please show or enlighten me which one you mean? Mar 30 07:18:54 iphone or android or whatever your device is. Mar 30 07:19:18 or the car should come with a pocketable smartdevice for $29K it costs. Mar 30 07:19:32 nokia something, 15 euros from the supermarket Mar 30 07:20:19 the car itself cost 29K. for that price it should at least come with a device. Mar 30 07:20:23 my ideal kinda phone Mar 30 07:20:26 and you're free to not buy a 29k car if the offer does not fit your expectations. Mar 30 07:20:40 instead of complaining afterwards. Mar 30 07:21:32 the market would be huge if the engineers but thought about it. Mar 30 07:21:43 sell all kinds of stuff to the car owner. Mar 30 07:21:48 manulas...parts. Mar 30 07:21:55 it is true though that the new smart tech is largely marketing, not really doing what you expect from it Mar 30 07:22:18 I know a lot of people in automotive. and believe me, there a lot of thought going into what they sell, and what they do not sell. Mar 30 07:22:29 this is absolutely non-trivial Mar 30 07:22:47 yes. But it takes a company like apple to do a good car/computer/human interface. Mar 30 07:23:15 apple needs to do the AppleDash. Mar 30 07:23:29 get in bed with detroit and take us forward. Mar 30 07:23:37 it takes a company like apple to make a user interface that meets the expectations of people already using apple. Mar 30 07:23:48 * LetoThe2nd is off for some coffee, nevertheless Mar 30 07:23:54 it can include android stuff too. Mar 30 08:37:43 cornel: the 'fuzz' branch is maybe 15 percent complete, so backporting it to jethro now is totally wasteful and pointless - instead I would appreciate if you could help us complete the work for the master branch Mar 30 08:49:50 RP: so how does one delete files at package time (rather than install time)? Is there a ready-to-copypaste example in recipes somewhere? Mar 30 09:01:16 rburton, kanavin: do we record the actual postinst intercept commands anywhere? That update_gio_module_cache discussion on list looks a bit worrying Mar 30 09:01:31 I thought we did but probably gave wrong advice... Mar 30 09:02:32 kanavin, what i plan to do is: backport to jethro what is already done (because we are working on jethro) and then look at what is not already fixed Mar 30 09:03:03 this is partly because right now i don't know of any proper procedure to fix a patch Mar 30 09:03:24 cornel: you are aware that most of the fixes are probably cosmetic Mar 30 09:03:30 if i would do, i would gladly do it first for morty and only later for jethro Mar 30 09:03:37 jku, sure Mar 30 09:05:45 RP: so /usr/bin/env usage is because SDK uses it to re-route to host tools? Mar 30 09:05:58 I knew i must be missing something Mar 30 09:06:23 can I read about this somewhere or is that all I need to know? Mar 30 09:09:43 cornel: I still think it's not a good use of your time. why do you want this done? Mar 30 09:10:15 jku: let me see the discussion - not feeling very well today :( Mar 30 09:11:03 kanavin, we are working on jethro, so we need the fixes there. but as i've said, i would gladly do it first for your branch but i do not know how to correctly create the fixed patches Mar 30 09:11:26 i can do things like diff -Nupr but git format-patch is foreign language to me Mar 30 09:11:59 cornel: why do you need the fixes? Mar 30 09:12:21 kanavin, we want to create more deterministic releases :) Mar 30 09:12:43 we've seen cases where patches were applied wrongly but it appeard to be ok Mar 30 09:13:14 what we've seen had no impact (those that allowed also do_compile) Mar 30 09:13:30 but we wonder if there are any other patches that do have impact Mar 30 09:14:32 cornel: it's a very tedious job, I'd suggest you wait for the tool support Mar 30 09:15:49 cornel: you'd be fixing some 80 failing recipes, and I fear for your sanity Mar 30 09:16:00 :) Mar 30 09:16:08 kanavin,ok Mar 30 09:16:20 kanavin, do we have an eta for the tool support? Mar 30 09:16:31 cornel: I would rather prefer you fix it in master - that would truly benefit everyone, but not straight away : Mar 30 09:17:33 cornel: paul eggleton has been silent about this for several weeks - I suggest you leave a comment in bug 8325 Mar 30 09:17:34 Bug https://bugzilla.yoctoproject.org/show_bug.cgi?id=8325 enhancement, Medium, 2.3, paul.eggleton, NEW , devtool modify could handle patch failure more usefully Mar 30 09:17:56 cornel: I'm not doing any further work on this, until there is a tool :) Mar 30 09:20:26 ok, thank you kanavin Mar 30 09:26:19 jku: I read the discussion, it looks to me that the real issue is that gio-querymodules is failing when it shouldn't Mar 30 09:26:36 jku: and I still don't see why chown would be at all necessary Mar 30 09:33:37 kanavin: that bug is a medium enhancement, if it's blocking work it probably needs its priority adjusting? Mar 30 09:36:37 RP: long story short, I believe Chris's patch is ok, but Peter's is not (as it breaks first boot deferral Mar 30 09:37:00 joshuagl: I guess it does yeah Mar 30 09:38:56 kanavin: did you get an answer to you package specific deletion? Mar 30 09:39:09 kanavin: highlighting here is acting odd this morning :/ Mar 30 09:39:20 jku: I think that is mostly what you need to know Mar 30 09:39:43 jku: seriously tempted just to patch those files to use /usr/bin/env perl and add "use warnings;" to the script Mar 30 09:40:18 kanavin: how does Peter's break first boot deferral? hides exit code? Mar 30 09:41:53 RP: yes Mar 30 09:42:46 RP: ugh, ambigous 'yes :) I did not get an answer about package specific deletion, but peter's patch does hide exit code :) Mar 30 09:44:18 kanavin: the expression to grep for is PACKAGE_PREPROCESS_FUNCS, e.g. the horrors in binconfig.bbclass Mar 30 09:44:39 RP: thanks :) Mar 30 10:09:22 ant_work: around? Mar 30 10:09:54 ant_work: I'd like to confirm whether without that patch, the postinst is still deferred to first boot or not Mar 30 10:10:09 hello Mar 30 10:11:21 without this patch there aren't warnings so I suppose no, it's not deferred Mar 30 10:12:33 ant_work: please don't suppose. It wouldn't necessarily show warnings Mar 30 10:13:59 what I did see, was the string ##libglib-2-0-0 in one of the postinst files Mar 30 10:14:21 ant_work: have you a do_rootfs log with the patch reverted you could share? Mar 30 10:14:57 not now, I can give one in a few hours Mar 30 10:15:18 ant_work: thats fine, thanks. I just want to make sure we understand what is going on (I think I might have some idea now) Mar 30 10:19:25 Apropos postinsts, I was trying to help with http://stackoverflow.com/questions/43067776/linux-capabilities-with-yocto/43109887#43109887 this morning, which is about setcap in postinst. Mar 30 10:19:57 I got stuck with ensuring that "setcap" is actually available when running the postinst during image construction. Mar 30 10:20:22 RDEPENDS_${PN} = "libcap" ensures that setcap is on the device. Mar 30 10:20:40 But that doesn't install it into recipe-sysroot-native, does it? Mar 30 10:20:50 RP: ^^^ Mar 30 10:20:56 pohly: no, it doesn't Mar 30 10:20:59 ant_work: "the string ##libglib-2-0-0 in one of the postinst files" is not an indication of first boot deferral at all - you need to inspect the log.do_rootfs to see if it happened or not Mar 30 10:21:09 RP: ^ Mar 30 10:21:30 RP: does one have to use a DEPENDS = "libcap-native" then instead? Mar 30 10:21:51 pohly: I wrote about this in the RSS emails, I don't remember offhand :/ Mar 30 10:22:00 Same here ;-} Mar 30 10:22:20 that string merely means that one of the recipes requested that script to be executed during image build. If the script fails, then first boot deferral happens. So we need to ensure the failing is not swept under carpet with various patches. Mar 30 10:22:36 (such as Peter's) Mar 30 10:22:43 pohly: I added a variable iirc Mar 30 10:23:02 pohly: PACKAGE_WRITE_DEPS Mar 30 10:23:16 Ah, now I remember. Mar 30 10:25:06 kanavin: it would be nice to collect some of this better in the logs... Mar 30 10:26:01 RP: I filed a bug for fixing the entire postinst scripts mess, you told me to defer that work to 2.4 ;) Mar 30 10:26:12 pohly: thanks for the rm_work fix btw, nicely done :) Mar 30 10:26:55 kanavin: I'd like to fix the logging and wasn't too keen on some of the other changes iirc Mar 30 10:27:26 RP: it's still only a workaround, though. Long-term it would be nice to have ordering dependencies in addition to hard dependencies, for cases like this. Mar 30 10:27:49 But yeah, for now it has to do ;-} Mar 30 10:28:20 pohly: Having spent some number of years working with dependencies, I don't think they divide into those two types as easily as you think Mar 30 10:28:41 Hi - why does the gstreamer1.0-omx recipe have a commercial license flag? Mar 30 10:29:12 RP: I can imagine that sometimes it is one, sometimes the other, and sometimes both. I think that's how systemd works. Mar 30 10:29:19 Anyway, definitely not trivial. Mar 30 10:29:39 pohly: right. Wish I had the time to think about that properly... Mar 30 10:30:04 Yes. Not now. Mar 30 10:30:34 RP: but we do need to robustly detect script failures, and clearly separate them from intented first boot deferrals - neither is happening right now Mar 30 10:31:14 and yes better logging for what is happening Mar 30 11:15:55 Hello there, we have a HPC cluster with 1280 cores. Is it possible to run bitbake using all this cores to speed up the build time ? Mar 30 11:16:26 Does bitbake have this feature or could it be implemented ? Mar 30 11:17:54 ilkmc2r: the system will automatically set BB_NUMBER_THREADS and PARALLEL_MAKE to take advantage of multiple cores. The bigger problem is the build is not able to parallelise in places (e.g. the toolchain is needed to cross compile anything) Mar 30 11:18:14 ilkmc2r: so the system will make best use of that is available but its not going to see 100% usage Mar 30 11:19:36 RP: But, our cluster was built by using 8 cores 16GB rams of blades, So if I run it on single blade, it will do 8 task at a time. I mean can we use whole HPC cluster, by using a job scheduler etc.. Mar 30 11:20:18 ilkmc2r: we don't support distributed builds currently, no Mar 30 11:20:55 ilkmc2r: its a nice idea but in reality we can't even fully use a big system like my 88 core machine so its not been a priority. There is an open bug for it Mar 30 11:20:57 ilkmc2r: Is this on your roadmap or is it Mar 30 11:21:35 RP: Is this on your roadmap ? Mar 30 11:21:36 ilkmc2r: we'd certainly like to do it but there have been other higher priority things. We're working on deciding what might go into 2.4 atm Mar 30 11:22:04 ilkmc2r: it really needs some people interested to work on it... Mar 30 11:22:33 RP: Well, we will try to implement it, and If we go success we will share it with you :) Mar 30 11:23:13 ilkmc2r: can I ask who "we" is? Mar 30 11:23:50 RP: By saying "we", I mean 12 people of an university's software development team. Near East University. Mar 30 11:24:02 ilkmc2r: Bitbake already has a strong "cooker/server" and worker abstraction where bitbake-worker does all the execution and talks over a pipe to the server Mar 30 11:24:27 ilkmc2r: hopefully that means some of the hard work has already been done Mar 30 11:24:45 ilkmc2r: in multiconfig builds we already support multiple worker execution too Mar 30 11:25:28 ilkmc2r: there's a icecc class but it's a bit tricky to keep it running smoothly (the class but also icecc itself). A beefier build host felt like a much better use of my time than dealing with weird issues Mar 30 11:26:07 ilkmc2r: please do talk to us as you do it rather that do this in isolation as whilst we've not implemented it, we do have ideas that could likely help (like the above abstractions) Mar 30 11:27:54 jku: icecc also only helps compile jobs which is about 20% of our build time Mar 30 11:28:28 right Mar 30 11:30:49 jku: o/ Mar 30 11:32:02 RP: I will discuss this with my team members and will keep in touch with you. And started to learning about cooker/server mechanism Mar 30 11:32:22 hey phako[m] Mar 30 11:37:49 so how do I build a nativesdk quilt ? Mar 30 11:40:49 nativesdk-quilt? Mar 30 11:42:14 rburton: it doesn't seem to inherit or extend anything related to nativesdk ... is it really part of nativesdk? Mar 30 11:42:20 hm Mar 30 11:42:27 jku: perhaps there isn't a nativesdk version of it :/ Mar 30 11:42:30 then the problem is hypothetical :) Mar 30 11:42:47 jku: I know we've had this issue with other nativesdk recipes... Mar 30 11:43:28 yeah this finally explains the weird obsession with env that I had been wondering about :) Mar 30 11:51:42 RP: so I assume you would still rather have a quilt patch that uses "${USRBINPATH}/env perl" in shebang ? Mar 30 11:59:24 question on TUNE-FEATURES: why does TUNE_FEATURES_tune-aarch64 = "aarch64" not include 'neon' and 'vfpv4' Mar 30 12:00:03 some packages check the features for neon and on aarch64 they don't catch it then Mar 30 12:00:11 e.g. libpng Mar 30 12:32:49 jku: I think its a better example to set Mar 30 12:33:09 jku: at some point we need to unify quilt and quilt-native and this should help with that too Mar 30 12:53:01 rburton: can you give this to autobuilder again? I fixed a few long hanging fruits there, so would like to get a build which is less cluttered with obvious issues :) https://git.yoctoproject.org/cgit/cgit.cgi/poky-contrib/log/?h=akanavin/openssl-1.1 Mar 30 12:55:36 kanavin: fired on yocto.io Mar 30 12:56:39 yocto.io ? Mar 30 12:57:19 rburton: thanks, time for nap I guess :) Mar 30 12:57:34 CTtpollard: second autobuilder cluster Mar 30 12:59:41 we're on trend with a .io domain ;) Mar 30 12:59:49 mkay... so, Have we really no way in 2017 to create a bootable filesystem image that A.) is not vfat and B.) does not require root ? Mar 30 13:00:20 .io is def for the cool kids Mar 30 13:00:22 I wondered if yocto had gone into the UX business for a second Mar 30 13:00:23 Ummm, scratch that, there is with libguestfs tools, but that is a huge amount of stuff Mar 30 13:00:43 .org is so 2000s Mar 30 13:00:43 .build is better Mar 30 13:01:53 .yocto , now only $185000 Mar 30 13:02:21 what, for a whole tld? bargain Mar 30 13:03:25 let me fire my question again: Does anyone know why gstreamre1.0-omx has LICENSE_FLAGS="commercial" ? because it's ususally not usable without the vendor OMX libs? Mar 30 13:04:12 phako[m]: probably worth asking whoever wrote the original recipe. really should be a comment to explain… Mar 30 13:05:09 phako[m]: you might be right, that could be a mistake Mar 30 13:07:12 phako[m]: does 'git blame' reveal anything? Mar 30 13:07:15 dv Mar 30 13:07:22 author Carlos Rafael Giani Mar 30 13:07:28 http://cgit.openembedded.org/openembedded-core/commit/meta/recipes-multimedia/gstreamer/gstreamer1.0-omx.inc?h=morty&id=e200e61529fa5a78354cde67442e90ace4b3857c Mar 30 13:08:07 ah, no, inherited from the 0.10 plugin Mar 30 13:08:36 http://cgit.openembedded.org/openembedded-core/commit/meta/recipes-multimedia/gstreamer/gst-openmax_0.10.1.bb?h=morty&id=43410523a07d9eb52a7d57ae3dc1cc320cbbc6f9 Mar 30 13:15:43 phako[m]: https://bugzilla.yoctoproject.org/show_bug.cgi?id=1262 Mar 30 13:15:44 Bug 1262: normal, Medium, 1.1, saul.wold, RESOLVED FIXED, Default build of world should not fail due to COMMERCIAL_LICENSE issues Mar 30 13:16:47 maybe omx recipes is no longer dependent on 'commercial' recpes, and so the line can be dropped Mar 30 13:22:57 phako[m]: oe-core c69453fe94a649c518b0e6d79616f05579b864ce Mar 30 13:23:23 looks like it can be removed Mar 30 13:23:53 get a patch on the list asap as 2.3 is freezing Mar 30 14:06:16 I was at ELC this year and took the Yocto Advanced workshop. Mar 30 14:06:48 I'm upgrading my Jetrho to Morty, attempting to anyway, and have a question about the devshell changes Mar 30 14:07:40 We used to be able to 'bitbake -c devshell u-boot', then edit files and, within devshell, '../temp/run.do_compile' to try the build interactively. Mar 30 14:09:59 Now, when we use devtool, how does that equate to what I just outlined? Is there some step-by-step example I can study? The Yocto Documentation is a bit too abstract about his this topic and I'm looking for a more concrete example of a devtool process. Mar 30 14:11:20 e.g. 'devtool -c configure u-boot' == 'bitbake -c configure u-boot', or, 'bitbake -c compile xxxx', seems devtool has little granularity? Mar 30 14:15:57 rburton: I'll check Mar 30 14:18:26 T0mW: once devtool has set up the tree, you can build either with 'devtool buid' or with usual bitbake commands Mar 30 14:19:15 shit, i need to look at the gio module intercept issue asap this morning, i had to set aside work and watch matthew last night so didn't get time to look at it :\ Mar 30 14:19:49 kergoth: like 'devtool -c configure blah' ? Mar 30 14:20:13 or just use 'bitbake -c configure blah' and it will go to the workspace Mar 30 14:21:17 kergoth: yeah, I pushed a commit yesterday that borked a customers code, I just remembered that I've got to fix that myself. :/ Mar 30 14:21:38 ls Mar 30 14:21:57 i hate it when i submit something that has been working for ages for us and it explodes and breaks stuff for other people. not the first time, won't be the last, i'm sure Mar 30 14:22:06 still quite irritating all around Mar 30 14:22:34 i did a quickie fix and didn't test it, arrogance on my part "my code doesn't stink" Mar 30 14:22:49 happens to us all :) Mar 30 14:29:44 how do I find out what pulls in virtual/kernel? I want a chroot image, so no kernel needed Mar 30 14:32:17 is it always on, and I should select linux-dummy? Mar 30 14:33:55 yes Mar 30 15:48:47 bitbake-diffsigs tells me my do_configure hash changed, but my bb has do_configure[noexec] = "1" Mar 30 15:49:07 how can I find the culprit? Mar 30 15:51:22 most likely it's a task that your do_configure depends on that changed Mar 30 16:23:00 https://twitter.com/siemens_press/status/847482515388600320 Mar 30 16:23:11 it's official, the acquisition of mentor is complete Mar 30 16:23:38 is this where you sell all your new stock and retire to an island somewhere? Mar 30 16:24:10 hah, it's not like mentor was private or something. Mar 30 16:47:55 anyone experienced with tegra-t18x and boot2qt? Mar 30 16:48:03 I can't find the nvidia-layer/meta-tegra layer Mar 30 16:48:18 have to assume it is proprietary Mar 30 16:49:02 tmpsantos: this is the most upto date I can find Solarwinds Application Users Mailing List Mar 30 16:49:11 erm, wrong buffer.... Mar 30 16:49:20 https://layers.openembedded.org/layerindex/branch/master/layer/meta-tegra/ tmpsantos Mar 30 16:51:23 CTtpollard: I found that one too, but doesn't seem to be the one boot2qt is expecting Mar 30 16:51:59 tmpsantos: is there public documentation for what you're following? Mar 30 16:52:46 CTtpollard: https://doc.qt.io/QtForDeviceCreation/ Mar 30 16:58:22 tmpsantos: what happens when you export the machine name & run the init script? Does it not bring in the necessary layers? Mar 30 16:58:57 CTtpollard: I think I found the issue, it requires a layer marked as "internal" Mar 30 16:59:47 points to... Mar 30 18:32:08 anyone seen this before I start investigating? Just appeared recently ; seems like x86-64 sysroot cruft contaminating my arm64 target binaries... Mar 30 18:32:12 | ERROR: oe_runmake failed Mar 30 18:32:12 | /home/paul/poky/build-arm64/tmp/work/aarch64-overc-linux/libsolv/0.6.26-r0/recipe-sysroot-native/usr/lib/librpmdb.so: error adding symbols: File in wrong format Mar 30 18:32:12 | collect2: error: ld returned 1 exit status Mar 30 18:44:45 so - in a devshell, if I simply remove the reference to that x86-64 librpmdb.so, then the link completes w/o error. Mar 30 18:45:24 shouldn't be looking for libs in that sysroot at all Mar 30 18:45:47 agreed ...appears the turds to link are crafted up in some cesspool of Cmake hell Mar 30 18:45:59 putting on my hip waders and rubber gloves now. Mar 30 18:46:31 good luck :) Mar 30 18:50:57 ha, normally I try -c cleansstate right after I try "up-arrow + enter" for build fails. Since this looked like host contamination, I didn't immediately go there... Mar 30 18:51:11 yet -c cleansstate fixed it. Go figure. :-/ Mar 30 18:51:21 lovely Mar 30 18:51:29 ya, no kidding. Mar 30 18:52:28 that said, I'm glad to have been spared the cmake drama. Mar 30 21:37:52 rburton: pohly's net-tools patch breaks builds... Mar 30 21:52:59 RP: fix is simple I think. added a return to the shell function and it worked fine. Mar 30 21:53:29 stephano: it needs a ":" Mar 30 21:53:40 stephano: but yes, same idea Mar 30 21:54:49 RP: ah right, the mythical ":". so much better than "true". Mar 30 21:57:16 stephano: its convention in our metadata by now... Mar 30 21:57:33 RP: makes sense Mar 30 21:58:31 stephano: I've just tweaked the patch and updated -next Mar 30 21:58:45 hopefully a rebuild of -next should be faster on hot sstate... Mar 30 21:58:58 Sounds good. I'll keep an eye on it. Mar 30 22:19:47 There is definitely some kind of bottleneck going on with our builds, far too much "bursting" activity then stalling :/ Mar 30 22:19:57 at least on large numbers of cores Mar 30 22:21:07 * paulg still sets jobs/packages manually. Mar 30 22:28:23 paulg, zeddii_home: Is https://pastebin.com/p8Jp8aqF a known kernel issue? Mar 30 22:28:30 seems I keep getting this on rebuilds :/ Mar 30 22:30:49 RP, have not seen that ; have not seen zeddii_home mention it either -- have a reproducer? Mar 30 22:31:52 nope. haven’t seen it either, and I do a lot of rebuilds. but yah, I’ll fix it if I can make it happen here. Mar 30 22:31:54 paulg: not yet but its the second time today, I cleaned it last time... Mar 30 22:32:23 I’m heading out now. but will try a few rebuilds, haven’t tweaked anything in the kernel build at all lately. Mar 30 22:32:27 once could be my local crazy builds, twice suggests something lurking :/ Mar 30 22:32:33 yup Mar 30 22:33:36 pretty sure zeddii_home does rebuilds in the same recycled build dir day after day as I do, so we _should_ see this if it is an issue. Mar 30 22:34:08 that said, I'm building linux-yocto-dev mostly... Mar 30 22:35:04 stephano, RP: just as an FYI, colon as no-op isn't solely an oe convention. http://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html#colon - "This utility shall only expand command arguments. It is used when a command is needed, as in the then condition of an if command, but nothing is to be done by the command." Mar 30 22:36:01 afk Mar 30 22:36:23 I’ll fire up some builds and see if I can make it happen .. but that will be later. have to head out for volleyball now. Mar 30 22:37:21 | 1490912493.7192538: mkdir: cannot create directory ‘.’: File exists Mar 30 22:37:32 sounds like a variable expanded to "" Mar 30 22:37:50 paulg: or the fence post, etc, blowing up. i.e. we cleaned up, and it didn’t expect to be cleaned up. Mar 30 22:38:18 my git pull of pky just died Mar 30 22:38:26 > git pull --rebase Mar 30 22:38:27 fatal: unable to connect to git.pokylinux.org: Mar 30 22:38:28 git.pokylinux.org: Temporary failure in name resolution Mar 30 22:38:36 nice. Mar 30 22:38:38 I guess I wasn’t meant to do a build now before I leave :D Mar 30 22:38:56 looks like all my nameservers are down. odd. Mar 30 22:38:56 walk away and enjoy volleyball. :) Mar 30 22:39:26 I just hope it sorts itself out. I don’t want to drive into the office just for that. Mar 30 22:39:38 kergoth: interesting. that's a good resource, thanks. Mar 30 22:40:08 np. yeah, it is, i often look up bits there to know what's portable shell Mar 30 22:40:42 http://pubs.opengroup.org/onlinepubs/9699919799/ being the main index Mar 30 22:41:39 kergoth: thanks :) Mar 30 22:42:12 zeddii_home, paulg: I probably should sleep sometime soon and let the rest of this build finish. I'll poke at it more tomorrow, at least I know its not a known issue... Mar 30 22:42:48 yah. no worries. if it happens, just yell. I’ll take care of it if it is broken. Mar 30 22:43:02 I started a build, will see if I can make it pop up. Mar 30 22:44:01 yah, if you find a solid reproducer, you know where to find us. :) Mar 30 22:46:00 paulg, zeddii_home: https://pastebin.com/LMfJa6iN is the order the tasks ran in this build. So it started again from the checkout on an existing build Mar 30 22:47:23 yah. that’s my only mode of operation, so I have a good chance at seeing it. Mar 30 22:47:40 if you have any fancy stuff .. and by fancy, I mean rm_work, etc, I’d have to add it to my local.conf Mar 30 22:48:42 Guess I should start a build w/o virtual provider kernel being dev. Mar 30 22:49:04 zeddii_home: depends how you define fancy. I did parallelise do_package_write_ipk. I don't think there is anything odd that would cause this though Mar 30 22:49:05 I'm assuming that is closer to what RP is doing.... Mar 30 22:49:18 mine is now compiling, so I got futher. will trigger another build later. Mar 30 22:49:43 * zeddii_home really goes, before he has to get a speeding ticket Mar 30 22:50:49 can't have zeddii_home being zeddii_homeless :) Mar 30 22:52:07 zeddii_home: have fun! Mar 30 22:55:56 paulg, zeddii_home: good news is it reproduces Mar 30 22:57:01 bitbake linux-yocto; ; bitbake linux-yocto -c compile -f; bitbake linux-yocto; bitbake linux-yocto -c kernel_checkout -f; bitbake linux-yocto Mar 30 22:57:39 mildly convoluted but... :). This was with MACHINE=qemux86-64 Mar 30 22:59:05 RP, without any PREFERRED_VERSION_linux-yocto lines in local.conf or similar? Mar 30 23:00:30 paulg: correct. I am on master-next so http://git.yoctoproject.org/cgit.cgi/poky/commit/?h=master-next&id=98efb6b7e430967cfc141149b9d5eeff423c6a91 is in there but shouldn't have any effect on this Mar 30 23:00:53 well, I'm on a local branch of craziness based on the above Mar 30 23:01:02 but nothing touching the kernel tooling or kernel Mar 30 23:01:50 oh, hmm. That may not be true Mar 30 23:02:29 paulg, zeddii_home: I'm going to guess this breaks it: http://git.yoctoproject.org/cgit.cgi/poky-contrib/commit/?h=rpurdie/wip-rss2&id=362e261e1a54f762bac762ac80030d0684c058dd Mar 30 23:02:37 * paulg plays Ozzie crazy train in his mind. Mar 30 23:02:50 but only on rebuilds Mar 30 23:03:28 In my defence, that error is not obvious ;-) Mar 30 23:06:46 RP, what specifically was the use case that the above fixed? I'm left wondering that after reading the commit log. Mar 30 23:07:49 paulg: I had a version of patchelf that created interesting binaries. The patchelf binaries refer to this as a bug in the kernel and I couldn't resist fixing it Mar 30 23:08:44 paulg: https://github.com/NixOS/patchelf/blob/master/BUGS Mar 30 23:09:26 paulg: in the end I had to use a different solution as waiting for everyone to fix their kernels wasn't going to work for me Mar 30 23:10:10 paulg: http://git.yoctoproject.org/cgit.cgi/poky/commit/?id=c4901328fe5cf912c0965e5b011b64a95a9bcb9d has more info Mar 30 23:10:58 paulg: Still can't decide whether to send that upstream or not (the kernel fix( Mar 30 23:11:19 * paulg returns and reads Mar 30 23:17:57 * RP -> Zzzz Mar 30 23:18:19 RP, yah, we'll continue this tomorrow. Mar 31 00:38:22 RP: is there anyway to always delete a file during do_install if found ? Mar 31 00:39:14 RP: on musl charset.alias gets generated quite often due to a gnulib bug but its spurious file Mar 31 00:39:44 fixing gnulib and propagating it to all relevant packages will take some time since gnulib is shared in source Mar 31 00:40:48 something like a global do_install_append() Mar 31 00:57:51 khem: could you do a do_install_append() or a do_install postfunc in tclibc-musl.inc ? Mar 31 00:58:37 bluelightning: yeah, I did something like https://github.com/kraj/openembedded-core/commit/beca51e7c4b595118f813df85470c4746e34a4e9 Mar 31 01:00:06 khem: right.. not sure if base or the inc file is the best place to put it, but either will work Mar 31 01:00:07 may be limiting it to tclibc-musl is better Mar 31 01:00:51 khem: you may need to limit it to class-native as well to avoid affecting native recipes Mar 31 01:00:56 er Mar 31 01:01:00 class-target, I meant Mar 31 01:01:35 for a workaround like this you probably should have a comment in the function explaining it as well Mar 31 01:03:28 libc-musl override is only effective for target Mar 31 01:03:30 https://github.com/kraj/openembedded-core/commit/5a776238a0c85ec13f603c16009b35dc5c80bb20 Mar 31 01:03:44 is probably more close to what u suggested Mar 31 01:30:00 overrides wont work there I believe Mar 31 02:09:30 bluelightning: this version seems to work https://github.com/kraj/openembedded-core/commit/325f1d574a0a3f7a2f5f8295bf4a81bc809472c4 Mar 31 02:12:48 khem: looks OK to me, but we ought to have a comment as previously noted (since it's a bit of an odd thing and it's not going to be clear why it's needed without going to the commit history) Mar 31 02:19:19 bluelightning:I updates the commit log Mar 31 02:19:25 hope thats better check again Mar 31 02:21:38 khem: sorry I meant in the function itself (but the commit message is a bit better thanks) Mar 31 02:22:54 o i c Mar 31 02:24:30 I know we don't have that many comments in our code, but when it's something slightly unusual I think it's good to have them Mar 31 02:25:08 added comment in the code too Mar 31 02:27:38 bluelightning: what are quick steps to do recipe version bumps using devtool Mar 31 02:27:45 I have figured few flows Mar 31 02:27:52 e.g. fixing a build breakages Mar 31 02:27:55 that works great Mar 31 02:33:25 khem: devtool upgrade recipename -V version [-S srcrev for git recipes] Mar 31 02:33:45 resolve any patch application conflicts Mar 31 02:34:28 then test/fix as appropriate Mar 31 02:34:29 once you're happy with it, devtool finish recipename layername Mar 31 02:39:43 I wonder how can I make do_install[postfuncs] specific to targrt Mar 31 02:41:18 absolute worst case you could use ${@bb.utils.contains()} Mar 31 02:41:36 or unconditionally add the postfunc, but make it empty for the non-target case Mar 31 02:41:45 either would do Mar 31 02:53:59 kergoth: the problem I think is this Mar 31 02:54:49 https://github.com/kraj/openembedded-core/commit/aee3b98d82132e9e6b4cd85654d28ffad0ad3434 Mar 31 02:55:09 now it will still add a postfunc even if empty Mar 31 02:55:17 and that will change signatures :( Mar 31 02:55:35 and we cant share the sstate between musl and glibc Mar 31 02:55:39 for native pkgs **** ENDING LOGGING AT Fri Mar 31 03:00:01 2017