**** BEGIN LOGGING AT Tue Aug 06 02:59:57 2019 Aug 06 06:28:48 Hello, http://git.yoctoproject.org/ seems to be down Aug 06 06:32:58 looks fine here (http, https, and git clone'd) Aug 06 06:34:10 tprrt: up and running at my place Aug 06 06:41:07 Chrusel: ok, so I will check the internal network. I have also the same issue on the https://wiki.yoctoproject.org/ but https://www.yoctoproject.org/ is accessible. Aug 06 06:45:27 tprrt: I have access to all sites/domains/subdomains ... good luck! Aug 06 06:50:51 Chrusel: Thanks you Aug 06 06:52:56 shauno: Thanks you Aug 06 06:53:15 New news from stackoverflow: How to add go-cross to Yocto SDK? Or otherwise compile Go for Yocto on other machines? Aug 06 08:53:33 New news from stackoverflow: dtc command not found in yocto Aug 06 10:21:31 I'm doing a fresh build (on a fresh machine) after returning from paternity-leave for a few months, and I'm having a problem that I believe I solved at some point in the past, but I can't for the life of me remember how.. I have a recipe that's calling a "home-grown" build-system (*sigh*), but one if the things that system does is resolve dependencies and then git-clones them from a local git server. I'm using an ssh keypair for authenti Aug 06 10:21:32 cation to those repos but under bitbake those keys can't be found (and the host is not a known_host).. Aug 06 10:21:42 Anyone who has had the same problem? Aug 06 11:20:58 For some reason it's trying to an ARM executable for git. Aug 06 11:46:49 Is it normal that a kernel binary isn't installed in an image (tar type)? Aug 06 11:52:54 alessioigor: it is. Aug 06 11:53:13 (unless requested otherwise, of course) Aug 06 11:53:36 LetoThe2nd: Could I request it? Aug 06 11:53:56 Using machine configuration file? Aug 06 11:53:57 New news from stackoverflow: use pip install command in recipes Yocto Project Aug 06 11:55:21 alessioigor: IMAGE_INSTALL += "kernel" :) Aug 06 11:57:50 * alessioigor runs to hide himself for the shame... Aug 06 11:58:41 alessioigor: happens, that. you're not the first one to expect things to be complicated. Aug 06 12:14:15 LetoThe2nd: Thanks! Aug 06 12:15:51 :-) Aug 06 12:29:49 Hello there! I have a question. Normally, at a computer that is running Ubuntu I compile my code for a project with these flags/arguments: -I/opt/etherlab/include, -L/opt/etherlab/lib, and -lethercat in order to include the proper headers and link the library. How can I do the same thing for my target computer? I have already created an image with the library I am using and created a recipe for that layer. I have also updated the SDK with respect to the new i Aug 06 12:32:48 fdav: 1) create a recipe and therefore package for this etherlab thing, and then depend on it. Aug 06 12:33:08 fdav: "and then" -> 2) Aug 06 12:34:16 I have already created a recipe for it Aug 06 12:34:43 fdav: so it gets built & installed properly, right? Aug 06 12:34:56 yes Aug 06 12:35:29 well then DEPEND with the other recipe on it Aug 06 12:36:25 it seems like I just can't link the library, it includes the proper headers accordingly. I get bunch of "undefined reference" errors Aug 06 12:36:56 I just compile my application using the SDK and scp it to the target computer, there is not a recipe for my application right now Aug 06 12:37:30 Do I make any sense, I am not sure I am kind of new to the Yocto world :D Aug 06 12:38:36 fdav: all i can say then is to find out why you can't link to that library. inspect the logs in tmp/work/yourpackage/somewhere, there should be extensive information Aug 06 12:39:02 thanks, I will do that! Aug 06 12:48:15 RP: Do you have any reference docs for setting up the Xvnc/QEMU configuration for the pokybuild user? Aug 06 13:07:35 Hi, I'm trying to build a poky distribution for a very specific card which use MFGTOOL. My question is how can I generate a MFGTOOL profile ? Aug 06 13:08:12 alexb3600: http://layers.openembedded.org/layerindex/branch/master/recipes/?q=mfgtool Aug 06 13:12:13 LetoThe2nd: So do I just need to add the meta-freescale layer and build the fsl-image-mfgtool-initramfs image ? Aug 06 13:13:08 alexb3600: it depends. generally you should expect the provider of your board to also provide the documentation needed to use it. Aug 06 13:14:45 LetoThe2nd: the BSP is already given by the provider but it use a non-yocto custom image builder Aug 06 13:15:28 LetoThe2nd: and I want to see if yocto is better Aug 06 13:16:08 LetoThe2nd: in other words I want to create the yocto BSP of this board Aug 06 13:16:11 tgoodwin: its really simple, just starting Xvnc and running the runqemu tap generation script Aug 06 13:16:16 alexb3600: as you can see from the list of hits in the search i linked you to, there's obviously a variety of ways to use mfgtool. i am no nxp user so i cannot properly comment, but in general you need to pick/create a setup that replicates the setting that you have been provided with. Aug 06 13:16:47 RP: I think one of the issues I'm running into is that "checkvnc" assumes it owns display 1, which it does not. Aug 06 13:17:09 The other is that I did not run vncserver once to set the password as the build user. Aug 06 13:17:12 tgoodwin: right, the display number is hardcoded :/ Aug 06 13:17:28 tgoodwin: ah, yes, that would also be needed... Aug 06 13:17:35 tgoodwin: we should document this :/ Aug 06 13:18:01 RP: I'm currently looking for a way to parse vncserver -list to pull the active one. Probably replace that tcp response with looking for pid files and checking their status. Aug 06 13:18:17 LetoThe2nd: Hum ok, i have the device table, the kernel config, the device tree source of my board. So do you think it's possible to create a custom yocto layer for my board thanks to these files ? Aug 06 13:19:04 RP: I noticed in one of the build logs that it looks like the builder attempts to use sudo too. Aug 06 13:19:29 tgoodwin: only if the tap devices don't exist Aug 06 13:19:34 alexb3600: anything is possible, just the needed effort varies Aug 06 13:19:46 tgoodwin: oh, the performance workers do have sudo to drop the caches Aug 06 13:21:04 LetoThe2nd: ok thanks a lot Aug 06 13:23:15 alexb3600: have fun Aug 06 13:24:47 LetoThe2nd: Ahah :) Aug 06 14:19:25 hi I've run into dhcp issues when using ini-ifupdown package in readonly-rootfs -> is there some how to described what must be symlinked to writable partition somewhere? thx Aug 06 14:44:18 opennandra> maybe one way could be to create several overlayfs over /var & /etc and see which file is created inside? Aug 06 14:44:58 nayfeL I have /data partition where I can write, but I plan to use overlayfs (is there some meta-layer for using that easily)? Aug 06 14:45:47 easiest way is to use volatile-binds with systemd, but it seems you use sysvinit? Aug 06 14:46:50 nayfe: nope I'm using systemd Aug 06 14:49:49 opennandra> so in a volatile-binds.bbappend file, if you add VOLATILE_BINDS_append = " /data/var-lib /var/lib \n " it will use overlayfs in latest yocto releases Aug 06 14:50:28 nayfe: I'm using thud Aug 06 14:53:37 zeddii, ping Aug 06 14:53:50 opennandra> more precisely, it's a systemd feature from 242 if I remember right, it's ok on thud Aug 06 14:54:13 nayfe: ok cool so I'll try to create it an lets see how it behave then Aug 06 14:54:16 thanks Aug 06 14:56:59 hum, I'm wrong, it's from https://git.yoctoproject.org/cgit.cgi/poky/commit/meta/recipes-core/volatile-binds/files/mount-copybind?id=c4acf1b531843497297c492456daea1744400faa but thud is ok, anyway Aug 06 14:58:12 opennandra> one prerequisite is to enable overlayfs kernel option of course Aug 06 14:58:26 YPTM - armin is on Aug 06 14:58:45 YPTM - Randy has joined. Aug 06 14:58:58 nayfe: yes sure, so it means data will be written to overlay and when deleted some default will remain in rootfs and I don't have issue with writing to rootfs Aug 06 15:00:38 YPTM - Jan-Simon has joined Aug 06 15:01:13 YPTM - Trevor has joined Aug 06 15:01:40 YPTM - Vineela ha joined Aug 06 15:01:55 monthly YP call: https://www.yoctoproject.org/public-virtual-meetings/ -- join if you are interested. Aug 06 15:02:00 YPTM - the "other" trevor has joined ;-) Aug 06 15:02:09 lol Aug 06 15:02:11 YPTM: Joshua Watt here Aug 06 15:04:04 did we ever change the wemu machine to an a9 with neon? Aug 06 15:04:47 YPTM - Tim has joined Aug 06 15:07:35 armpit: lol Aug 06 15:08:04 RP: are there any specifics you can give me for the tap needs? Also is a "dummy" interface alright (the server is running netplan) Aug 06 15:08:11 RP: I can probably take a look at the locale-problem, but I'd need full access to a build carcass. Aug 06 15:08:13 tgamblin: out of curiosity, do you get called "kevin" a lot? Aug 06 15:08:40 RP: I didn't have a mic on the computer I am using, or I would have spoken up. Aug 06 15:08:59 tlwoerner: you know it. Also "Travis" Aug 06 15:09:28 I can obtain a build carcass from Randy or fray, locally after we get the change pushed in. Aug 06 15:09:35 tgamblin: LOL! "... things only a 'trevor' would understand... " Aug 06 15:19:33 vmeson: https://realpython.com/jupyter-notebook-introduction/ Aug 06 15:21:17 thx, FYI: The layer is here: https://github.com/sybotech/meta-jupyter Aug 06 15:47:41 fyi: version numbers: https://wiki.yoctoproject.org/wiki/Releases Aug 06 15:52:12 YPTM is over Aug 06 16:02:17 tgoodwin: I'd run something like sudo `which runqemu-gen-tapdevs ` 1000 sharedsource 8 tmp/sysroots-components/x86_64/qemu-helper-native/usr/bin Aug 06 16:03:11 jwessel: I'll try and get you the rpm package list but we can also get you ssh access to an ab worker, we need halstead for that Aug 06 16:28:59 RP, that would probably be even faster to solving the problem. Aug 06 16:30:22 Always some kind of a new problem. It would be interesting to know where the sqlite config binary it is that is calling, .e.g. how did your system get to that state. Aug 06 16:30:53 I haven't seen any kind of build failure like the one you had referenced. Aug 06 16:32:05 jwessel: its several of our autobuilder workers so not isolated Aug 06 16:32:34 Do you have other problems with it building pseudo, or just as a result of this upgrade? Aug 06 16:32:53 jwessel: everything fine in previous builds Aug 06 16:33:04 Does it use sstate? Aug 06 16:33:17 jwessel: yes, heavily Aug 06 16:34:09 I have a guess as to how you get to this problem then. A version of the sqlite3 will build against libz, if it is there. Aug 06 16:34:37 jwessel: that is entirely possible Aug 06 16:34:52 But on the pkg uprev, if it called the sqlite3 lib print util, but libz-native isn't there yet, you would hit it. Aug 06 16:35:17 It has to be a parallel build + some heavy state to get to the exact situation. Aug 06 16:35:26 jwessel: are you saying sqlite-native doesn't build deterministically ? Aug 06 16:35:28 Which is solveable with some dependencies. Aug 06 16:35:33 correct. Aug 06 16:35:43 I am 100% certain based on a scan of the lib printer. Aug 06 16:35:45 jwessel: we should fix that then, its probably the cause Aug 06 16:35:58 Let me setup a test. Aug 06 16:36:02 I can probably prove it. Aug 06 16:36:41 Everything I have done with the OS Tree work has been stepping on various hidden bear traps... Aug 06 16:36:49 Eventually I will have tripped them all. Aug 06 16:37:27 jwessel: I think that is my life these days... Aug 06 16:37:43 jwessel: the new runqueue hashequiv stuff as primed up a load more... Aug 06 16:37:44 heh... I know your pain. Aug 06 16:37:53 s/as/has/ Aug 06 16:38:18 I didn't expect to be fixing wic, pseudo and a myriad of other problems. Aug 06 16:38:50 I have a fix in my branch for dpkg too where it has a conflict with busybox we never have seen before due to usrmerge. Aug 06 16:39:23 I am guessing we can probably track down the glibc-locale thing too. Aug 06 16:45:40 jwessel: tracking that one down would be awesome, its been bugging us for ages Aug 06 16:46:17 I make no immediate promise that it will be easy to solve of course. But I will take a look at it. Aug 06 16:46:46 The code is fairly racy with the glibc-locale "stashlocale" + the glibc build. Aug 06 16:47:14 Like we uncovered with the openat() pseudo, we know there are some cases where DB entry can be dropped. Aug 06 16:47:49 A really careful analysis of the build along with the event sequencing will be needed. :-) Aug 06 17:06:58 RP: I haven't been able to duplicate exactly, but I can tell you for sure that sqlite3 builds with and without zlib differently. I have the configure logs. Aug 06 17:07:19 Simply adding a DEPENDS = "zlib-native" to sqlite3 fixes it. Aug 06 17:17:27 Hello, I'm getting an issue on the thud branch with files not showing up in the SDK...we've found an issue in the "debian.bbclass" where it is trying to call objdump on the library files to get their soname, but it's failing because the symlinks are broken because it's running objdump in the packages-split directory Aug 06 17:18:00 this is causing some packages to get renamed such as the "main" package but not the "-dev" package Aug 06 17:19:26 marler8997: is it running cross-objdump or native ? Aug 06 17:19:32 cross Aug 06 17:19:37 we're getting "No such file or directory" Aug 06 17:19:53 and it's because the symlink points to a file in another package, so if all the dependencies were installed the symlink works Aug 06 17:19:59 yeah I was hoping the link sections might help but seems not Aug 06 17:20:39 hmm cross-package symlinks Aug 06 17:21:01 I think it should just ignore then Aug 06 17:21:32 the debian class detects if the package has a library in it Aug 06 17:21:53 it does this by looking for all the "*.so*" files in the package, then running objdump on it to get the SONAME from the elf binary Aug 06 17:22:06 so it's failing the objdump because it's looking in the "package-split" directory Aug 06 17:22:14 it shouldn't be failing the objdump Aug 06 17:22:32 if it ran objdump on the file in the "package" directory, then it would work Aug 06 17:22:44 so then the debian class doesn't detect that it's a library and doesn't rename the package accordingly Aug 06 17:26:01 is anyone else seeing this issue? Is this a known issue? Aug 06 17:32:31 marler8997: Does it fail for only your specific package or all of them? Aug 06 17:32:46 it's failing for a bunch of packages in oe-poky Aug 06 17:32:48 I was just wondering if it is a symptom of the build output. Aug 06 17:32:59 It is probably a generic problem then. :-/ Aug 06 17:33:46 I only use the opkg or rpm format. I am wondering why the rule is significantly different for the dpkg? Aug 06 17:34:04 Because we don't see a similar problem there, that I am aware of. Aug 06 17:34:26 I'm also using opkg Aug 06 17:34:27 And the "build binaries" are the same in all 3 cases. opkg, dpkg, rpm. Aug 06 17:34:41 The debian renames packages for all formats Aug 06 17:34:47 The *debian class Aug 06 17:36:30 RP: I sent you the patch to fix the pseudo-native link issue. The pseudo-native build was just the victim. Aug 06 17:44:01 running before split wont help since its trying to calculate rdeps which has to be per package Aug 06 17:44:24 or also for calculating debian renaming rules Aug 06 17:44:41 I'm looking into it more and it looks like it actually may be purposely failing Aug 06 17:44:55 It looks like when it detects one package with the actually library file, it renames all the packages Aug 06 17:45:08 so it looks like the problem of missing " Aug 06 17:45:17 so it looks like the problem of missing "-dev" packages is somewhere else Aug 06 20:41:55 jwessel: thanks, I just had a look at this Aug 06 20:42:14 RP I will make a v2 that uses the pkgconfig. Aug 06 20:42:31 jwessel: It looks very much like sqlite3-native was built on a system with zlib-devel and then used from sstate on a system that doesn't have it Aug 06 20:42:42 Ah... That would cause it too. Aug 06 20:43:06 So you do fall into that camp like I explained the mail where you can build with it and later not have it. Aug 06 20:43:19 But for a different reason, multi-host sstate. Aug 06 20:43:23 jwessel: think of the scenario where two workers start up slightly after each other where one has built sqlite3-native but the other is just starting Aug 06 20:43:46 Sure. I totally understand how it can happen. Aug 06 20:43:51 jwessel: right, our sstate is shared Aug 06 20:44:05 jwessel: your patch probably could just do DEPENDS = "zlib" Aug 06 20:44:17 But how does that work for -native ? Aug 06 20:44:29 then its correct for the target and native cases (as the native code will remap it as usual) Aug 06 20:44:30 Does it append in a zlib-native automatically? Aug 06 20:44:47 jwessel: it will turn "zlib" into "zlib-native" in the native case Aug 06 20:45:00 Fair enough. I was looking at what zlib is used for in sqlite3 and we never use it. Aug 06 20:45:10 Why make the thing bigger than it already is. Aug 06 20:45:12 jwessel: that was my other question. Can we just disable it Aug 06 20:45:25 The target version was not building with it. Aug 06 20:45:35 Like I said, for the the v2, I'll make it a pkgconfig option. Aug 06 20:45:39 So someone can turn it on. Aug 06 20:45:54 jwessel: I think we can change it if we set ac_cv_search_deflate Aug 06 20:45:56 It has a use, for doing "compressed page support". Aug 06 20:46:05 That is another option. Aug 06 20:46:18 I wasn't sure how pkgconfig could be used to change those options. Aug 06 20:46:32 jwessel: if we patch it I'd just ask we let upstream see the patch Aug 06 20:46:43 we have far too many patches :/ Aug 06 20:46:43 I always do. Aug 06 20:46:53 Doesn't mean they accept the patches. Aug 06 20:46:57 But I always try. Aug 06 20:47:14 jwessel: right, I was specific about them seeing it, not accepting :) Aug 06 20:48:18 It is probably reasonable try the config check and fake out the pkgconfig so you can build it both ways. Aug 06 20:48:29 Then you don't have to change any code at all. Aug 06 20:48:41 Just the recipe that is. Aug 06 20:49:04 CACHED_CONFIGUREVARS += "ac_cv_search_deflate=no" Aug 06 20:49:07 that might do it Aug 06 20:49:19 It will because we autoconf it, from the ac file. Aug 06 20:49:28 I'll try it out and send another patch in a bit. Aug 06 20:50:40 jwessel: FWIW that seems to work here Aug 06 20:51:01 hey, if you want to patch it and roll with it, be my guest :-) Aug 06 20:51:22 jwessel: I'm tempted to try and unbreak master-next :) Aug 06 20:51:43 Exactly. I'd like to get the pseudo patch more soak time. Aug 06 20:53:22 I'll quickly try out the pkgconfig one I was thinking about. Aug 06 20:53:36 Where we just use the variable, so you can compile it both ways. Aug 06 20:57:29 jwessel: patch sent to the list FWIW :) Aug 06 20:58:14 +PACKAGECONFIG[zlib] = ",,zlib" Aug 06 20:58:14 + Aug 06 20:58:14 +CACHED_CONFIGUREVARS += "${@bb.utils.contains('PACKAGECONFIG', 'zlib', '', 'ac_cv_search_deflate=no',d)}" Aug 06 20:58:29 I believe that does what we want, but I wanted to build test it. Aug 06 20:59:19 jwessel: I guess. The thought of what the parser has to do with this always puts me off writing it! Aug 06 20:59:55 True, but that is the only mechanism we have since there is no other syntax way of dealing with it. Aug 06 21:00:03 I find those expressions to be ugly myself. Aug 06 21:00:20 jwessel: I know. I wish we had a better way but I've never come up with one Aug 06 21:00:58 Considering we only have 6 types of common expressions, we could use something other than a python eval. Aug 06 21:01:07 If you were looking to optimize it anyway... Aug 06 21:01:11 :-) Aug 06 21:01:40 another time another day on that one... Aug 06 21:02:14 jwessel: Yes, FWW I would welcome any thoughts on what this other syntax could look like Aug 06 21:04:24 BTW, with the PACKAGE_CONFIG option, it fixes khem's concern because you can safely use the PACKAGECONFIG or PACKAGECONFIG_class-native Aug 06 21:05:20 jwessel: yes, I just replied as such Aug 06 21:05:29 jwessel: I did say your patch was uglier :) Aug 06 21:05:58 Without a doubt. I don't use bitbake stuff like that because it is pretty... Only because that it is the only way. Aug 06 21:06:25 jwessel: I'm aiming to fix another issue before I run the next build so if you have the other patch sent out I'll use it Aug 06 21:06:34 I'll have it in a few min. Aug 06 21:06:59 "checking for zlib.h... yes" was just checking to see it actually does what it is supposed to. Aug 06 21:07:38 jwessel: it will say yes or no, its the next line about deflate which is key Aug 06 21:08:11 log.do_configure.2521:checking for library containing deflate... -lz Aug 06 21:08:11 log.do_configure.5482:checking for library containing deflate... (cached) no Aug 06 21:08:31 Correct. I was just making sure it worked in the yes case adding the dependency etc... and the no case. Aug 06 21:08:35 It is still building all the configs. Aug 06 21:09:17 Forgot to re-install zlib. Aug 06 21:09:27 So the test didn't work. almost there... Aug 06 21:14:22 RP: I followed up with the v2. Aug 06 21:15:00 oops looks like it mis sent... Damn... Aug 06 21:16:30 I sent it from the wrong repo the first time. Sorry about that. Aug 06 21:16:34 It is fixed now. Aug 06 21:17:38 It passed all the test configurations. Aug 06 21:31:36 RP: Does the QA glibc-locale look something like this? Aug 06 21:31:39 WARNING: glibc-locale-2.29-r0 do_package_qa: QA Issue: glibc-locale: /usr/lib64/locale/ar_YE/LC_COLLATE is owned by uid 5002, which is the same as the user running bitbake. This may be due to host contamination [host-user-contaminated] Aug 06 21:32:25 I hit it on one of my builders while using the new pseudo patch and some others. Aug 06 21:42:14 jwessel: yes, that looks like it Aug 06 21:42:41 I am poking around with it now, out of curiosity... Aug 06 21:42:50 jwessel: I actually have my own theory about it. I suspect its unfortunate inode number reuse :/ Aug 06 21:43:12 jwessel: a file is deleted which is in psuedo's db and then a new file is created with the old inode number Aug 06 21:43:15 That is possible I suppose. Aug 06 21:43:35 jwessel: never found a good way to test that theory :/ Aug 06 21:43:50 jwessel: autobuilder build running fwiw Aug 06 21:44:24 I could make a kernel that gave you the same inode or edit the fs or something really sly... There are other ways to cheat it, if you really think that is the issue. Aug 06 21:44:47 jwessel: I'm just mentioning it as one possibility Aug 06 22:05:05 What is really odd, is that it is a single file out of 7000+ Aug 06 22:07:25 jwessel: and it varies randomly within the 7000 files Aug 06 22:07:58 jwessel: it also is usually in rebuilds when there is more inode churn in my experience... Aug 06 22:11:58 I'll have to think about this one a bit more... In theory you could have it happen running: "bitbake glibc-locale -C compile" in a loop Aug 06 22:12:14 Based on what you have observed. Aug 06 22:14:02 But there would need to be some more activity probably in the same pseudo context. Aug 06 23:02:52 jwessel: Juro posted a reproducer script which was pretty much that on the list **** ENDING LOGGING AT Wed Aug 07 03:02:32 2019