**** BEGIN LOGGING AT Tue Aug 23 02:59:58 2016 Aug 23 06:07:28 RP: https://github.com/01org/meta-security-isafw/pull/70 fixes https://autobuilder.yoctoproject.org/main/builders/nightly-musl/builds/227/steps/BuildImages/logs/stdio Aug 23 06:07:42 Hi @all ... is there somewere a description on how to use the bitbake "domains" feature ... from bitbake help --log-domains=DEBUG_DOMAINS? Aug 23 06:08:47 I am facing problem while building kernel version 3.14.4 for beagle bone black Aug 23 06:13:01 MDNNeo: look for BBDEBUG, but I think it is not well documented, you can file a bug for it Aug 23 06:21:31 khem: so basically it seems to be kind of straight forward ... I can do following in a recipe python func ... logger = logging.getLogger("BitBake.Test") logger.debug(1, "test dbg1") ... like here https://github.com/openembedded/bitbake/blob/master/lib/bb/persist_data.py Aug 23 06:22:16 but if I call bitbake -v -D -l Test myimage ... I still see all the debug and notes not only the test domain Aug 23 06:24:36 yep thats because it will print all messages at that level Aug 23 06:26:13 hmm ... basically what I would want is to just get the dbg msg from "my domain" ... Aug 23 06:26:51 not sure if the domain thing is what I need :) Aug 23 06:40:23 you could emit a unique identifier for your messages and filter them Aug 23 06:46:50 khme: hmm ... but then I also could just have a look in the log file instead Aug 23 07:13:06 hi, is it possible for the fetcher or unpacker to place a git pull into a existing subdir without overwriting it? Aug 23 07:14:06 i already tried subdir and destsuffix, but it does not seem to work. Using the svn fetcher subdir worked. Aug 23 07:40:43 nobody knows? Aug 23 08:18:53 heey guys, I got a question w.r.t. using Yocto in an air-gapped enviroment Aug 23 08:19:39 I'm introducing Yocto to a company and they are air gapped. How could Yocto cope with this? Aug 23 08:20:39 flitjes: you mean "the machine where yocto builds is not connected to the internet" right ? Aug 23 08:21:07 @boucman_work yes exactly Aug 23 08:21:27 ok. yocto has various facilities for that Aug 23 08:21:29 @boucman_work I got everything to work on a public wifi, which they kinda not like ;-) Aug 23 08:21:51 hehe Aug 23 08:21:53 they like yocto, not the work around w.r.t. the air gap Aug 23 08:22:10 Ok could you point me to some documentation which explains they facilities? Aug 23 08:22:26 http://www.yoctoproject.org/docs/latest/mega-manual/mega-manual.html#setting-up-effective-mirrors Aug 23 08:22:41 basically, you have a machine with your yocto setup, you put it on the net Aug 23 08:22:53 you run "bitbake -c fetchall " Aug 23 08:23:11 then you use its download directory to build a mirror on your intranet Aug 23 08:23:43 but whenever someone wants to add a new software on the target, you will need to redo that round to get it on the mirror Aug 23 08:24:04 Ok, that will work Aug 23 08:24:07 (so maybe do a fetchall world for a start. It will do a huge mirror, but you will have everything) Aug 23 08:24:13 that's what we do now as well for the other bsp Aug 23 08:24:23 that's nice! Aug 23 08:24:51 fetchall world, it will be big, but then we could work freely within the air gapped build system Aug 23 08:25:00 Space will not be the issue :) Aug 23 08:26:01 :) Aug 23 08:28:37 Hii, Aug 23 08:29:45 I want to change the downloads directory path in yocto Aug 23 08:31:06 Dhaval: http://www.yoctoproject.org/docs/latest/mega-manual/mega-manual.html#var-DL_DIR Aug 23 08:33:29 anybody pls, is it possible for the fetcher or unpacker to place a git download into a existing subdir without overwriting it? Aug 23 08:39:10 qwert: I think you can specify the name of the directory to unpack to somehow, but I don't know how it's done... Aug 23 08:39:41 qwert: https://www.yoctoproject.org/docs/latest/bitbake-user-manual/bitbake-user-manual.html#bb-the-unpack Aug 23 08:40:02 apperenty adding a subdir key to your SRC_URI should do the trick Aug 23 08:41:07 boucman_work: thank you for your response. I am aware of the subdir parameter and i also tried the destsuffix parameter to the git fetcher. Both attempts failed Aug 23 08:42:02 I know this questions was here for a few days but How can I remove Recpies from IMAGE_INSTALL, which are included in deeper layers. Aug 23 08:42:16 I want to remove connmand and insert network manager Aug 23 08:44:22 IMAGE_INSTALL_remove I guess Aug 23 08:52:03 since dora thats basically correct Aug 23 08:52:24 boucman_work: I am not an python expert, but it seems like within bb/git.phy :258 (krogoth) the destdir will be deleted if it exists using bb.utils.prunedir Aug 23 08:52:37 but i'd rather find out *why* they are included deeper inside. usually theres a reason. and if not, change it there. Aug 23 08:53:29 plus, i'm pretty sure that connman vs. nm is more than just a package selection. Aug 23 08:55:47 hello Aug 23 08:56:02 how is the checksum of the license computed? Aug 23 08:57:45 like in LIC_FILES_CHKSUM Aug 23 08:57:53 c0rnel: md5 of the file, usually Aug 23 08:58:41 c0rnel: take a look at bitbake/lib/bb/checksum.py Aug 23 08:58:52 LetoThe2nd, it seems that in certain cases, when the license text is surrounded by some other text, only the checksum of the license is computd but i don't know how Aug 23 08:59:02 qwert, thank you! Aug 23 09:03:50 I wish the repo management built on top of Yocto were standardized! We are struggling with 3 different HW vendors, all Yocto based, but with three completely different schemes for doing configuration/repo management Aug 23 09:05:20 people love overcomplicating things sometimes :P Aug 23 09:07:09 (and we're probably going to make our own, as there is no governing template for it) Aug 23 09:09:28 sveinse: I know I said that here a dozen time, but there is no consensus. :( Aug 23 09:09:45 some people agree, other consider this a way for people to differentiate. Aug 23 09:10:11 sveinse: the best one I have found so far is repo, but it's not yocto specific in any way Aug 23 09:10:32 (and my company already did it's own tool too, and it's no better than the others) Aug 23 09:14:31 boucman_work: this IMAGE_INSTALL_remove seems not to work: "The following packages conflict with networkmanager check_conflicts_for: connman *" Aug 23 09:19:49 HyP3r: thats what i mean. connman vs. nm is more than just adding removing packages. check what gets pulled in where. maybe DISTRO, or such. Aug 23 09:20:13 hi all, what is the best way to make a recipe for an external kernel module which depends on modules from the kernel itself? Aug 23 09:20:35 in my case, I want to build the qcacld-2.0 module which depends on cfg80211 from the kernel Aug 23 09:21:05 hi guys Aug 23 09:21:09 but by default the KERNEL_SRC seems to be set to STAGING_KERNEL_DIR which doesn't include any object Aug 23 09:21:28 changing it to STAGING_KERNEL_BUILDDIR doesn't help since it doesn't include module objects Aug 23 09:45:36 pidge: ping ? Aug 23 09:45:53 boucman_work: pong Aug 23 09:46:01 hey Aug 23 09:46:20 I just sent v1 of my patch to remove hardcoded references to gcc Aug 23 09:46:36 ah, grand. I'll take a look at them. Aug 23 09:46:37 it's... bigger than I expected. quite a few core stuff has references to gcc Aug 23 09:47:05 heh. it always is. Aug 23 09:47:31 I was coming back to you to see if there is a way to somehow test that automatically in the builders. I tested with core-image-minimal, but I don't have the time/CPU to test with world... and that's what's really needed :( Aug 23 09:48:03 I explain in my cover letter how I tested, something similar could be done to the autobuilder, but it will probably trigger many build failures Aug 23 09:49:46 boucman_work: i cannot offer you a autobuild test, but if you need a specific one-shot test at world i can do that. Aug 23 09:51:43 LetoThe2nd: hmm, interesting. I'll come back to you about that... but from a QA point of view it would be better to do it at the yocto autobuilders level... Aug 23 09:52:47 boucman_work: yes, certainly. no doubt. Aug 23 10:06:25 gbisson: you just have to inherit modules. Aug 23 10:07:13 gbisson: take a look at: http://www.yoctoproject.org/docs/2.2/mega-manual/mega-manual.html#var-KERNEL_MODULE_AUTOLOAD http://www.yoctoproject.org/docs/2.2/mega-manual/mega-manual.html#working-with-out-of-tree-modules Aug 23 10:07:47 HyP3r: yes I inherit 'modules' but this seems to set KERNEL_SRC to the source without objects, since my modules depends on cfg80211 it doesn't find the symbols... Aug 23 10:08:34 HyP3r: I will paste my recipe and the log here after lunch so you can see the problem Aug 23 10:08:55 gbisson: I know what you mean. I was developing the a kernel module last week Aug 23 10:09:09 boucman_work: And I endorse freedom to chose. But when you're a end-user product builder, working on systems delivered from different vendors, extra labor is added to integrate these systems together. I've already been met with the argument "but I thought the reason for having Yocto was that it is standard" from my manager :P Aug 23 10:09:18 HyP3r: but it seems to be a known issue: https://www.mail-archive.com/yocto@yoctoproject.org/msg29197.html Aug 23 10:09:39 HyP3r: was your kernel depending on other modules? Aug 23 10:09:50 s/kernel/kernel module/ Aug 23 10:10:59 sveinse: i don't need convincing :P i'm a heavy proponent of having some sort of "bitbake-deploy" tool that would do all steps needed to have a proper yocto setup out of nothing Aug 23 10:12:47 boucman_work: Are there any efforts active for such an initiative? Or is this just wishful thinking? Aug 23 10:17:08 gbisson: gbisson the KERNEL_SRC Variable is pointing to a directory which is pretty 'clear' configurations are in the directory 'kernel-build-artifacts' Aug 23 10:17:21 gbisson: but no, my kernel module was out-of-tree but without any dependencys Aug 23 10:18:30 gbisson: but you could write a recpie whcih compiles both kernel modules which a Makefile which tries to compile both at one runtime Aug 23 10:23:55 HyP3r: thanks, but that looks hacky, I mean building this external module manually is very straight forward, don't understand why it gets so complicated just because it needs another module. I wish there was a variable that would point to the kernel build dir with both kernel and module objects... Aug 23 10:25:23 gbisson: that was my problem too. My kernel module (wf111) was a bit shitty, and the makefile needs to see the .config file and some header files in the same directory Aug 23 10:25:39 gbisson: I was only able to solve that by patching the Makefile Aug 23 10:25:55 gbisson: the yocto project should be renamed to 'pachting makefile til it works' Aug 23 10:27:46 hello Aug 23 10:27:54 a question regarding toaster Aug 23 10:28:10 is it possible to build a sdk with toaster ? Aug 23 10:28:17 i'm using krogoth Aug 23 10:30:33 iskander: it is. You need to add :populate_sdk after the target. So if you are building an image called core-image-myimage, you type in the build form: core-image-myimage:populate_sdk (or populate_sdk_ext) for the extensible sdk Aug 23 10:31:09 cool, thanks, that's what i need Aug 23 10:31:41 i found no documentation regarding toaster+sdk Aug 23 10:32:10 iskander_work: yes, sorry about that. We should add that to the manual for the upcoming release Aug 23 10:32:32 i understand Aug 23 10:33:14 cool, it's building :) Aug 23 10:33:16 hmm, just had a gtk+ recipe fail, tried to clean/cleansstate it 'gtk+ was skipped: one of '['directfb', 'x11']' needs to be in DISTRO_FEATURES' so?....... Aug 23 10:35:24 btw, PREMIRRORS variable works great now after patching views.py :) Aug 23 10:36:27 iskander_work: excellent. Thanks for the patch. We'll try to upstream it today :) Aug 23 10:36:38 np, thanks for support Aug 23 10:37:29 odd, i'm subscribed to toaster mailing list but i got a message that my mail is awaiting approval Aug 23 10:37:51 sveinse: wishfull thinking... I could probably get a 6month trainee to work on that, but the subject is pretty complicated... on the design part. Not something you want to give to a trainee until you have a very complete idea of what you want Aug 23 10:38:34 iskander_work: that's weird. You normally need approval if you are not subscribed. I've released the email, and it's already being delivered. No need to resend Aug 23 10:38:57 ok Aug 23 10:41:47 boucman_work: I hear you brother. I'm thinking the same thing. We can wish as much as we want from a yocto consumer pov, but unless this is being deployed as a general consensus in the community, this will never be implemented Aug 23 10:43:23 belen, i subscribed again, odd, i thought i was already, hmm Aug 23 10:43:24 ..or alternatively do the same approach as the systemd changeover in the distros... Aug 23 10:45:30 iskander_work: mailman hiccup maybe ;) Aug 23 10:50:03 Hi, I have problem I have to install some directories in qemu-native, do_install_append-native() doesnt works Aug 23 10:51:43 kukos: do_install_append_class-native() (note the missing "_class" Aug 23 10:51:54 ) Aug 23 10:52:52 thanks, I try this now Aug 23 10:56:26 It works, thanks a lot, in nativesdk I need class too ? Aug 23 11:01:26 kukos: only if you need to do something special for nativesdk Aug 23 11:01:33 in that case, _class-nativesdk will work Aug 23 11:01:50 try to use general code if you can though, that will work regardless of the type of recipe Aug 23 11:02:54 see the note in http://www.yoctoproject.org/docs/2.2/ref-manual/ref-manual.html#var-FILES for one way to make code general for example Aug 23 11:03:56 look up BBCLASSEXTEND as well if you haven't seen it Aug 23 11:06:39 Has anyone any experience with web servers? I'm trying to figure out which one would be the fit. A comparision between nginx and lighttpd concerning embedded devices would be nice. Aug 23 11:07:33 anticom: do not judge upon names, judge upon the feature set you actually need. Aug 23 11:08:02 LetoThe2nd: Basically i just need to run a c/c++ fcgi app and serve some static content Aug 23 11:08:27 But i've read, that lighttpd has some year-old bugs that haven't been fixed yet and hence some people are quite upseet Aug 23 11:08:35 anticom: we've successfully used lighttpd for that on an arm9 Aug 23 11:08:45 on the other hand i read that nginx might be a little more greedy when it comes to resource consumption Aug 23 11:09:45 anticom: greedyness concerning resources alone does not really mean anything. the point is, are you short on cpu/ram, or can you happily use them up to gain faster development or better results? Aug 23 11:10:42 LetoThe2nd: There is some ram+cpu available but it should be used as minimal as possible since there are some other daemons running that really need especially the computing power Aug 23 11:11:16 LetoThe2nd: But also the fcgi app should be able to serve a couple of requests per second Aug 23 11:11:42 anticom: ok. the trick in embedded web always is to whift workload towards the client - which is opposed to what a lot of web devs to. hence the hint. Aug 23 11:13:16 anticom: but in the end, it really depends on what you need and can spare. for example, we've lately setup a project which uses nodejs+socket.io as its single webserver. eats up a whopping 100+X megabytes of ram at startup. but after that, stays low on cpu and gives good response times. Aug 23 11:13:46 LetoThe2nd: the fcgi app is just serving some json that will be processed client-side Aug 23 11:14:07 Also it shouldn't be big on computation Aug 23 11:14:20 anticom: fait enough Aug 23 11:14:23 *fair Aug 23 11:14:44 LetoThe2nd: And have you used nginx already aswell? Aug 23 11:14:50 Or no experiences there Aug 23 11:14:59 anticom: sounds very much like what we already did with lighttpd, so i could not advocate against it. we didn't run into major problems. and no, never used nginx. Aug 23 11:15:55 LetoThe2nd: Okay thanks Aug 23 11:16:26 One more thing although it's offtopic here: Since you did some fcgi stuff and i'm rather new to this. Can you recommend a fcgi library? Preferable C++ Aug 23 11:16:30 Preferably* Aug 23 11:33:43 anticom: nope. we were stuck with simple cgis, and knda abandoned them for nodejs Aug 23 11:34:27 LetoThe2nd: Like hand-crafted everything? Aug 23 11:34:48 anticom: libcgi/libcgi++ + handcrafting, yeah Aug 23 11:34:55 Well that's kind of the issue i'm having right now. All the (few) available libraries out there are from the stone age Aug 23 11:35:01 libcgi included Aug 23 11:37:20 belen, i got another problem with toaster :( Aug 23 11:37:49 sdk was built successfully but i'm not able to download it Aug 23 11:38:10 [Errno 2] No such file or directory: u'/var/www/toaster/poky/build-toaster-2/tmp/deploy/sdk/poky-glibc-x86_64-core-image-sato-cortexa8hf-ne' Aug 23 11:38:20 something is wrong with this link Aug 23 11:38:44 it's incomplete Aug 23 11:40:32 anticom: thats exactly why we switched to node+express Aug 23 11:40:49 the filename 'poky-glibc-x86_64-core-image-sato-cortexa8hf-neon-toolchain-2.1.1.sh' got truncated Aug 23 11:40:54 anticom: super easy creation of api endpoints, and json is built-in. Aug 23 11:41:48 anticom: you could look at the poco libraries though, and use those to create a simple server process on a second port, or behind internal forwarding. Aug 23 11:46:04 LetoThe2nd: Thing is need to be able to communicate with a daemon to exchange data which should be exposed via the web-server on an endpoint. So would you do that via fcgi in node (if that's even a thing) or is it in the js bindings already? It's been month i've last looked into node Aug 23 11:46:33 iskander_work: Have you checked the path on you fs? Aug 23 11:46:54 anticom: node needs no (f)cgi for me, it is the server process itself Aug 23 11:46:55 oh you have, didn't see your 2nd message Aug 23 11:47:18 LetoThe2nd: Well as i said, the reason we're doing all this is to expose information another daemon receives Aug 23 11:47:20 belen, yes, it's correct but truncated Aug 23 11:47:25 anticom: and it can communicate with whatever daemon you want through sockets or such Aug 23 11:47:42 So there is socket support in js inside node? Aug 23 11:47:50 that's basically what i wanted to know :) Aug 23 11:47:58 anticom: there is Aug 23 11:48:07 great! Aug 23 11:48:22 anticom: mind again, this is something that works well for us. it is not a general solution for all problems. Aug 23 11:48:44 anticom: node has a rather long startup time and needs way more ram than lighttpd or such. Aug 23 11:49:56 LetoThe2nd: Well I'd have to just try out lighttpd, node and maybe nginx and see what works best Aug 23 11:50:04 belen: the same problem with core-image-sato, it's truncated too Aug 23 11:50:15 iskander_work: that's weird Aug 23 11:50:32 anticom: exactly Aug 23 11:53:14 belen: the problem occurs only in the frame 'Other artifacts' Aug 23 11:53:39 the frame 'Images' is OK, i'm able to download images Aug 23 11:54:16 unfortunately, sdk is only visible in 'Other artifacts' Aug 23 11:55:19 the message 'This build did not create any image files' appears in the frame 'Images' and there is no link to download sdk Aug 23 11:55:38 'This is probably because valid image and license manifest files from a previous build already exist in your .../poky/build/tmp/deploy directory. You can also view the license manifest information in Toaster. ' Aug 23 12:19:13 has anyone experienced some issues with qtwebengine? it crashes when i run my application on an intel machine generated by yocto Aug 23 12:21:50 are you able to get a stacktrace? we use qt a fair amount on arm, maybe i might be able to help Aug 23 12:23:47 Has anyone attempted sshfs for sstate cache sharing? Aug 23 12:24:20 actually i tried to compare a non-working execution vs a working one by using strace, and it leads to a mmap system call, so i guess there is something wrong with the memory Aug 23 12:30:37 belen: all paths are truncated after 100 characters Aug 23 12:31:13 does it ring a bell ? maybe some constant for max path length in SQL db ? Aug 23 12:33:27 iskander_work: that's really annoying. We have removed that section and improve the artifacts display in our dev branch, but that doesn't help you on krogoth. If you have access to the machine running the builds, you can grab them from their location I guess … Aug 23 12:41:23 damn Aug 23 12:41:27 mysql> describe orm_buildartifact; Aug 23 12:41:27 +-----------+--------------+------+-----+---------+----------------+ Aug 23 12:41:27 | Field | Type | Null | Key | Default | Extra | Aug 23 12:41:27 +-----------+--------------+------+-----+---------+----------------+ Aug 23 12:41:27 | id | int(11) | NO | PRI | NULL | auto_increment | Aug 23 12:41:30 | file_name | varchar(100) | NO | | NULL | | Aug 23 12:41:32 | file_size | int(11) | NO | | NULL | | Aug 23 12:41:35 | build_id | int(11) | NO | MUL | NULL | | Aug 23 12:41:37 +-----------+--------------+------+-----+---------+----------------+ Aug 23 12:41:41 file_name is only 100 chars :( argh Aug 23 12:41:56 as i thought Aug 23 12:42:06 i have to change sql db Aug 23 12:42:36 belen: where can i find the script which creates the database ? Aug 23 12:45:43 iskander: I guess you need to change the db model. /bitbake/lib/toaster/orm/models.py Aug 23 12:46:08 iskander_work: the class might be BuildArtifact Aug 23 12:50:00 anyone have any idea here? http://paste.baserock.org/okapuyejah Not applying any appends, and successfully building for other targets (gtk+3 in krogoth) Aug 23 12:53:29 CTtpollard looks like this is needed https://github.com/GNOME/gtk/commit/2f3cb31e55d698fe6ddcc8126b9d1342268a1a98 Aug 23 12:54:53 belen: bitbake/lib/toaster/orm/migrations/0001_initial.py, right ? Aug 23 12:55:28 jku: ah ok, it could be due to the specific bsp kernel version Aug 23 12:55:34 jku: thanks Aug 23 12:56:49 iskander_work: I would assume it should be /bitbake/lib/toaster/orm/models.py but I could be completely wrong Aug 23 12:58:02 belen: i found it Aug 23 12:58:12 it's django stuff Aug 23 12:58:20 migrations.CreateModel( Aug 23 12:58:20 52 name='BuildArtifact', Aug 23 12:58:20 53 fields=[ Aug 23 12:58:20 54 ('id', models.AutoField(verbose_name='ID', serialize=False, auto_created=True, primary_key=True)), Aug 23 12:58:23 55 ('file_name', models.FilePathField()), Aug 23 12:58:26 56 ('file_size', models.IntegerField()), Aug 23 12:58:28 57 ('build', models.ForeignKey(to='orm.Build')), Aug 23 12:58:31 58 ], Aug 23 12:58:33 59 ), Aug 23 12:58:56 models.FilePathField(), i checked django docs, the default length is 100 :) Aug 23 12:59:16 theoretically, i have to patch this python script and create the db anew Aug 23 12:59:26 iskander_work: nicely done :) I hope it works Aug 23 12:59:50 why only 100 for a file path, that's not much on UNIX :( Aug 23 13:00:15 i wonder why noone stumbled upon it until now Aug 23 13:01:46 iskander_work: not many people are building sdks with toaster :) we did know the artifacts stuff wasn't working very well, but hadn't come across this specific problem before Aug 23 13:02:05 np, i'll fix it Aug 23 14:03:12 belen: i'm testing my 'BuildArtifact' patch, does it make sense to submit a bug report ? Aug 23 14:05:26 iskander_work: no need. I would just submit the patch to the mailing list and explain the problem in the commit message. That will do Aug 23 14:06:03 ok Aug 23 14:06:23 iskander_work: if you can, add [krogoth] to the subject line. That tells the reviewers the patch is for 2.1 Aug 23 14:07:43 good Aug 23 14:08:06 will do, thanks for support Aug 23 14:48:04 Is samba completely out of the question for sharing sstate cache? Aug 23 14:48:55 sveinse: I never really got shared sstate to work. share mirroring works perfectly, though Aug 23 14:48:58 I've just spent 2 hours trying to get nfs4 up and running on our server... It is certainly iffy, and I still haven't worked out all the iffies Aug 23 14:49:43 boucman_work: Share mirroring? Like in http you mean? Aug 23 14:50:28 in any way... but my company never has two project with the same MACHINE/yocto version, so I never pushed that very far... Aug 23 14:51:15 I notice that the build server does not create the same sstate objects as our users do, so feeding back to a common sstate cache would be a lifesaver in terms of time Aug 23 14:51:46 I dunno why the server is different from the user's thou. Same OS, but different host CPU (xeon vs i7) Aug 23 14:53:27 same host compiler ? (or are you using uninative) Aug 23 14:53:57 same ubuntu based compiler, yes Aug 23 14:55:26 hmm Aug 23 14:55:41 * boucman_work never had to debug signatures so far, so can only provide limited help... Aug 23 14:56:31 sveinse: this section just got added btw, in case you need to debug signature stuff: http://www.yoctoproject.org/docs/2.2/ref-manual/ref-manual.html#usingpoky-viewing-task-variable-dependencies Aug 23 14:57:42 by looking at the siginfo files, you can see what variables tasks depend on and what values those variables had when the signature was generated Aug 23 14:58:33 maybe you could run bitbake-diffsigs on a local siginfo file and one from the server Aug 23 14:58:49 Ulfalize, thx, I hadn't spotted it Aug 23 14:58:50 just a curious note: I've built the same project on the build server as on the server with completely independent sstate caches. Very many siginfo files which equal filename have different contents on the two machines Aug 23 14:58:58 boucman_work: added yesterday i think :P Aug 23 15:00:04 hehe Aug 23 15:00:07 sveinse: maybe pickle files aren't deterministic Aug 23 15:00:18 That surely must be a stopper for rsync'ing user's sstate cache to the central server, as there apparently are several versions for the same hash Aug 23 15:00:24 so that they can hold the same data but still be different Aug 23 15:01:34 Then one must trying inspect the object to determine if it should be uploaded... This is not going to be optimal Aug 23 15:02:24 most likely you're only meant to have one machine that updates the central sstate cache Aug 23 15:02:31 and then a bunch of clients Aug 23 15:02:52 that seems like the simplest solution at least Aug 23 15:03:29 yep. I had hoped that the build server could do that. But we see that even when the server has built everything, our clients have to make lots (hours) of local building despite sstate cache. Aug 23 15:03:44 Server rebuild takes 90 seconds, so the cache is working :o Aug 23 15:04:04 yeah, probably something wonky with the cache configuration in that case Aug 23 15:04:10 *20minutes for wiped tmp and rebuild Aug 23 15:04:33 hmm. how to proceed.... Aug 23 15:05:20 could check how many _setscene tasks run on a client after removing the local sstate cache Aug 23 15:05:33 those correspond to stuff accelerated via the sstate cache Aug 23 15:05:41 so would be a sanity check to see if it can get anything from there Aug 23 15:06:04 the _setscene tasks run in a separate pass before other tasks btw Aug 23 15:13:50 when sstate mirror is setup, it will download from the mirror and into the local sstate cache, right? Aug 23 15:14:23 so the local sstata cache will become a mixture of already cached object and newly generated Aug 23 15:14:51 yeah, guessing it has to work like that Aug 23 15:18:12 sveinse: yes, unless you use NFS, where only symlinks are added Aug 23 15:18:23 allright. Then a scheme for syncing can be: don't compare file contents or time stamps in the cache. Just assume if the file is present both places, it represents the same contents Aug 23 15:18:53 neverpanic: I have spent too long time getting nfs up today, and are close at giving it up Aug 23 15:19:38 been googling my problems, but it seems nfs is no longer very widespread Aug 23 17:44:47 I'm having trouble finding the proper way to add nginx into a yocto build. Aug 23 17:45:24 Since the nginx files do not seem to use PROVIDES or RPROVIDES, any attempt at making a dependency on nginx fails with "Nothing RPROVIDES 'nginx'" Aug 23 17:56:08 How do I force Yocto to regenerate bblayers.conf? Aug 23 17:58:22 hweaving: it's oe-init-build-env that creates it, only when it doesn't exist. remove/rename the existing file and re-source the setup script and its hould re-create it from the template Aug 23 18:06:55 kergoth: That was my mistake, I was trying to recreate using bitbake Aug 23 18:09:34 ah, yeah, bitbake reads it, it doesn't create it :) Aug 23 18:47:53 Hey, wondering if someone can clear up how the downloads folder works when git repos are used. I see that when I change the requested commit in the recipe and rebuild, it changes the .done stamp, and the git2 clone of the repo is unchanged. But when I try git commands on the repository, it doesn't work because it was a bare clone. How does it checkout the right commit if it was a bare clone? Aug 23 18:53:34 there's a local clone from DL_DIR into WORKDIR when the recipe is unpacked Aug 23 18:53:43 the checkout is done in the repo in workdir (${WORKDIR}/git, specifically) Aug 23 18:54:08 fetch = download from upstream into DL_DIR, unpack = unpack/copy/extract/clone from DL_DIR to WORKDIR Aug 23 18:57:19 kergoth: Or ${WORKDIR}/git2 ? Aug 23 18:57:52 no. git2 is a subdir of WORKDIR changed from git for historical reasons Aug 23 18:58:05 Ah Aug 23 18:58:13 subdir of DL_DIR, that is Aug 23 18:58:22 the dir in WORKDIR is always git, unless overridden Aug 23 18:58:32 WORKDIR == the dir buried down under tmp/work/ Aug 23 18:58:37 DL_DIR == downloads Aug 23 19:10:40 * kergoth sighs Aug 23 19:12:09 ? Aug 23 19:13:00 just noticed my bitbake shallow git tarball support has been broken since may, including the current patches on the list waiting for richard to review Aug 23 19:13:03 * kergoth bangs head against desk Aug 23 19:14:17 yeah, I think it's time to write more unit tests. would give both richard and myself more confidence, and increase my chances of getting it merged before feature freeze Aug 23 19:15:33 :) Aug 23 19:17:20 I'm still not 100% happy with this, but it meets a need, and anything better would require more hefty changes to the fetch core that wouldn't be ideal at this juncture Aug 23 19:17:37 guess ppl aren't beating down the door to be early adopter/testers... :-/ Aug 23 19:18:22 that's how it usually goes with changes of any kind that touch anything core, everyone has their own issues to deal with without adding more risk to their builds :) Aug 23 19:18:47 course in this case it has no real effect unlesss you opt into it by setting a variable, but still Aug 23 19:18:51 * paulg can't help but think of Winnebago Man when he hears "do me a kindness" or "at this juncture" Aug 23 19:22:46 Ah I'm running an old yocto where git2 is still being used, does it still clone into workdir from DL_DIR/git2? Aug 23 19:26:11 ? Aug 23 19:26:21 DL_DIR/git2 is used by everyone today, that hasn't changed Aug 23 19:26:25 so yes Aug 23 19:29:49 Ah sorry, alright, thanks Aug 23 19:32:19 Is there a way to avoid a checkout in the workdir every time? I think back when we were on openembedded, tarballs with checked-out versions of the repo at the commit were stored in DL_DIR and unpacked without needing to checkout again Aug 23 19:44:16 oh, this must be what you are talking about with shallow git tarball support? Aug 23 20:50:40 hi, I would like to apply patch to opencv recipe from meta-oe. This recipe pulls several git repositories, and from what I see all patches are applied to only one of them Aug 23 20:50:50 namely the main opencv repo Aug 23 20:51:05 but my patch is to contrib repo which is also pulled Aug 23 20:52:10 here is the recipe: http://git.openembedded.org/meta-openembedded/tree/meta-oe/recipes-support/opencv/opencv_3.1.bb?h=krogoth Aug 23 21:18:12 maciejjo: there's a url parameter to control the path (relative to S) where the patch is applied. i.e. ;patchdir=foo or ;patchdir=../foo Aug 23 21:23:21 thanks, that did the job Aug 23 21:26:53 np **** ENDING LOGGING AT Wed Aug 24 02:59:58 2016