**** BEGIN LOGGING AT Thu Aug 20 02:59:59 2015 Aug 20 05:46:51 <[w00t]> Hello Aug 20 05:49:42 <[w00t]> Is it possible to check out two git repos in your SRC_URI with different GITDIRs? I get an error in do_fetch when I've got two repos, but individually they both work. The error is "sh: 0: getcwd() failed: No such file or directory" Aug 20 05:54:02 <[w00t]> nevermind :) Aug 20 09:11:27 morning all Aug 20 10:08:07 morning bluelightning... although it is some time from then :) Aug 20 10:08:46 abelal: http://www.total-knowledge.com/~ilya/mips/ugt.html Aug 20 10:08:55 but it is still morning here FWIW ;) Aug 20 10:09:31 good morning Aug 20 10:25:28 well that's enlightening :) Aug 20 10:25:34 bluelightning: ^ Aug 20 10:48:02 hello channel Aug 20 10:48:26 I am having a weird thing using meta-qt5 and was wondering if anyone has an idea Aug 20 10:49:18 for the yocto environment, qt.conf is generated in https://github.com/meta-qt5/meta-qt5/blob/master/recipes-qt/qt5/nativesdk-qtbase_git.bb and Sysroot is set Aug 20 10:49:46 but for the SDK, it is generate from https://github.com/meta-qt5/meta-qt5/blob/master/classes/populate_sdk_qt5.bbclass and Sysroot is not set Aug 20 10:50:10 the result is that a .pro file that uses QT_SYSROOT will work under Yocto but not under the SDK Aug 20 10:50:17 I was wondering if this is intentional Aug 20 10:58:00 aaaanyway, I wil probably post to the list then Aug 20 10:58:05 maybe with a patch already Aug 20 10:58:08 we'll see Aug 20 10:58:44 it's a bit chilly for this time of year, don't you think? Aug 20 13:27:32 I wish the meta-iot-middleware layer people would stop copying recipes out of meta-oe Aug 20 13:27:45 and if they really need to, please credit where the recipe came from Aug 20 13:36:53 rburton: ^^^ Aug 20 13:38:15 Crofton: you mean meta-intel-iot-middleware ? Aug 20 13:38:33 http://layers.openembedded.org/layerindex/branch/master/layer/meta-intel-iot-middleware/ says it depends on meta-oe Aug 20 13:40:53 interesting Aug 20 13:41:10 http://layers.openembedded.org/layerindex/branch/master/recipes/?q=zmq Aug 20 13:41:12 hopefully the whitelist class will stop all of that Aug 20 13:42:16 whitelist class? Aug 20 13:42:52 well if you paid attention to oe-core instead of swanning around caving and at hack camps... Aug 20 13:42:52 besides the duping, the thing that really annoyes me is the failure to credit the source Aug 20 13:42:59 rofl Aug 20 13:43:05 so much to pay attention to Aug 20 13:43:11 http://permalink.gmane.org/gmane.comp.handhelds.openembedded.core/67780 Aug 20 13:43:50 that Crofton .. always out having fun! Aug 20 13:43:52 excellent idead Aug 20 13:44:25 within hours of arriving at camp I was getting OE feedback Aug 20 13:45:00 Strange... getting errno=No route to host when doing 'recipetool create -o test.bb git://gitlabserver/Test/test.git;protocol=ssh;user=git' even though I can ssh to gitlabserver. Has anyone else tried protocol=ssh with git fetcher? Aug 20 13:45:54 I'll be amused by the first person to argue against the whitelist using the "We need to make sure the recipe is maintained" argument Aug 20 13:52:00 Crofton: you seem to be too energetic this morning after your vacation - don't worry, it will wear off quickly... :) Aug 20 13:52:09 * zeddii experienced that this week. Aug 20 13:52:14 crushed in two hours Aug 20 13:52:19 * zeddii sobs Aug 20 13:52:54 * paulg_ works hard at crushing zeddii 's will to live. Aug 20 13:53:14 * zeddii hands a bit more of his dead soul to paulg as this weeks payment Aug 20 14:02:22 denix, I am still in CLT Aug 20 14:02:28 waiting for plane at 1120 Aug 20 14:04:20 Crofton: flying home? it's just few hours drive from there... Aug 20 14:05:48 I know Aug 20 14:05:59 I woul dneed to rent a car and find my lugguage Aug 20 14:06:28 by the time I dealt with overhead flying is better Aug 20 14:06:47 there was no way I could have drove home last night without falling asleep Aug 20 15:38:43 how is called the concept of virtual/any or RUNTIME_any variables in yocto? Aug 20 15:39:32 an example that what I'm talking about is virtual/kernel variable Aug 20 15:45:46 what exactly is the question? Aug 20 15:46:52 So, need an opinion, I'm going to submit my host-user-contaminated test to oe-core insane.bbclass, do folks think it should be in WARN_QA or ERROR_QA by default? See https://github.com/MentorEmbedded/meta-sourcery/blob/master/classes/package_qa_sourcery.bbclass#L23 Aug 20 15:51:52 looking the documentation i answer my own question, was looking about the PREFERRED_PROVIDER and VIRTUAL-RUNTIME Aug 20 15:55:03 aj_c: VIRTUAL-RUNTIME is just a convention for variable naming, there's nothing special about it Aug 20 15:55:32 the virtual/ prefix for "virtual" targets is mostly just convention as well although there are some isolated bits of code that look at it Aug 20 15:56:39 bha, another flight cancelled Aug 20 15:57:10 I retreated to the lounge, i have a seat on the next flight, but I think there is about to be a riot Aug 20 16:40:18 JaMa: for python-pyparted should we use a git tag instead? I don't see another publicly available tarball Aug 20 16:40:42 hmm Aug 20 16:48:29 moto-timo: just set SRCREV to match with version tag Aug 20 17:14:19 JaMa: you got it :) The perl modules I'll do another way with patches Aug 20 19:28:53 kergoth: sounds like it _should_ be an error. No doubt that will trip someone up but they should fix the error. Aug 20 19:29:20 Errror now, revert to warning if people squeal with the plan of making it a hard error for 2.1 ? Aug 20 19:43:37 Hi, I'm new to yocto creating a recipe for my package, how can I build a package for the BUILD machine to be part of the SDK? Aug 20 19:44:06 BBCLASSEXTEND += "nativesdk", then add nativesdk-yourrecipe to TOOLCHAIN_HOST_TASK to add it to the sdk Aug 20 19:44:33 nativesdk is like native, except the output of the build will run on SDKMACHINE, not the build machine, and it emits packages for use in the sdk Aug 20 19:45:56 And that must be a separate recipe that the one used to build the HOST packages? Aug 20 21:26:59 WIP from meta-mel: https://github.com/kergoth/poky/compare/master...kergoth:for-upstream Aug 20 21:27:04 lets see, what next.. Aug 20 22:56:30 Hey, if people want to experiment with New Pseudo, there is now a new branch in the pseudo git tree. seebs/reshuffle. Adds a profiling feature, a couple of reasonably mild performance tweaks, and a major feature allowing it to be configured to use extended attributes (whenever possible) instead of IPC to a database server. Aug 20 22:57:17 Performance is... slightly odd. On smallish real tasks like using tar, it appears to reduce the pseudo performance hit by 30-50% for a lot of workloads. On the entire-rootfs-build-process, it may actually make wall clock time worse, and I've not managed to track down why. Aug 20 22:57:35 But! It's around if people are curious or want to see what madness I've been up to. Aug 20 22:58:26 I've always been wondering if I could patch pseudo to hide files a recipe doesn't explicitly depend on to increase build reproducibility Aug 20 22:58:40 I guess that's quite a large endeavour, though. Aug 20 22:59:06 Huh. Interesting thought. Getting the list in place would be hard. Aug 20 22:59:25 But it's a neat idea. Some kind of "don't allow access to any file unless it's being-created or is already in the database". Aug 20 22:59:27 Yeah -- and the performance hit would probably also be significant Aug 20 23:01:30 or you could just do all your builds in a chroot, so even if its influenced by the host, it at least always does so the same way :) Aug 20 23:02:16 aboriginal has an interesting technique, though it only helps for binaries, it replaces the user's PATH with a path it creates that has symlinks to the specific host binaries it needs to have available Aug 20 23:02:20 so e.g. /usr/bin isn't in the PATH Aug 20 23:02:39 My concern is autoconf automatically picking up a dependency that's in the sysroot but I don't want it to pick up Aug 20 23:02:46 doesn't help other inspections of host files, but interesting nontheless, particularly since it's pretty trivial to do Aug 20 23:03:02 That basically means using a separate chroot (or sysroot) for each recipe build Aug 20 23:03:07 neverpanic: I think we really need per-recipe sysroots constructed from DEPENDS. Aug 20 23:03:09 right Aug 20 23:03:30 we have the info available in the metadata to do it, now, at least a task has access to its dep info, thats how the aclocal copying in autotools.bbclass works. Aug 20 23:03:36 Yeah, we've discussed that at work, and we may just end up implementing that and providing a patch some time in the future Aug 20 23:03:40 it'd need to figure out the sstate archive paths or something Aug 20 23:03:42 For what it's worth, one of the things I found out was that my beliefs about where the performance hits of pseudo were coming from were mostly wrong. Aug 20 23:03:55 So one of the changes is to not constantly allocate/free paths, and that seems to help noticably. Aug 20 23:03:57 then we'd want to cache the sysroots its constructing for performance, for common combinations Aug 20 23:04:20 seebs: heh, thats always fun.. profiling is good at blowing apart our assumptions :) Aug 20 23:04:24 heh Aug 20 23:04:43 Yeah, that's always the question, when does creating the sysroots start becoming less overhead than doing it at runtime with LD_PRELOAD Aug 20 23:04:53 Basically, the "simple" code that just computes canonical paths appears to consume a great deal of time. Aug 20 23:04:57 seebs: So the IPC isn't really the bottleneck? Aug 20 23:05:12 It's *a* bottleneck, but a very large reduction in IPC doesn't have as much effect as I anticipated. Aug 20 23:05:24 Basically, the more you reduce the IPC cost, the less it matters compared to other parts. Aug 20 23:05:57 That's good to know, I have a similar pet-peeve project where I was thinking about using a lock-free shared memory cache to avoid the constant IPC overhead, but know that I hear this, I should probably fire up a profiler first… Aug 20 23:06:07 The other thing I'm thinking about is... There's a lot of things that are using paths, and I think there's a fair number of cases where I might not actually *need* the canonical path unless something happens with it that will require storing something in the database, and if I could avoid computing them except in those cases, that might save time. Aug 20 23:06:18 hmmmm Aug 20 23:06:29 FWIW, that was how fakeroot did it, and the nightmarish problems we had with that in longer builds were why pseudo exists. Aug 20 23:07:06 Although some of those problems were specific to a previous build system which was doing more long-term persistance than we do now. Aug 20 23:08:17 What really stunned me was when I found out that fakeroot had an option to store path names instead of device/inode pairs. Aug 20 23:08:19 And then I read the implementation. Aug 20 23:08:27 For each device/inode pair: Walk the filesystem looking for it. Record the path. Aug 20 23:09:08 Fun fact: OS X has a direct way to do this kind of lookup. I don't know how they do it, but it's possible. Aug 20 23:09:51 basically /.vol/$fsid/$inode gives you that file Aug 20 23:12:19 seebs: good god, that's unpleasant Aug 20 23:15:28 The reason they can do that, as I recall, is that their underlying FS implemenation doesn't really HAVE inodes, so the inode numbers are basically already in a database. Aug 20 23:16:22 * paulg is pretty sure seebs gets to deal with more unpleasantness than the rest of us on a daily basis. Aug 20 23:16:40 Not in terms of work. But Do Not Ask about my hobbies. Aug 20 23:17:17 (Turns out: If you have an open anonymous advice request thing, you end up attracting a lot of abused kids. It's sort of spectacular.) Aug 20 23:21:15 things can't be good if folks are turning to the internet to get advice... :-/ Aug 20 23:21:54 such is the world we live in Aug 20 23:48:40 hmm, if libjpeg-turbo is default in debian, fedora, opensuse, and chrome's own builds, why is it still relegated to meta-oe? Aug 20 23:49:39 "turbo" is so 1985... Aug 20 23:51:03 what the.. the jpeg recipe, for a linuxstdbase build, symlinks libjpeg8 to libjpeg6 Aug 20 23:51:15 but 8 broke compat with 7 which broke compat with 6 Aug 20 23:51:25 so that would certainly cause problems Aug 21 02:43:53 Hmm, to add teh host user contamination test to image.bbclass, will have to define its vars in an INHERIT'd class or config file. guess bitbake.conf would be appropriate Aug 21 02:44:00 hmm Aug 21 02:44:07 i hate adding more to that, though Aug 21 02:44:19 maybe i could just make the rootfs check dependent on the inherit of insane Aug 21 02:54:22 that's not exactly pretty either, adds an implicit cross-inherit dependency **** ENDING LOGGING AT Fri Aug 21 02:59:58 2015