**** BEGIN LOGGING AT Fri Jan 06 03:00:01 2017 Jan 06 10:54:16 pohly: Looking at your anonymous python ordering code, I'm trying to figure out if there is any better API we could use Jan 06 10:55:06 RP: I'm open for suggestions, but couldn't come up with something myself. Jan 06 10:55:14 pohly: Whilst numbers do kind of solve the problem, we don't have any real defined meaning to them. I'm even wondering if some kind of dedicated events would be better Jan 06 10:55:40 pohly: I have sort of similar but different problems in the recipe specific sysroot code Jan 06 10:56:14 I can imagine that events might work. Jan 06 10:56:59 What I find problematic about events is mostly lack of documentation: in which process do they fire? What can be done in each event? Jan 06 10:57:23 pohly: we should add a section to the manual about them, that would make a good bug Jan 06 11:01:18 pohly: the event section of the bitbake manual is horrible, I think I might quickly write a new version Jan 06 11:03:46 RP: it's not just bitbake - there also events in OE-core, albeit less of them. Jan 06 11:04:23 Not sure whether those deserve their own documentation. Jan 06 11:05:34 pohly: there should be comparatively few of those, mostly toaster related? Jan 06 11:13:46 pohly: https://bugzilla.yoctoproject.org/show_bug.cgi?id=10886 - see if my proposed changes help at all Jan 06 11:13:47 Bug 10886: normal, Medium, 2.3 M2, srifenbark, NEW , Improve event documentation Jan 06 11:14:26 pohly: I guess describing each event in the first section would be the next step Jan 06 11:15:04 pohly: another question - would we want to replace the completion scheduler with your rm_work one? Jan 06 11:41:39 Hi, how to create a secound u-boot image with another environment? Jan 06 11:42:10 rburton: not sure what happened to mut :/ Jan 06 11:42:43 RP: no just staring at that now Jan 06 11:43:16 urgh, I haven't done SWAT yet today :-/ Jan 06 11:43:39 if its mut, don't bother. doing it now. Jan 06 11:44:28 rburton: I looked and then decided to come back later when you were online Jan 06 11:44:46 new musl broke connman Jan 06 11:46:47 lots of beaglebone/wic failures and a patch from ed Jan 06 11:49:41 chmod: cannot access '/srv/www/vhosts/autobuilder.yoctoproject.org/pub/sstate//de/sstate:libxml2:aarch64-poky-linux:2.9.4:r0:aarch64:3:de9ad622922ce2c9f57457aeea02ab33_package.tgz.rRaDVA6L': No such file or directory Jan 06 11:49:53 i do love that race Jan 06 11:51:04 rburton: I wish I knew how that was possible :/ Jan 06 11:51:52 RP: yes, long-term the completion scheduler could be replaced. Would you prefer to do that immediately? We could continue to use the same name, so nothing would break for anyone (somehow) using that directly. Jan 06 11:52:41 pohly: I think I would, yours seems to be much better and the intent of both is the same. We shouldn't really refer to rm_work in bitbake Jan 06 11:53:11 pohly: did my comments in that bug answer any of your events questions or is there more info needed? Jan 06 11:53:40 pohly: does on the heartbeat event as a further comment might be good! :) Jan 06 12:46:50 pohly: "Interestingly enough, builds with rm_work.bbclass were also faster than those without." <— what is this magic Jan 06 12:48:04 how :O Jan 06 12:49:37 rburton: my theory was that the page cache became more efficient because it could drop cached data for deleted files. There was an interesting drop in the number of cached pages that only occurred when using rm_work.bbclass - see the OE-core mail thread for a screenshot. Jan 06 12:50:07 interesting theory Jan 06 12:50:13 how much memory does your build machine have? Jan 06 12:50:35 rburton: 32GB Jan 06 12:50:55 * rburton looks forward to replicating :) Jan 06 12:53:12 rburton, pohly: Its all down to machine setup. We run the performance tests with and without rm_work on the same hardware and its marginal, some machines win with rm_work, some win without. The numbers with rm_work are usually more stable though Jan 06 12:53:22 will be interesting to see the effect the patches have though Jan 06 12:53:44 pohly: do you have a measure of how much faster? Jan 06 12:53:52 on this machine, enabling rm_work was a small performance hit Jan 06 13:14:47 RP: I posted numbers in the "rm_work + pybootchart enhancements" mail thread. For core-image-sato, build time was 45:38.18 without rm_work and 42:58.54 with the enhancements. Jan 06 13:15:32 However, building Ostro OS XT got a bit slower (1:52:59 -> 1:53:13). Jan 06 13:17:18 pohly: right, that sounds about the same as the perf test machines Jan 06 13:20:49 * Crofton|work wonders how repeatable those build times are Jan 06 13:53:20 pohly: here turning on rm_work made the build about 30 seconds slower, close enough to be noise to be honest Jan 06 13:54:24 rburton: fair enough. It would also be interesting to compare the normal scheduler with the "rm_work" scheduler, without actually using the rm_work.bbclass. Jan 06 14:06:53 pohly: yes! Jan 06 14:07:26 Crofton|work: we have perf test scripts running on machines which show variance of around a minute but trends do become clear with averaging Jan 06 14:07:43 sure Jan 06 14:07:49 I was curious about the variance Jan 06 14:08:19 Crofton|work: I'm trying to get all the performance data visible as a git repo... Jan 06 14:08:54 marquiz: where are we at with that? :) Jan 06 16:00:56 RP: today is a public holiday in Finland Jan 06 16:03:15 sgw_: systemd-boot in meta-intel doesn't want to fetch here Jan 06 16:03:28 ERROR: systemd-bootchart-231+gitAUTOINC+3a74d4fc90-r0 do_fetch: Fetcher failure: Unable to find revision 3a74d4fc90cb322a4784a3515bef7118c8f8c5ba in branch master even from upstream Jan 06 16:06:34 rburton: gah Jan 06 16:08:11 not too keen on how meta-intel pins the revision of systemd-boot tbh Jan 06 16:08:14 rburton: that's systemd-bootchart not meta-intel Jan 06 16:08:32 ah Jan 06 16:08:38 oh, the old wildcard was _boot% Jan 06 16:08:48 * rburton updates Jan 06 16:09:00 ed2: I hadn't realised thanks Jan 06 16:10:01 RP: how is it going with release build? Are we lucky with sdk_ext installer? Jan 06 16:10:07 rburton: does the autobuilder always include the meta-intel layer for all builds including non-intel? Jan 06 16:10:43 sgw_: don't think so Jan 06 16:12:09 sgw_: rburton: if it does it's probably a bug Jan 06 16:12:38 * joshuagl has an issue to ensure any changes to .conf files are undone as appropriate Jan 06 16:12:42 ed2: We've decided now we understand the other problems with rc1, we'll released rc1 as was and concentrate on M2. Your patches are merged, passed testing on the AB and will unblock 2.2.1 though Jan 06 16:17:25 rburton: fixed the boost x32 patches to use Incompatible [configuration] on the branch, if you want to swap them on mut, or at least before it hits master-next. thanks again Jan 06 16:17:28 * kergoth gets coffee Jan 06 16:17:30 RP: great! thanks Jan 06 16:18:21 kergoth: thanks for those patches, its nice to see that fixed :) Jan 06 16:18:52 kergoth: have you any thoughts on ordering of anon python and task api? Jan 06 16:18:56 np. i haven't done a ton of testing on target, other than making sure it boots, though. will do more testing and run some ptests Jan 06 16:18:56 kergoth: thanks Jan 06 16:19:10 rburton: where is that error coming from, which build?, I confused it with the next line from YoctoAutoBuilder Jan 06 16:20:05 kergoth: dropping anon python in favour of specific events is one thing I keep wondering about. There is the problem of event handler ordering too though Jan 06 16:20:39 sgw_: locally, but its been fixed in meta-intel already Jan 06 16:21:38 Yeah.. I'm not a huge fan of the priority scheme in that patch, but we clearly need something. I like the idea of moving to more event handlers and adding explicit deps between handlers of the same event via the addhandler, but that doesn't solve the rm-work need, since you'd have to list each handler. I also think it'd be worth adding a syntactic shortcut to reduce the syntactic overhead of event handling. i.e. addhandler do_something at ConfigParsed Jan 06 16:21:38 or at bb.event.ConfigParsed or so. Jan 06 16:25:22 kergoth: the rm_work issue would be dealt with through a "preTaskFinalise" type event? Jan 06 16:25:38 kergoth: I like the idea of better syntax, that was bothering me earlier but I hadn't thought of it that way Jan 06 16:25:55 kergoth: might be worth a bug Jan 06 16:25:59 or better a patch :) Jan 06 16:27:34 indeed :) Jan 06 16:27:37 * kergoth adds to his list Jan 06 16:33:37 pohly: I think I'm coming down in favour of a new event just before the tasks are processed Jan 06 16:36:11 kergoth: had to chuckle at the bbmake reference still in one of the files Jan 06 16:38:06 wow, i thought we'd gotten rid of those long ago. that is funny Jan 06 16:40:41 hey folks, I'm trying to add a kernel config fragment to a private layer, but it seems to be ignored. Jan 06 16:40:58 I'm guessing it's a layer priority issue, but could use adice on ways to figure out the issue. Jan 06 16:41:06 anyone have some advice on how to run this down? Jan 06 16:41:16 most kernel recipes dont' support fragments at all Jan 06 16:41:18 so i'd check that first Jan 06 16:41:54 kergoth: hmmm, good tip Jan 06 16:42:15 if it's not linux-yocto, and not the qoriq kernel, most likely it doesn't as is, but check the the recipe Jan 06 18:01:00 hi all Jan 06 18:01:19 seebs, kergoth you there? Jan 06 18:04:09 anyway... seebs, kergoth .. i don't know if you remember of me, i was close to throw beaglebone black and everything else out from the window 'cause i cannot obtain a booting image... i develop using a VirtualMachine [on VirtualBox]. I need USBPcap libraries installed on my Windows host because of another project i work at...removing it solved the problem, so USBPcap and VirtualBox usb layer are not friend at all. Uninstalling it allowed me to write Jan 06 18:04:09 correctly the image to a sdcard and booting the BBB. Thanks in any case for your help Jan 06 18:05:11 now... is there anyone that had experience on custom psplash logos ? Jan 06 18:21:53 I am trying to create a symlink in the my root that points to a directory on another partition that is mounted as /data, how can I do this? It keeps pointing to TMPDIR Jan 06 18:30:26 you should be able to make a broken symlink that resolves once the partition is mounted Jan 06 18:33:23 in my do_install of my recipe I have: ln -s ${D}/data/foo ${D}/foo but that doesn't work Jan 06 18:35:43 it creates this link: foo -> /home/builder/workspace/OS/build/tmp/work/cortexa9hf-neon-poky-linux-gnueabi/system-configuration/1.0-r0/image/data/foo Jan 06 18:38:36 Talorno: ah, interesting. glad you got it taken care of at least Jan 06 18:39:28 kergoth, probably that could be useful to someone else one day or another! :) Jan 06 18:39:34 * kergoth nods Jan 06 18:46:22 surprisewitness: that would be because ${D} is the build path. you don't want to use that in the target. Jan 06 18:46:38 surprisewitness: you want the link target to be the path you want the symlink to have in the rootfs Jan 06 18:46:44 ie, /foo Jan 06 18:47:02 there's a lot of misunderstanding about how ln works. often they assume the first argument is a path on disk, but it's just what the new link will point to.. Jan 06 18:47:11 get someone running into that fairly regularly, i swear Jan 06 18:47:54 i think that the argument order should be exactly reversed ,btw :P Jan 06 18:48:37 looks like what he wants as is, actually. he wants a symlink in tht eroot that points into /data Jan 06 18:50:22 that is correct kergoth Jan 06 18:58:18 thanks all, that worked Jan 06 18:59:15 why i cannot see my custom psplash boot logo even if i changed the bitmap .h file [generated by the needed script] and i bitbaked and booted the new image? Jan 06 19:00:01 how did you go about making the recipe use yours? most likely the problem is there. i.e. using a bbappend but forgetting to alter FILESEXTRAPATHS Jan 06 19:00:42 well....i've gone for the hard and brute way.... i swapped the file in the original psplash folder.... is it so bad? :D Jan 06 19:02:38 not a good idea for the long term, but as a gross hack just to validate that it works, i guess Jan 06 19:02:46 we don't recommend directly modifying any core layers at all Jan 06 19:02:47 use your own Jan 06 19:03:18 that said, clearly it should have shown up given what you did, i'd: bitbake -c clean psplash; bitbake -C fetch psplash Jan 06 19:04:08 i did not do -C fetch. What does the -C switch does? Jan 06 19:04:19 i just did clean and built the whole thing again Jan 06 19:05:54 just cleaning alone isn't going to make it rebuild from scratch Jan 06 19:05:59 it'll just pull it from sstate instead Jan 06 19:06:14 you'd need cleansstate or -C to force it to rebuild from scratch Jan 06 19:06:21 i did cleansstate Jan 06 19:06:22 *shouldnt'* be needed, but doesn't hurt to give it a go Jan 06 19:06:29 let me try again Jan 06 19:06:55 worth noting that cleansstate only removes the local sstate, if you're using an sstate mirror, it'll re-download and use it instead of building from scratch, that's why i developed the habit of using -C instead of cleansstate. but that's probably not an issue for you :) Jan 06 19:08:06 i will try again... today i'm too happy i reached kinda 8sec boot time to my app :D Jan 06 19:08:49 will bitbake -C download sources and stuff again? i mean... won't it delete the custom .h file i made? Jan 06 19:09:07 i don't understandt he question. where exactly did you put your .h? Jan 06 19:09:24 -C just makes it re-run fetch and all later tasks. fetch doesn't re-download when it doesn't have to Jan 06 19:09:36 so re-running the fetch task won't re-download anything Jan 06 19:09:56 i put my .h file in BLABLABLAH.../poky-morty/meta/recipes-core/psplash/files Jan 06 19:10:11 it will *never* remove a file from a layer Jan 06 19:10:17 layers are input only Jan 06 19:10:26 it writes to BUILDDIR and TMPDIR, that's it Jan 06 19:10:53 it's compiling again Jan 06 19:10:56 let's see Jan 06 19:11:57 whoa, working again 725 packages :D Jan 06 19:12:02 *tasks Jan 06 19:59:37 kergoth, i don't know if psplash writes stuff in the first stages of the kernel... do i have to write again boot also on my sd or rootfs is enough? Jan 06 20:06:02 uhm. rootfs was enough but still the logo doesn't comes Jan 06 20:27:30 hello Jan 06 20:30:52 odd problem here, i am building an app which uses pcap.h. In my recipe I have it as a depends. I link statically against it. It does not get installed on my sysroot when i build an sdk. I explictly added it to my image IMAGE_INSTALL_append += libpcap but it also does not produce a sdk sysroot with pcap.h. In my build dir i have a sysroot with /usr/include/pcap.h and /usr/include/pcap/pcap.h. Jan 06 20:31:54 hmm. ok I take that back. I did not the right find. it is there. Jan 06 20:31:56 nvm Jan 06 20:36:53 davis: wouldn't pcap.h be in libpcap-dev, anyway? Jan 06 20:43:42 neverpanic: yah, lol, i was coming back to say, scratch that I did the wrong find WHEN i found it. it is indeed still nto there. Jan 06 20:43:51 ill all libcap-dev and try again Jan 06 20:44:03 neverpanic: vilen danke Jan 06 20:46:44 you're welcome Jan 06 20:47:55 add libpcap to DEPENDS in your app recipe Jan 06 20:49:02 so you dont forget it when you add the app back into build system Jan 06 20:52:18 DEPENDS_class-target += "libpcap" Jan 06 20:52:24 i have that Jan 06 20:52:30 and it builds Jan 06 20:52:55 it just does not put the pcap.h that it uses in sysroots in the sdk sysroot Jan 06 20:53:51 libpcap-dev in my image install appends breaks the build 8 -( Jan 06 20:55:01 you cant add dev packages to IMAGE_INSTALL Jan 06 20:55:09 how do you generate your SDK Jan 06 20:55:22 -cpopulate_sdk ? Jan 06 20:55:29 yes Jan 06 20:55:43 one sec, ill find my exact cmd Jan 06 20:56:01 if your image does not need libpcap then it wont have it then Jan 06 20:56:06 bitbake -c populate_sdk someimgname Jan 06 20:56:10 ok Jan 06 20:56:24 you might want to specifically add it to SDK Jan 06 20:56:32 so this component i'm building uses pcap via a static link Jan 06 20:56:50 that is the only thing i think is causing it not be installed Jan 06 20:58:45 if you have it in build deps then it should be pulled into SDK as well Jan 06 20:58:58 but if your app is the only one needing it Jan 06 20:59:09 and your app is built externally Jan 06 20:59:21 then its not there in direct deps for the image yet Jan 06 20:59:29 unless you wire in your app Jan 06 20:59:29 hmm. Jan 06 20:59:40 but you can add it to SDK Jan 06 21:00:42 i'm thinking Jan 06 21:01:05 add TOOLCHAIN_TARGET_TASK += "libpcap" in local.conf Jan 06 21:01:57 ok many thanks Jan 06 21:04:28 im trying it now Jan 06 21:04:41 I sincerly appreciate it. tip of the hat to you khem **** ENDING LOGGING AT Sat Jan 07 03:00:01 2017