**** BEGIN LOGGING AT Fri Aug 02 03:01:07 2019 Aug 02 04:28:18 Hi - yocto/bitbake question: is there a configswitch (or similar) that controls the output format of the yocto-sdk? or is there only the self-extracting shar? Aug 02 04:29:18 (looking for a way to generate a .deb instead) Aug 02 04:56:25 jschneider: we only self extracting shar Aug 02 05:31:41 Good Yocto community! Aug 02 05:33:14 Are other software archaeologists who have to use (-.-) clearcase fetcher ccrc out there? Aug 02 05:38:47 The fetcher, (as it is in warrior/master), does not work properly - I will provide a fix/change soon and post it to the bitbake mailing list. It would be smart to double check the change with another ClearCase setup. Damn, ClearCase is a load of crap ... -.- Aug 02 05:43:08 Chrusel: I haven't noticed any issues or patches there lately. So sure go ahead, patches welcome. Aug 02 05:49:16 @khem: ah, ok - thanks! Aug 02 07:47:58 Hi, is this correct? The name of the layer e.g. ‘meta-mylayer’ doesn’t have any affect on file priority (which file ends up on the image) but the folder underneath does e.g. ‘recipes-core’ but the folder under that also doesn’t e.g. ‘base-files’? Aug 02 07:49:50 bentech: yeah, the name is totally irrelevant. Aug 02 08:01:25 Hi all, I'm debugging a size related issue. We're upgrading from krogoth to thud with very few changes in our layers (mostly fixing the build) so aside from build fixes, recipes are identical. Aug 02 08:02:14 The cpio image more than doubled (66M to 140M) and from the list of installed packages I can see a non-negligeable increase in number of python3 packages installed in the image Aug 02 08:03:04 But I'm quite lost on where to start, since it appears to be using this python3-manifest.json thing for dependencies. Aug 02 08:03:32 Does anyone have any hint on where to start? Aug 02 08:52:33 qschulz: do a buildhistory diff before and after the update. may be as simple as openssl 1.1.x depending on perl... Aug 02 08:55:34 I find the recipe-depends.dot almost impossible to use, especially after creating an svg out of it Aug 02 08:56:53 python3-core is also 14 times bigger in thud and I've got a cpp package (coming from gcc AFAICT) taking 18Mo Aug 02 11:00:31 10:30 pike| c'est un simple spam sur une mailing ou je suis Aug 02 11:00:38 ah... well :D Aug 02 11:00:54 http://cgit.openembedded.org/meta-openembedded/commit/meta-python/recipes-devtools/python?id=befa59e6de53ac72ebf5876156eb9192598967c8 Aug 02 11:01:15 This saved me 50MB on the additional 80MB I have Aug 02 12:07:58 * alessioigor waves all Aug 02 12:08:06 Is there a way to override or extend create_merged_usr_symlinks_rootfs() in meta/classes/image.bbclass whitout change the latter? Aug 02 12:08:09 Thanks! Aug 02 12:10:31 New news from stackoverflow: "audit" doesn't get included in Yocto final image Aug 02 12:16:35 RP: Is there anything different about the environment of the build worker? I have a rootfs postprocess command that is failing only at the worker; I can't reproduce it at my desk or when I manually run it at the worker. It's a "find" call, like: "find dir/{a,b} ...". Aug 02 12:31:24 tgoodwin: there shouldn't be Aug 02 12:31:34 The behavior is as if it's running in sh instead of bash. Aug 02 12:32:12 tgoodwin: we do use /bin/sh and scripts are meant to be sh safe Aug 02 12:34:05 Good to know. It's funny I've never seen this fail before. Aug 02 12:35:50 tgoodwin: different /bin/sh link desktop to worker? Aug 02 12:37:03 Yep... that was it. Ubuntu 18.04 server has it linked to dash vs. desktop is linked to bash. Aug 02 12:37:19 Thanks Aug 02 12:42:08 rburton, right it's a bit of a mess. I'd say we should standardize on virtio (you do *not* need working virgl to have it). Aug 02 12:42:29 basically because that is the future, and is also where everyone else is heading :) Aug 02 12:46:34 kanavin: virtio has always made a lot of sense for the qemu machines Aug 02 12:46:42 just didn't exist when we set out Aug 02 12:47:45 RP: there's talk about convincing SoC vendors to support virtio in their hypervisors, so that the guest Linux doesn't need SoC specific drivers at all Aug 02 12:48:00 I personally can't see the likes of nvidia doing that, but who knows :) Aug 02 12:48:43 both because it surely affects gpu performance, and reduces the level of lock-in from the hardware vendor Aug 02 12:49:43 kanavin: that is an interesting one :) Aug 02 12:50:04 RP: in the cloud it has already happened more or less Aug 02 13:13:38 RP: have you ever seen "chown failed -- inappropriate ioctl for device" when running populate_sdk? That's another one I cannot seem to reproduce at my desk. Aug 02 13:14:07 tgoodwin: no Aug 02 13:41:18 RP: also getting "python callback ??? failed, aborting" with populate_sdks. Aug 02 13:45:55 tgoodwin: That doesn't sound familiar Aug 02 14:35:26 RP: I'm at a loss for this one. I thought it might be the lack of a shell for the build user, so I set it to /bin/bash, logged in as the user and ran bitbake. The populate_sdk task works fine. Aug 02 14:36:03 I let buildbot re-run the task and it fails. Aug 02 14:40:18 tgoodwin: can you pastebin the actual error. Context may help me Aug 02 15:23:41 RP: I was inheriting rm_work. Once I removed that, populate_sdk worked. Aug 02 15:25:03 tgoodwin: that is interesting. I wonder why that was breaking things? :/ Aug 02 15:26:03 RP: Right. And exclusively when buildbot is running things. Aug 02 15:26:43 tgoodwin: I'd love to understand how that happens Aug 02 15:27:42 With rm_work enabled, I was noticing that there was no recipe-sysroot-native for the image, and the log for populate_sdk showed it was trying to run dnf from there. The log.task_order indicated rm_work had already run. Aug 02 15:34:30 tgoodwin: oh, that would indeed break it Aug 02 15:35:45 RP: yeah it's like the one beat the other to the punch. Down in the populate_sdk log it showed an error when it tried to execute dnf, but the log continued for pages including a DNF version output like it got executed somehow. Aug 02 15:36:03 weird stuff Aug 02 15:37:07 tgoodwin: its because populate_sdk is a leaf task, it isn't a dependency of do_build Aug 02 15:37:18 tgoodwin: rm_work adds itself before do_build Aug 02 16:08:31 RP: congrats on getting the hash euqivalency work done! Aug 02 16:10:11 kanavin: thanks :) Its a start, not sure its "done" yet! Aug 02 16:12:18 RP: it's a bit ironic that I am reluctant to advocate this feature here, because we are trying to convince management to buy even more powerful workstations and build servers for yocto work. Aug 02 16:12:31 saying that 'yocto will get faster' is a bit counter to that :D Aug 02 16:13:23 kanavin: I can see the dilemma :) Aug 02 16:14:03 kanavin: Reality is it will help for some things, not others Aug 02 16:15:05 RP: also I got a ticket for ELC-E, not sure if you or others are coming, but just fyi Aug 02 16:15:14 this year I should be in good health to finally make it there Aug 02 16:16:07 there are direct flights Berlin-Lyon too apparently, which is nice Aug 02 16:18:13 kanavin: I'm intending to go and have budget approval for that one. Not sure I can get direct flights but will figure it out Aug 02 16:19:32 Plane to Paris, train to Lyon Aug 02 16:22:16 Crofton|work: maybe, may be easier to fly via AMS Aug 02 16:23:17 RP: cool, thanks Aug 02 16:24:49 sure, but for me, this is the way to go Aug 02 16:26:12 I would take a train any time over flying Aug 02 16:26:17 TGV is nice :) Aug 02 16:26:42 and there is a station directly in Paris CDG airport Aug 02 16:27:31 travel time is 2 hours apparently Aug 02 16:28:10 but not frequent service to Lyon. Might spend a night or two in Paris and inspect beer places Aug 02 16:31:05 kanavin: I tried trains to get to Grenoble for an ELC-E once. It didn't work out so well Aug 02 16:33:44 rp has a cloud that follows him around Aug 02 16:34:57 tools-sdk doesn't include gdb ..... Aug 02 16:39:22 Crofton|work: nah, was just unfortunate last time... Aug 02 17:39:33 okay so in the patch to fix yesterday's pseudo patch Aug 02 17:39:51 you will notice that i am replacing "flags|O_NOFOLLOW" with "flags&O_NOFOLLOW" as a test for "whether O_NOFOLLOW is present in flags" Aug 02 17:40:22 i just want to state for the record that i'm actually sort of proud of managing to do something that insanely stupid, and especially of the fact that it *actually passed any tests at all*. Aug 02 18:02:14 nice work seebs Aug 02 18:07:07 ha, i'm always amazed by the stupid mistakes i make at times, usually something miniscule like that with large effects. single character makes a lot of difference.. Aug 02 18:15:19 rburton: yeah removing qemu user mode didnt improve things for me Aug 02 18:15:37 rburton: I had already disabled gobject introspection which is the only thing that it dit Aug 02 18:16:16 but stuff like chromium is wired in to execute qemu as well, and apparently the v8 engine in it requires to run code on QEMU, and AFAIC I cant disable it the same way Aug 02 18:17:06 I probably will just have to wait for skylake-client/avx support on qemu upstream Aug 02 18:18:35 Crofton|work: I'd be interested in ispecting beer places in Paris as well Aug 02 18:19:22 the beer scene isn't grim Aug 02 19:08:50 seebs: It goes to show you also how much of a corner case it was. There were not actually that many failures. It did expose something in the OE QA that isn't supposed to be happening. Aug 02 19:09:12 fray was separately send out some patches for that problem. Aug 02 19:09:42 This time around all the regression testing is green so far. Aug 02 21:02:02 kanavin: the problem with virtio (and why i said standardize on std first) is that virtio-gpu is only enabled in qemu for x86 and one other arch that i can't remember, so we're going to be hitting a lot of codepaths that have never been tested before Aug 03 01:42:37 New news from stackoverflow: Install .deb package on a Yocto system **** ENDING LOGGING AT Sat Aug 03 02:59:57 2019