**** BEGIN LOGGING AT Sat Aug 22 02:59:58 2015 Aug 22 04:05:19 there we go, i think this way of doing the sstate distro fallbacks is a bit better than what i was using before: https://github.com/MentorEmbedded/meta-mentor/commit/cd551d83b5750f8eb5565d4a188b5ef6b18180f7 Aug 22 05:19:40 hello Aug 22 05:20:34 weston not starting with yocto layer for beagle bone black Aug 22 10:04:50 kergoth: that sanity checking is causing problems for me too Aug 22 10:58:59 RP, good analysis of the whitelist class Aug 22 14:36:52 Hi, what the difference between BUILDDIR and TOPDIR ? Aug 22 14:38:07 I set the TOPDIR to my new build folder before souring the oe-init-build-env, but didn't work as expected Aug 22 14:54:22 BUILDDIR is used/set by the setup scripts, TOPDIR is used/set by bitbake. they're commonly the same, but not guaranteed to be. you shouldn't set or modify TOPDIR, there's no need, and as you've seen, the setup scripts don't obey it Aug 22 15:08:08 Ok, Thanks. So the only way to start with a different build folder is call oe-init-build-env* with the build parameter Aug 22 18:47:49 i've successfully create an image which contains qt5 (based on meta-qt5). now i want to create an SDK. i did a bitbake myimage -c populate_sdk. this creates an SDK me but there is one problem. the image does _not_ contain the host-qmake program that is needed to cross-compile for my target Aug 22 18:48:08 how do i do this? Aug 22 18:48:41 how do i create an SDK that also contains the needed host tools for qt cross development? Aug 22 18:54:55 either use meta-toolchain-qt5 or add the right nativesdk package for qmake to TOOLCHAIN_HOST_TASK Aug 22 18:55:00 afaik anyway Aug 22 18:56:18 e.g. TOOLCHAIN_HOST_TASK_append = " nativesdk-qtbase-tools qtbase-mkspecs" Aug 22 18:56:40 something of the sort Aug 22 19:15:15 ideally -cpopulate_sdk shoud know it and do it automagikally Aug 22 19:17:21 yeah, would be nice to have a metadata driven mechanism 'if this is in the image, then these extra packages need to go into the sdk' Aug 22 19:17:22 woot, I've gotten closer. I'm now down to just four files in the rootfs with unexpected permissions/ownership. Aug 22 19:17:27 ... group, passwd, shadow, gshadow. Aug 22 19:17:56 kergoth: thanks! Aug 22 19:18:00 will try that Aug 22 19:18:34 i don't know enough about yocto, though populate_sdk would do the trick Aug 22 19:19:42 kergoth: will/would meta-toolchain-qt5 pull in the dev packages for all the libs i have in my image? Aug 22 19:19:57 no, that's a standalone toolchain recipe, not image-bound Aug 22 19:20:05 thought so Aug 22 19:20:34 but TOOLCHAIN_HOST_TASK_append = " nativesdk-qtbase-tools qtbase-mkspecs" would do the trick? if i just add this to my local.conf? Aug 22 19:21:09 after doing that a -c populate_sdk would contain qmake? Aug 22 19:22:47 kergoth: ^ Aug 22 19:26:05 if those are the correct package names, yes. that's untested Aug 22 19:26:43 kergoth: i'll have a look. thanks for your support! Aug 22 19:26:55 np Aug 22 19:31:34 seebs, meh, those aren't important files... :) Aug 22 19:59:14 kergoth: TOOLCHAIN_HOST_TASK_append = " nativesdk-packagegroup-qt5-toolchain-host" worked nicely Aug 22 20:01:18 nice, glad to hear it Aug 22 21:14:11 so I studied it more and it got worse, there are more things resulting in weird db entries than I expected, so I am going to do more studying to see if I can figure out what's causing them. Aug 22 21:14:22 d'oh Aug 22 21:14:28 I think I just realized what one of them is, though. Maybe. Aug 22 23:02:43 Beaverton, OR Aug 22 23:02:58 doh copy/paste error Aug 22 23:32:36 Well, I just found a fascinating thing that I'm surprised has never broken anything before. Aug 22 23:32:57 So I have this fancy code in pseudo that tries to make sure that files are opened with a *real* mode that is not 000. Aug 22 23:33:02 And allows root to read/write them. And this mostly works. Aug 22 23:33:14 But if you set umask 777, and create a file, it actually *does* get created mode 000 at first. Aug 22 23:33:24 And that doesn't matter, because the first chmod op on it fixes the mode. Aug 22 23:33:34 ... But it would matter if you were going to try to setfattr() on it to store attributes locally. Aug 22 23:34:06 And if you then fix the mode later, the "but can I store attributes to this file at all?" test succeeds, and concludes that nothing has been saved to that file because it has no database entry, while there actually IS a database entry for that file. Aug 22 23:34:08 Whoops. Aug 22 23:35:43 And that may well explain a number of database-weirdness things I was seeing. Reason it was affecting only passwd/group is those are probably the only things being initially set up in a code path that appears to create files with umask 777. Aug 22 23:40:24 wget-finder is handy Aug 22 23:41:59 So I found a bug that only shows up with the xattr version, and have probably fixed it now. Aug 22 23:42:01 $ umask 777 Aug 22 23:42:03 $ touch file Aug 22 23:42:06 $ chmod 600 file Aug 22 23:42:57 At this point, there's a db entry in sqlite for "file", but the xattr stuff doesn't check for it because it can write to file's extended attributes just fine. So it grabs the existing owner/mode and stores them in the extended attributes along with the new permissions. Aug 22 23:43:03 Existing meaning "in the real filesystem". Whoops. **** ENDING LOGGING AT Sun Aug 23 02:59:58 2015