**** BEGIN LOGGING AT Wed Oct 28 02:59:58 2015 Oct 28 07:24:16 Morning. I've added a new header file to linux, and I'm trying to figure out how to make it includable from userspace (meaning it must be added to the sdk). Oct 28 07:36:24 tasslehoff: i'd say the basic point is, make sure that the kernels install_headers process also properly includes it. Oct 28 07:37:53 LetoThe2nd: is that something in the kernel tree or something in openembedded? Oct 28 07:38:21 tasslehoff: kernel Oct 28 07:39:11 LetoThe2nd: ah. never seen those scripts before :) Oct 28 07:52:34 If I run 'make headers_install' I find my header file in usr/include/linux/spi/, but it is not present in the sdk. Oct 28 08:12:16 morning all Oct 28 08:12:29 morning Oct 28 08:34:26 tasslehoff: The kernel recipe isn't used to generate userspace headers. You need to add the header to the linux-libc-headers recipe (e.g. via bbappend). At least that's what I've done in the past. Oct 28 08:36:53 andre_d_: tasslehoff: worth reading the header at the top of linux-libc-headers.inc before doing that: http://cgit.openembedded.org/openembedded-core/tree/meta/recipes-kernel/linux-libc-headers/linux-libc-headers.inc Oct 28 08:40:11 great! thanks andre_d_ && bluelightning Oct 28 08:41:19 bluelightning: I would contend that there are cases where it is necessary to modify linux-libc-headers :-). In tasslehoff's case it probably isn't indeed, though. Oct 28 08:41:36 andre_d_: most likely not :) Oct 28 08:42:43 andre_d_: Richard's comment doesn't rule that out, but it does lead to some fallout though so it's worth considering whether it's the right place to make the change, whatever that change might be Oct 28 08:54:43 for me it's just a do_install_append where I somehow copy the file I need from STAGING_KERNEL_DIR? Oct 28 09:13:11 Hi. Do sombody know how to enable spread spectrum in uboot for imx6 processors? I found some information here https://community.freescale.com/docs/DOC-100036, but in my uboot i don't see flash_header.S file. In fact i don't see it in freescale official uboot either. Is this instruction obsolete? Oct 28 09:17:01 mornin Oct 28 09:21:27 sorry if I'm being dense here, but could someone give me an example of a do_install_append that adds an include file to linux-libc-headers? Oct 28 09:41:59 ah. nevermind. found an example in meta-ti Oct 28 09:46:00 bluelightning: I'll continue to debug today, it happened again here (the md5sum problem) Oct 28 09:46:23 Jin^eLD: ok, thanks Oct 28 09:47:52 I'll add my findings to the ticket Oct 28 09:48:41 maybe not. is there a recipe that can show me how to do this? I'm in foreign territory now.. Oct 28 09:51:17 tasslehoff: so am I unfortunately :( I've just added https://bugzilla.yoctoproject.org/show_bug.cgi?id=8596 although that doesn't really help you immediately... Oct 28 09:51:44 tasslehoff: this is associated with some actual code in the kernel right? Oct 28 09:52:36 bluelightning: yes. a custom spi driver Oct 28 09:53:04 ok, can the header not be installed as part of the kernel then? Oct 28 09:53:52 I guess all that means is your userspace code needs to have the appropriate -I option pointing under ${STAGING_KERNEL_DIR} Oct 28 09:54:03 bluelightning: yes, but I don't know how to do that either :) Oct 28 09:56:05 bluelightning: you mean if the userspace code is compiled in yocto? Oct 28 09:56:18 tasslehoff: right - is it not? Oct 28 09:56:30 bluelightning: no. I'm trying to get the header file into the sdk. Oct 28 09:58:07 hmm, so worst case could you not just create a trivial recipe that inherits nativesdk and installs the header into ${D}${includedir}, then add the -dev package produced by that to your SDK ? Oct 28 09:59:37 that's probably not the best solution but it seems to me it ought to work Oct 28 09:59:38 bluelightning: sounds like that will work. that also should inherit virtual/kernel to get to the header file? Oct 28 10:00:02 it should have a DEPENDS on virtual/kernel if that's what you mean Oct 28 10:00:27 bluelightning: that is what I mean :) Oct 28 10:06:19 guys, how can I do a cleansstate on all recipes that compose an image? Oct 28 10:06:28 is there any builtin way of doing so? Oct 28 10:07:18 will just delete tmp/ do so? Oct 28 10:15:44 bluelightning: initial attempt: http://pastebin.com/HtV2FdiM Oct 28 10:19:18 ftonello: for cleansstate? no, you'd need to rm -rf both sstate-cache and tmp for that Oct 28 10:19:24 ftonello: what problem are you trying to solve? Oct 28 10:24:00 Again this same question. Really nobody know how to do that? Oct 28 10:24:04 Do sombody know how to enable spread spectrum in uboot for imx6 processors? I found some information here https://community.freescale.com/docs/DOC-100036, but in my uboot i don't see flash_header.S file. In fact i don't see it in freescale official uboot either. Is this instruction obsolete? Oct 28 10:26:27 bluelightning: inheriting crosssdk worked, but not inheriting nativesdk. That gives: Nothing PROVIDES 'virtual/x86_64-angstromsdk-linux-gcc' Oct 28 10:26:39 erm Oct 28 10:26:47 crosssdk is definitely not what you want Oct 28 10:27:19 bluelightning: no. that was just an experiment :) Oct 28 10:27:31 if you can't build a simple nativesdk recipe then how on earth are you building the SDK? Oct 28 10:27:53 can you pastebin the recipe? Oct 28 10:28:06 bluelightning: already did. :) Oct 28 10:28:15 oh, so you did, sorry Oct 28 10:28:46 np. btw, I build the sdk with "bitbake -cpopulate_sdk my_rootfs" Oct 28 10:28:51 bluelightning: lunch. back in 20. Oct 28 10:46:55 bluelightning: do you by chance know who generates/keeps up to date Packages.stamps ? Oct 28 10:47:20 Jin^eLD: I don't, I would have assumed opkg-make-index Oct 28 10:49:09 hmm indeed, I first thought maybe its generated outside Oct 28 10:53:39 bluelightning: ok... here's something interesting Oct 28 10:53:53 the mtime in the Packages.stamps is indeed the same as the mtime of the file when checked via python console Oct 28 10:54:20 this is why the checksum is not recalculated because opkg-make-index assumes that the file has not changed since the last run Oct 28 10:55:14 ok, so it's whatever is in the stamps file that it uses to determine what needs updating? Oct 28 10:55:34 yes Oct 28 10:55:44 stamps is some sort of a cache to optimize things Oct 28 10:55:56 the question is how does that stamps file get updated in advance of the md5sum getting updated... Oct 28 10:55:56 so it does not recalc all md5sums always but only for packages that have changed Oct 28 10:56:00 exactly... Oct 28 10:56:48 either they aren't properly tied together, or an error occurred during the md5sum calculation and it blindly carred on and updated the stamp Oct 28 10:57:05 two possibilities, at least... Oct 28 10:57:19 possible... I'll now try to figure out who or what is responsible for updating the stamps file in the first place and then see where I get from there Oct 28 11:01:26 bluelightning: did the recipe seem sane? Oct 28 11:01:42 tasslehoff: near as I can tell, I have no idea why it wouldn't work Oct 28 11:01:56 tasslehoff: if you drop the DEPENDS does it make a difference? Oct 28 11:02:08 or comment it out, anyway Oct 28 11:02:54 bluelightning: no. same error. Oct 28 11:17:10 tasslehoff: what's this recipe called? Oct 28 11:20:10 bluelightning: misc-linux-headers.bb Oct 28 11:20:59 bluelightning: ok.. thanks.. I am trying to use another linux kernel source, but I believe I already worked that out. Oct 28 11:21:03 tasslehoff: and you're building nativesdk-misc-linux-headers right? Oct 28 11:21:22 bluelightning: I did a cleansstate on the old recipe and it worked.. Thanks :) Oct 28 11:21:35 bluelightning: no! I did not know I should prefix it. Oct 28 11:25:22 what makes the nativesdk- prefix valid. it doesn't work for my recipe Oct 28 11:25:31 the inherit nativesdk Oct 28 11:25:49 hm. I though so. for some reason that does not work. Oct 28 11:25:58 which version of the build system are you using? Oct 28 11:27:14 bluelightning: 1.8 Oct 28 11:27:47 oh Oct 28 11:27:53 I'm being an idiot, sorry Oct 28 11:28:15 bluelightning: no problem. nice to have company :p Oct 28 11:28:41 rename your recipe to nativesdk-misc-linux-headers.bb Oct 28 11:28:58 I'm guessing it won't actually fix the issue but at least that's what it ought to be called... Oct 28 11:29:32 you might as well add INHIBIT_DEFAULT_DEPS = "1", that may fix the error (and you don't need those dependencies anyway) Oct 28 11:30:27 still puzzled as to how you'd be getting that error in the first place though Oct 28 11:31:27 bluelightning: renaming the recipe fixed the build error. Oct 28 11:31:40 hmm, ok, interesting Oct 28 11:31:56 you can put back the DEPENDS now I guess Oct 28 11:32:15 bluelightning: that was there when I compiled Oct 28 11:32:19 ok great :) Oct 28 11:32:39 bluelightning: is adding this recipe to TOOLCHAIN_TARGET_TASK_append now correct? Oct 28 11:32:52 yes, that would be the next step Oct 28 11:33:23 actually, hmm Oct 28 11:34:05 maybe nativesdk isn't correct here and I've led you down the wrong path, since that would be for the host side of things... Oct 28 11:35:39 bluelightning: mhm.. problem is that it seems this stamps mess needs to happen only once, and if it does you're screwed because it won't update the md5sum until the ipk really changes Oct 28 11:35:53 and I can't seem to be able to figure out how or why that "first time" happens Oct 28 11:36:05 tasslehoff: ok, sorry to be a pain, but I'm now thinking the correct thing would be to rename the recipe back to what it was, drop the inherit nativesdk, then add misc-linux-headers-dev to your TOOLCHAIN_TARGET_TASK_append Oct 28 11:36:29 bluelightning: will try :) Oct 28 11:37:17 Jin^eLD: well as for workarounds you can just delete the index files or move them out of the way - or was that not what you meant? Oct 28 11:37:45 I can come up with a workaround, that would not be a problem, I wanted to figure out why this is happening so we get a proper fix... Oct 28 11:37:59 at some point it must go wrong Oct 28 11:38:19 and once the stamps file gets an updated mtime without the md5sum being updated its "over" Oct 28 11:38:40 because then it never will be unless the ipk really undergoes a change or unless Packages.stamps is deleted Oct 28 11:38:59 right, I get that... but that's not the heart of the issue though Oct 28 11:39:10 surely? Oct 28 11:40:09 the heart of the issue is that a current m_time is written to the stamps without an md5sum update in the Packages.. at least thats my idea of it Oct 28 11:40:20 right, that's a given Oct 28 11:40:30 open question is the "why" or "how" that happens Oct 28 11:40:49 the stamps file is always rewritten Oct 28 11:41:08 except for the case where there would be an exception Oct 28 11:41:09 why doesn't mtime change always trigger an md5 update? Oct 28 11:41:25 yes that is exactly the question... Oct 28 11:41:33 that "first time" when its happening would be interesting Oct 28 11:41:36 only digging through the code will answer that... Oct 28 11:41:55 I did not see anything obvious yet Oct 28 11:42:46 no matter if the package was in stamps already or not: it will get written there Oct 28 11:44:07 well, eventually I'll find it Oct 28 11:44:17 so far I could not see any obvious flaws in the code Oct 28 11:45:51 but at this point I guess I can only leave in my debug messages, remove the stamps file and hope for the problem to happen again Oct 28 11:46:13 bluelightning: I don't think my recipe does much now. I don't inherit anything, and all I have in there is a do_install :) Oct 28 11:46:19 * tasslehoff needs infrastructure Oct 28 11:46:55 tasslehoff: in this case there doesn't need to be much Oct 28 11:50:32 "except IOError: pass" ... that's not great... Oct 28 11:51:05 I can understand ignoring ENOENT but that will ignore any kind of error reading that file Oct 28 11:59:47 bluelightning: true. feel I'm closing in now. the -dev package contains what I want, but has an unsatisfied dependency to misc-linux-headers Oct 28 12:02:38 -dev is the only non-empty folder in packages-split Oct 28 12:18:33 bluelightning: the final piece of the puzzle was RDEPENDS_${PN}-dev = "" Oct 28 12:18:57 at least I hope it was the final piece.. and the right puzzle.. Oct 28 12:25:11 tasslehoff: ah right yes, I should have remembered that Oct 28 12:27:57 bluelightning: it worked like a charm. thanks for helping. Oct 28 12:28:05 np Oct 28 15:40:56 can somone remind me how to build gfortran Oct 28 17:06:58 why is qemu being built even for non-qemu machines? Oct 28 17:09:52 qemu-native is used by some other recipes Oct 28 17:09:57 like everything with fonts Oct 28 17:10:35 locale generation as well, IIRC Oct 28 17:12:19 dv_: there's a few places at rootfs time where something approximating the target machine is sufficient to pregenerate files such as locales or fontconfig caches, which can't be generated in an arch-independent way Oct 28 17:13:11 qemu-native doesn't build on ubuntu 15.10 Oct 28 17:13:21 the config.log shows errors like libpulsecommon-6.0.so: undefined reference to `dbus_watch_get_enabled@LIBDBUS_1_3' Oct 28 17:13:44 yeah Oct 28 17:13:47 there s a bug for that Oct 28 17:13:48 one sec Oct 28 17:14:37 8553 Oct 28 17:14:40 has a workaround too Oct 28 17:14:44 "qemu: drop hooking of libsdl from the host" ? Oct 28 17:14:57 oh ok Oct 28 17:15:44 hours of linkage fun Oct 28 17:15:54 :) Oct 28 17:16:13 otavio: found out anything about the sysroot gcc issue you saw in the autobuilders? Oct 28 17:16:52 we are switching our codebase here to jethro, but have a freakin weird bug with gcc 5 Oct 28 17:17:10 so we are considering to file a bug report and stick to 4.9 for now Oct 28 17:28:30 dv_ which bug? Oct 28 17:29:17 dv_: yes; we fixed it Oct 28 17:29:20 dv_: it was on our side Oct 28 17:30:06 if anyone cares about prelink -- I've tracked down one issue with glibc-2.22 (not in 2.21) that prevents prelink working properly on IA and ARM.. Oct 28 17:30:21 (working on a fix... :) Oct 28 17:31:38 fray: it was going to be disabledfor 2.0, no? Oct 28 17:32:59 yes.. Oct 28 17:33:05 but that is because it doesn't work safely.. Oct 28 17:33:08 "I'm fixing that" Oct 28 17:33:23 it'll likely go into a 2.0.1 if I can Oct 28 17:33:29 otavio: oh, and it is in meta-fsl-arm now? Oct 28 17:34:01 dv_: no; it was our autobuilder setup Oct 28 17:34:07 dv_: now fixed and working Oct 28 17:34:43 hm perhaps I had a similar issue. what was the problem? Oct 28 17:36:07 dv_: when a build fails we cannot change the machine and reuse tmp Oct 28 17:36:23 dv_: i can explain to you by voice latr Oct 28 20:49:22 is there anything in the krnel do_path that modifies new files flags? Oct 28 20:59:27 urg Oct 28 20:59:28 reset Oct 28 21:00:01 So I have a patch made from git that adds a python script with the exec flag set to the sourcce Oct 28 21:00:12 late in the build, something will run that python script Oct 28 21:00:58 does bitbake do_patch respect the Oct 28 21:00:59 diff --git a/tools/zynq-boot-bin.py b/tools/zynq-boot-bin.py Oct 28 21:00:59 new file mode 100755 Oct 28 21:01:05 new file mode in the diff? Oct 28 21:02:16 also behvaior is different betwen Fedora 20 and 21 Oct 28 22:11:08 BTW the fix for prelink on IA32 systems: https://sourceware.org/ml/libc-alpha/2015-10/msg01028.html Oct 28 22:11:18 (I'm going to look into ARM next as it suffers from the same problem) Oct 28 22:24:14 * Crofton is going insane Oct 28 22:24:23 go with it.. it helps Oct 28 22:24:25 one build pathces a file in and sets exec flag fine Oct 28 22:24:29 other masks it off Oct 28 23:33:39 FFS, most evil OE issue in ages Oct 28 23:33:54 had to copy patch binary from F20 machine to F21 Oct 28 23:45:10 F21 is doing something wrong or? Oct 28 23:49:00 yeah Oct 28 23:49:03 email sent Oct 28 23:49:17 copying patch from F20 to F21 "fixed" my problem Oct 28 23:49:37 I attempted to use patch-native, but that fails with a depends loop :) **** ENDING LOGGING AT Thu Oct 29 02:59:59 2015