**** BEGIN LOGGING AT Tue Nov 28 05:51:43 2006 Nov 28 07:57:39 yo Nov 28 09:29:21 hey koen Nov 28 09:29:32 hey rwhitby Nov 28 09:29:44 you koen Nov 28 09:29:54 hey XorA Nov 28 09:30:04 koen: part of my problem 12 hours ago was that MACHINE=nslu2 does not do the ARCH_BYTE_SEX override that gets done for ixp4xx Nov 28 09:30:17 so I'm rebuilding angstrom now with MACHINE=ixp4xx Nov 28 09:30:53 koen: the prebuilt images at http://www.angstrom-distribution.org/unstable/images/20061127/ixp4xx/ don't work for most nslu2-linux devices, cause the jffs2 erase block size is incorrect. Can that be changed? Nov 28 09:31:04 sure Nov 28 09:31:16 s/can be/should be/ :) Nov 28 09:31:39 we're also working on removing the need for the 6 different kernel images, by using Apex as a second stage bootloader. Nov 28 09:31:47 s/should be/will be/ ;D Nov 28 09:57:13 koen: even with MACHINE=ixp4xx, the libgcc ipk still ends up as an i686 package Nov 28 09:58:03 hmmm Nov 28 09:58:43 was the http://www.openembedded.org/viewmtn/diff.psp?id1=2152f3464031a4bf3e4bc5c8203b46ebb0e44dff&id2=9a4fb914b7b1a293ffcc388e0301d3158ea42794 patch meant to change TARGET_ARCH to PACKAGE_ARCH ? Nov 28 09:58:58 (via MY_ARCH) Nov 28 09:59:27 yes Nov 28 09:59:40 otherwise it will say 'arm' instead of 'armv5te' Nov 28 09:59:58 or 'i586' instead of '' Nov 28 10:00:02 'arm' would be better for me ;-) Nov 28 10:00:07 (than i686) Nov 28 10:01:12 the work directory for gcc-cross is meant to be tmp/work/i686-angstrom-linux-gnueabi/..., right? Nov 28 10:01:45 (that's what it is for me) Nov 28 17:30:52 angstrom@serenity:~/website/unstable$ find . -name *.ipk | wc -l Nov 28 17:30:53 26251 Nov 28 17:31:17 after deleting 16k packages from the morgues Nov 28 17:31:44 that shows we have ~10k locale packages Nov 28 17:33:35 locale + debug, that is Nov 28 20:17:43 koen: my libgcc1 packages have now changed from i686 to armeb :-( Nov 28 20:18:00 they should be arm (el) Nov 28 20:18:19 then where you have armeb defined at all? Nov 28 20:18:36 very good question Nov 28 20:18:48 it uses PACKAGE_ARCH, so if they get armeb, everything should get armeb Nov 28 20:18:57 I'm building angstrom 2007.1 for machine ixp4xx with a very simple local.conf Nov 28 20:18:58 bitbake -e gcc-cross|grep armeb Nov 28 20:19:16 or even bitbake -e -b gcc-cross_4.1.1.bb|grep armeb Nov 28 20:19:28 http://pastebin.ca/260956 Nov 28 20:20:17 ARCH_BYTE_SEX = "be" Nov 28 20:21:16 right - the question is where is that being set Nov 28 20:21:34 include/ipx4xx.conf Nov 28 20:21:41 conf/machine/include/ipx4xx.conf Nov 28 20:22:11 ARCH_BYTE_SEX_ixp4xx = "le" Nov 28 20:22:25 in conf/distro/angstrom-2007.1.conf Nov 28 20:22:42 which should override that machine default Nov 28 20:22:48 112 MACHINE_ARCH = "ixp4xx${ARCH_BYTE_SEX}" Nov 28 20:23:22 yep - I'm getting ixp4xxle-... for my kernel and kernel modules ipks Nov 28 20:23:41 so that "le" line is working for some packages, but not others. Nov 28 20:23:55 98 TARGET_ARCH = "${@['armeb', 'arm'][bb.data.getVar('ARCH_BYTE_SEX', d, 1) == 'le']}" Nov 28 20:23:59 here? Nov 28 20:24:50 which file is that in? Nov 28 20:25:18 conf/machine/include/ipx4xx.conf Nov 28 20:25:28 ah, so it's an include file ordering problem? Nov 28 20:25:41 distro is after machine Nov 28 20:25:51 I don't understand how koen got it to build then. Nov 28 20:25:58 so distro will need to recreate machien stuff probably Nov 28 20:26:19 ugh Nov 28 20:26:41 which is why I stayed away from ixp stuff so long Nov 28 20:26:42 or include conf/machine/ipx4xx.conf again Nov 28 20:27:09 rwhitby: I use multimachine, so the armv5te toolchain gets built the first time for another machine Nov 28 20:27:53 any suggestions on how I fix this? Nov 28 20:30:01 try to include conf/machine/ipx4xx.conf again? Nov 28 20:34:42 how do other machines/distros in angstrom do switching between be and le ? Nov 28 20:34:51 not rather Nov 28 20:34:59 (we want to do things the "angstrom way") Nov 28 20:36:08 (or does angstrom not yet handle non-default-endian situations?) Nov 28 20:36:17 it does not Nov 28 20:36:30 ipx4xx are the only one OE machines with changing sex Nov 28 20:36:52 ok, so instead of doing an ixp4xx-specific hack, we should work out how to do it properly in OE Nov 28 20:37:18 yes Nov 28 20:37:58 which is why nail and I suddenly hide when people talk about that ;) Nov 28 20:38:17 ok, I've run out of time right now, but will be back and available in about 2 hours if that's a convenient time to discuss this Nov 28 20:38:31 (it's time to get ready for work here) Nov 28 20:38:39 I'll probably be asleep Nov 28 20:38:45 but feel free to mail the OE list :) Nov 28 20:39:23 I wonder why SlugOS doesn't fall over this - it's using the same machine.conf ... Nov 28 20:39:43 local.conf? Nov 28 20:40:15 nope Nov 28 20:40:57 ARCH_BYTE_SEX is set in (for instance) slugos-bag.conf Nov 28 20:46:54 of course the simple solution is to change the sex default in ixp4xx.conf (since angstrom only does le for arm eabi) Nov 28 20:47:07 (since we know that slugos BE can change it successfully) Nov 28 22:17:58 hooray, setting default ARCH_BYTE_SEX to le in ixp4xx.conf finally gives me arm packages :-) Nov 28 22:18:08 * rwhitby now waits for the rest to build so he can flash and test. Nov 28 22:21:50 rwhitby: a day of curing freaky shit then :-) Nov 28 22:22:26 XorA|gone: same as every day .... ;-) Nov 28 23:53:23 * rwhitby 's angstrom build completes **** ENDING LOGGING AT Wed Nov 29 02:59:58 2006