14:56 < zumbi_> /usr/mips-linux-gnu/bin/ld: skipping incompatible /usr/mips-linux-gnu/lib/libc.so when searching for -lc 14:57 < zumbi_> it happens with other arches too (the arches that have 32/64 bits ABI) 14:59 < zumbi_> i'm currently doing another build, so error log is not yet ready, in sometime it will get back to the error 15:02 < buxy> zumbi_: check /usr/share/perl5/Dpkg/Shlibs.pm if you want to add other directories to look into 15:04 < guillem> something I forgot to comment on the bug report about dpkg-cross support was that I think this is kind of a hack 15:05 < guillem> and I'd really like switching to use multiarch instead 15:05 < guillem> zumbi_: and hi 15:10 < codehelp> buxy: hi, the directory is in /usr/share/perl5/Dpkg/Shlibs.pm but dpkg-shlibdeps appears to find /usr/$crossprefix/lib and bail out 15:11 < codehelp> when it could go on to /usr/$crossprefix/lib32/ 15:11 < codehelp> and then bail out if that fails also 15:11 < codehelp> ditto for ../lib64 15:11 < buxy> I doubt this, maybe objdump bails out because it can't parse the files 15:11 < codehelp> ah 15:12 < codehelp> binutils-multiarch is installed 15:12 < codehelp> just checking on objdump 15:18 < codehelp> it's ok - looks like a bug in dpkg-cross 15:18 < codehelp> thanks for your help, buxy & guillem 15:30 < guillem> codehelp: also if we had proper multiarch support, you could stop doing the package name mangling, and path shuffling on those packages 15:31 < guillem> and cross-building would become a more natural and integrated thing 15:37 < codehelp> by proper multiarch support, you mean that the "foreign" binaries from mips would be automatically installed in /mips-linux-gnu/ ? 15:42 < guillem> codehelp: well only libraries (or objects you want to link to) would need that, and those would go into /lib// 15:42 < codehelp> yes, that's it 15:42 < codehelp> guillem: who is looking after multiarch support nowadays? 15:42 < guillem> you don't need crosscompiled binaries installed, those of course can be generated, but as normal packages with the normal target dirs 15:43 < guillem> there's a #multiarch channel and the few web pages 15:43 < codehelp> thanks 15:43 < guillem> codehelp: the problem has been mostly missing support from the toolchain 15:43 < guillem> so the rest has been mostly put on hold 15:43 < codehelp> it's the toolchain that we're working on (zumbi and I) here and next week 15:43 < guillem> last time binutils was missing a patch, and I'm not sure if gcc had a regression recently 15:45 < guillem> codehelp: well the patches are around, they have just not been applied 15:45 < codehelp> that's something I think Emdebian will need to work on for gcc-4.4 and Lenny+1 15:45 < guillem> codehelp: in my mind that be the clean way to get rid of dpkg-cross 15:46 < codehelp> true 15:46 < guillem> codehelp: as I said at the time, I'm fine with the current support in dpkg, as long as it's understood as a temporary workaround 15:46 < guillem> codehelp: there's some links at http://wiki.debian.org/multiarch/ 15:47 < codehelp> agreed - but we're coming up against bugs like this where the workaround is showing signs of flakiness 15:47 < codehelp> i.e. a reason to find the fix and implement it 15:50 < codehelp> wouldn't the package renaming still be needed, guillem? 15:50 < codehelp> even with multiarch, dpkg -l will need to show that libc6 and libc6-mips-cross is installed/ 15:50 < codehelp> even if we change the way the renaming takes place 15:50 < guillem> codehelp: supposedly once dpkg support multiarch you would be able to install both at the same time 15:51 < codehelp> how to tell that one is installed but one is not? 15:51 < guillem> codehelp: the tuple to distinguish the packages would include the arch for multiarch packages 15:53 < guillem> codehelp: Mithrandir already implemented initial multiarch support for dpkg long time ago 15:53 < codehelp> as a branch? 15:54 < guillem> there's a patch in the bts I think 15:57 < codehelp> I can see #324546 and #369507 - from Goswin - are those what you are thinking of? 16:00 < guillem> I've the patch on my home machine, probably took it from somewhere else then 16:03 < codehelp> ok - if you can find it, it would be handy if you could send to me - we'll take a look at it next week 16:03 < codehelp> (Emdebian WorkSession in Cambridge all next week so lots of available dev time) 16:04 < guillem> yeah sure, or even to one of those bug reports if it's not there, will check later 16:09 < zumbi_> btw, guillem think about extremadura worksession 16:17 < zumbi_> guillem: is #369064 what you are talking about? --- 23:21 < mrvn> zumbi_: no mail found 23:21 < zumbi_> i wrote a follow up to your bug report 23:21 < zumbi_> on binutils multiarch 23:22 < zumbi_> 369064 23:22 < mrvn> zumbi_: ahh, that would be in "bugs" then. 23:23 -!- ifvoid|0120804 is now known as ifvoid 23:24 < mrvn> zumbi_: that doesn't include /s390x-linux-gnu/lib nor /usr/s390x-linux-gnu/lib 23:25 < mrvn> zumbi_: and multiarch would be /lib/s390x-linux-gnu and /usr/lib/s390x-linux-gnu 23:25 < AzaTht> mrvn: didn't get any reply before 23:25 < mrvn> zumbi_: are you sure that patch works at all? 23:26 < zumbi_> mrvn: it used to do its job for what it was meant 23:26 < zumbi_> i suggest you to sent your patches to binutils upstream 23:27 -!- christoph [~christoph@f049032020.adsl.alicedsl.de] has quit [Quit: Verlassend] 23:27 < AzaTht> mrvn: any chance I could see your script? 23:27 < AzaTht> or is it #er-seecret prop-code 23:27 * zumbi_ would like to see multiarch running 23:28 < zumbi_> AzaTht: one think at a time... people is not multitask :-) 23:28 < zumbi_> s/think/thing 23:28 < AzaTht> zumbi_: prio queue 23:28 < AzaTht> ? 23:29 -!- streuner [~streuner@p54A5E32E.dip.t-dialin.net] has quit [Quit: Verlassend] 23:30 < mrvn> AzaTht: http://mrvn.homeip.net/filter-dctrl/ 23:30 < mrvn> AzaTht: go nuts 23:31 < mrvn> zumbi_: If we allow the namespace pollution of the cross-compile dirs for multiarch then I guess we can live with your patch too. 23:31 < mrvn> zumbi_: can you paste an ldscript example output? I'm not sure what dirs would get added overall. 23:32 < helix> are you using that as a smiley? 23:32 < ari> OGC 23:33 < zumbi_> mrvn: it adds /lib64/ dirs 23:33 < mrvn> zumbi_: but not lib? 23:33 < AzaTht> thanks 23:34 < zumbi_> yes, lib as weel 23:34 < zumbi_> s/weel/well 23:34 < mrvn> zumbi_: but not the /s390x-linux-gnu/lib[64] dirs, right? 23:35 < helix> hi ari 23:35 < zumbi_> those too, at least for cross compiling gcc it did well for all official debian arches 23:36 < AzaTht> helix: me? yes 23:36 < helix> ok 23:36 < zumbi_> anyway, do you want to rework all that and try a testcase for multiarch? 23:36 < zumbi_> i can do some work during the weekend and next week 23:37 < mrvn> zumbi_: One problem remains though. On i386 it uses /usr/i486-linux-gnu/lib and on amd64 (32bit) it uses /usr/i386-linux-gnu/lib. 23:37 < zumbi_> yes 23:37 < mrvn> zumbi_: That is the other patch in my bugreport. 23:37 < mrvn> How do you solve that? 23:37 < Manoj> heh. http://buildd.debian.org/fetch.cgi?pkg=setools;ver=3.3.4.ds-3.1;arch=alpha;stamp=1212150215 23:37 < zumbi_> ummm.. that is why amd64<->i386 toolchain fails 23:38 < mrvn> zumbi_: one could just link i386-linux-gnu -> i486-linux-gnu 23:38 < Manoj> if they had asked, I would have told them that a simplistic solution like that does not work 23:38 < Manoj> Hmm. it did work on mipsel, though ... 23:39 < zumbi_> mrvn: in my case i did not really care much about that arches, but it is something that has been hanging arround 23:39 * Manoj looks deeper 23:40 < Manoj> so. http://buildd.debian.org/build.cgi?pkg=setools 23:41 -!- streuner [~streuner@p54A5E32E.dip.t-dialin.net] has joined #debian-devel 23:41 -!- esaym [~user@cpe-70-120-89-6.satx.res.rr.com] has quit [Remote host closed the connection] 23:42 < Manoj> no patch in BTS. no mail to me. and a NMU that does not fix the bug, even though they closed it. 23:42 < Manoj> so, no test whether they actually fixed the bug or not. 23:42 < zumbi_> mrvn: now, i'm having some trouble with mips triarch and some other arches.. see http://www.emdebian.org/~zumbi/toolchain/unstable-toolchain-amd64/logs//amd64-mips-4.3.log (quitte big file) 23:43 < mrvn> zumbi_: do they even have different TOOL_LIB strings? 23:44 < mrvn> /usr/mips-linux-gnu/bin/ld: /usr/mips-linux-gnu/lib/crti.o: ABI is incompatible with that of the selected emulation 23:44 < mrvn> zumbi_: looks like gcc gets the wrong path. 23:44 < zumbi_> /usr/mips-linux-gnu/bin/ld: skipping incompatible /usr/mips-linux-gnu/lib/libc.so when searching for -lc 23:44 < mrvn> zumbi_: That is fine as long as it eventually finds one. 23:44 < zumbi_> it should look under /usr/mips-linux-gnu/bin/ld: skipping incompatible /usr/mips-linux-gnu/lib32/libc.so when searching for -lc 23:45 < zumbi_> rm last line 23:45 < zumbi_> it should look under /usr/mips-linux-gnu/lib32/libc.so 23:45 < zumbi_> but it doesn't 23:45 < mrvn> zumbi_: for n32? 23:45 < zumbi_> yes 23:46 < mrvn> isn#t lib32 for o32? 23:46 < zumbi_> no 23:46 -!- schasi [~schasi@p54A27651.dip.t-dialin.net] has joined #debian-devel 23:46 < zumbi_> are you sure 23:46 < zumbi_> i believe lib is for o32 and lib32 for n32 and lib64 for n64 23:46 < mrvn> zumbi_: it should look in something like /usr/mipsn32-linux-gnu/lib/ 23:47 < zumbi_> but that would be another new arch 23:47 < mrvn> For multiarch you need uniq TOOL_LIB strings for every abi. 23:47 < Manoj> what is the quick way to find who uploaded a package? 23:47 < mrvn> zumbi_: the lib32 and lib64 dirs are just ugly hacks. 23:48 < zumbi_> yes - that is exactly the problem i'm facing right now 23:48 < mrvn> Manoj: read the changes file 23:48 < Manoj> Maintainer: Manoj Srivastava 23:48 < Manoj> Changed-By: Arthur Loiret 23:48 < zumbi_> mrvn: that is the reason why i want multiarch 23:48 < mrvn> Manoj: and signed by? 23:48 < mrvn> zumbi_: for the mips arch someone has to extend the TOOL_LIB thing in binutils to make it uniq for each abi. 23:49 < zumbi_> what about powerpc 23:49 < Manoj> mrvn: there is none, so perhaps arthur.loiret@gmail.com is the one who uploaded it? 23:49 < mrvn> zumbi_: powerpc64-linux-gnu or ppc64-linux-gnu 23:49 < zumbi_> do you need to do that work as well for the 32 and 64 ABIs 23:49 < zumbi_> powerpc64-linux-gnu 23:49 < mrvn> s390 and s390x, sparc and sparc64, i486 and x86_64 23:50 * Manoj sends a notfixed and a found message to control 23:50 < MadCoder> Manoj: who-uploads 23:50 < mrvn> I think only mips doesn't have one arch per abi 23:50 -!- MoDaX [~nth@lan-84-240-22-131.vln.skynet.lt] has quit [Remote host closed the connection] 23:51 < mrvn> zumbi_: anyway, the "when searching for -lc" just means it found the bad one first. If it continues that means it eventually found a good one. 23:51 < zumbi_> mrvn: so that needs to be implemented on binutils? 23:51 < mrvn> zumbi_: yes. 23:51 < Manoj> MadCoder: Hmm. who-uploads is taking a while. 23:51 < zumbi_> no patches arround? 23:51 < mrvn> zumbi_: never looked at it. 23:51 -!- sabdfl [~sabdfl@87-194-36-33.bethere.co.uk] has joined #debian-devel 23:51 < Manoj> heh. maks. Why am I not surprised. 23:52 < mrvn> zumbi_: isn't mips n64 mips64-linux-gnu? 23:52 -!- zack [~zack@did75-20-88-183-33-56.fbx.proxad.net] has quit [Quit: Leaving.] 23:52 < MadCoder> Manoj: of course it does, it grabs things on d-d-c archives IIRC or sth similar 23:52 < zumbi_> mrvn: yes, and i guess n32 is mipsn32 23:52 < mrvn> zumbi_: then you just have to get that corrected in genscript.sh 23:53 < mrvn> zumbi_: That might be the same problem that causes amd64 to have i386-linux-gnu. There is no way to pass the right arch into the build and it has some strings hacked in for multilib. 23:54 < mrvn> zumbi_: as a test you could edit the ldscripts. I think for not o32 abi it should still parse the files. 23:56 -!- zack [~zack@did75-20-88-183-33-56.fbx.proxad.net] has joined #debian-devel 23:56 -!- MoDaX [~nth@lan-84-240-22-131.vln.skynet.lt] has joined #debian-devel 23:56 < zumbi_> mrvn: what do you mean? 23:57 < mrvn> zumbi_: edit /usr/lib/ldscripts/elf_mipsn32.* 23:57 < zumbi_> let's go step-by-step 23:57 < mrvn> or maybe first check with strace -f if it opens a file in /usr/lib/ldscripts 23:57 < zumbi_> where should i start to have multiarch working? on binutils? 23:57 < mrvn> zumbi_: yes 23:59 < zumbi_> add /lib/ and /usr/lib/< 23:59 < zumbi_> by arch_abi i mean arches called by "ABI names" Day changed to 31 mai 2008 00:00 < mrvn> zumbi_: first fix it so it uses the right arch as in mipsn32-linux-gnu 00:00 < zumbi_> ok - i have the help of your patches 00:01 -!- mind [~mind@host-84-222-13-183.cust-adsl.tiscali.it] has joined #debian-devel 00:01 < zumbi_> then, glibc and gcc already include multiarch, so we have a toolchain 00:01 < zumbi_> then we start to work with dpkg 00:01 < zumbi_> sounds right? 00:01 < mrvn> zumbi_: gcc only needs the support for /usr/include/arch-os-libc/ and it already has that. Check that it works for mips, mipsn32 and mipsn64. 00:02 < mrvn> Then you need to pick one library package and convert it to use the new dirs. 00:04 < mrvn> zumbi_: Then for dpkg you need the following: 1) allow different archs, 2) extend the db to allow multiple packages with the same name but different arch, 3) figure something out for /usr/share/doc/package, 4) fix the depends resolver to use the arch and Multi-Arch: yes|no field. 00:05 -!- d0rt [~ni@69.37.102.188] has joined #debian-devel 00:05 < mrvn> 5) add the arch to /usr/lib/dpkg/info/package* 00:06 < zumbi_> guillem and some other people can help on that 00:06 < zumbi_> maybe you are interested as weel 00:06 < mrvn> Not this week. But in general yes. 00:07 < zumbi_> ok, we can not do that in one day, i guess :-) 00:08 < mrvn> zumbi_: 5 will need some discussion with dpkg. It should be compatible with what we have now and reversible. 00:09 < mrvn> 2 and 4 are the really hard parts. ---