[00:01] stgraber: so is that latter sysvinit change cleaning up after previous incomplete sysvinit migration code? [00:01] (I suppose that's why it might be in sysvinit instead of lxc?) [00:02] stgraber: also, can you please put a version guard on that upgrade code? [00:54] slangasek: yeah, and lxc doesn't have any package inside the containers, so we don't have a way of fixing them other than SRUing initscripts for that. In theory the check I uploaded shouldn't even match a setup that's not broken but I should have indeed added a version guard on top of that if someome actually wants the "broken" setup. [01:00] stgraber: the risk of a false positive is small, but I'm paranoid about upgrade code not having version checks; so please fix that in quantal and in the SRU :) === apachelogger___ is now known as apachelogger [02:01] slangasek: please reject the SRU then, I'll prepare an updated postinst [02:23] stgraber: rejected, thanks [02:25] slangasek: re-uploaded to -proposed, testing the one to quantal to make sure I didn't mess up anything in the version check === mthaddon` is now known as mthaddon [03:36] so I'm trying to figure out how to bootstrap an x32 system by practicing building i386 packages from amd64... if I run debuild -ai386, it seems to build an i386 binary package ( though dh_strip tries to run i686-linux-strip instead of strip, so I had to make a symlink ), but I can't figure out how to get debuild to pass gcc -march=i386 [03:38] psusi: so debuild -ai386 will set DEB_HOST_ARCH and friends; packages that know about cross-building will take this as a hint to *do* cross-building, which means they'll look for a cross-compiler [03:38] psusi: so they will look for i686-linux-gnu-gcc - provide this and have it do the right thing [03:38] slangasek, right... I guess that's why dh_strip was trying to run i686-linux-strip instead of just strip.... [03:39] but the actual build didn't complain about not being able to find i686-linux-gnu-gcc [03:39] it seemed to run gcc and build just fine... just barfed at dh_strip, which I got around with a symlink [03:39] ideally i686-linux-gnu-gcc could be implemented by gcc itself just knowing what to do; but at a push, you can make it a short wrapper script that calls exec gcc -m32 $@ [03:39] right, then your test package seems to not be cross-build-friendly [03:39] (which package?) [03:40] but... while it built an i386 binary deb, the actual binaries are still amd64 since it isn't telling gcc the right -m [03:40] slangasek, yes i tested using multi-arch for that, it works great! [03:40] depending on the package, it may be using gcc as a fallback [03:40] psusi: Yeah, it's obviously hardcoding the compiler, which is not cross-friendly. [03:40] hrm.... that could be [03:40] and maybe creating i686-linux-gnu-gcc will be enough to make it happy [03:40] just using a -m32/-m64 wrapper for -afoo that can be satified by multilib [03:40] again, which package? [03:40] Or, I suppose it could be cross-friendly but falling back in the absense of triplet-cc, yeah. [03:41] slangasek, you need symlinks for binutils, and gcc/cpp/etc, and then it works [03:41] yes [03:41] so... you are saying to get it to work, I have to write a wrapper gcc script to run gcc -m32? can't get dh to be happy just running gcc and adding -m32 to the CFLAGS? [03:41] but yeah i had that working great [03:41] paulliu, yep, i'll go find what i was using [03:42] paulliu, slacker_nl this works: http://paste.ubuntu.com/1102847/ [03:43] i called it "gcc-32", and you could easily make a similar gcc-64 [03:43] it seems kind of silly to have to invoke all of the cross compiling machinery when the native gcc supports the "cross" arch, you just need to add a flag [03:43] but you're saying that's just how you have to do it? [03:43] psusi, the "c99" "c89" commands work exactly the same [03:43] of course this is called by the full qualified name [03:44] i got blocked on trying to wrap my head around how the packaging for gcc-defaults-cross might work [03:45] i think it should be packaged like this IMHO [03:45] so if you have to pretend you are doing a full cross compile... why doesn't our gcc package come with the cross compiler wrapper script to add -m32? [03:46] psusi, it could either be part of (gcc|g++|gobjc)-multilib packages, or a seperate package [03:48] psusi: You could certainly file a bug to that effect. [03:58] how do you smack bzr bd into passing options to debuild? it pukes on -ai386 [03:58] -- perhaps? [03:58] (I don't use it, just a wild guess) [03:59] psusi: -- $OPTIONS [03:59] good guess... I was thinking that but the man page doesn't show it ;) [03:59] Yes. I've written bugs about it. [03:59] It's not a guess. I knew because I couldn't find it either and had to ask. [04:00] "inconsitent interface" is on my reasons to avoid UDD kinds of things. [04:00] Who'd have thought -S -- -sa -us -uc was a good plan? [04:02] hrm... when I add -ai386, it says /usr/include/linux/errno.h can't find asm/errno.h [04:04] psusi, install libc6-dev:i386 ? [04:04] ohh... I guess gcc expects a separate 32bit headers directory [04:05] psusi, or rather apt-get build-dep -ai386 [04:12] ScottK: it's equally valid to call it as -- -S -sa -us -uc, if that makes you happier ;P [04:13] hrm.. so why does apt-get want to remove the native amd64 -dev packages to install the i386 ones?