From: <ast...@el...> - 2007-03-12 23:51:33
|
Got a bit of a surprise shock when I decided to go a little bleeding edge with a trunk compile. Seems bash has a minor issue, though I hadn't seen it before when I either was able to complete a compile, or at least get much farther along. This is trunk 757 with gcc 4.2 snapshot 20070307 , compile with ObjC, shared libraries, and libtool. This is on an older svn checkout where I had wiped out toolchain, toolchain_buildi586, and build_i586 and did a svn update to rebuild those parts cleanly while retaining the dl directory. make[2]: Leaving directory `/home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/bash-3.0/po' make[1]: Leaving directory `/home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/bash-3.0' rm -f /home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/root/bin/bash* mv /home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/root/usr/bin/bash* /home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/root/bin/ (cd /home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/root/bin; /bin/ln -fs bash sh) rm -rf /home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/root/share/locale /home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/root/usr/info \ /home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/root/usr/man /home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/root/usr/share/doc /home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/staging_dir/bin/i586-linux-uclibc-strip --remove-section=.comment --remove-section=.note /home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/root/usr/bin/bash /home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/staging_dir/bin/i586-linux-uclibc-strip: '/home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/root/usr/bin/bash': No such file make: *** [/home/ouroboros/astlinuxbuildenv/astlinux/trunk/build_i586/root/bin/bash] Error 1 |
From: Kristian K. <kri...@gm...> - 2007-03-15 05:42:13
|
On 3/12/07, ast...@el... <ast...@el...> wrote: > Got a bit of a surprise shock when I decided to go a little bleeding edge > with a trunk compile. Seems bash has a minor issue, though I hadn't seen > it before when I either was able to complete a compile, or at least get > much farther along. > > This is trunk 757 with gcc 4.2 snapshot 20070307 , compile with ObjC, > shared libraries, and libtool. This is on an older svn checkout where I > had wiped out toolchain, toolchain_buildi586, and build_i586 and did a svn > update to rebuild those parts cleanly while retaining the dl directory. > asterisk, Wow, that is bleeding edge! Anyways, I can't really make much sense out of that error. Can you turn the number of parallel builds down and run make after script? -- Kristian Kielhofner |
From: <ast...@el...> - 2007-03-16 09:08:36
|
On 3/15/07, Kristian Kielhofner <kri...@gm...> wrote: > On 3/12/07, ast...@el... <ast...@el...> wrote: > > Got a bit of a surprise shock when I decided to go a little bleeding edge > > with a trunk compile. Seems bash has a minor issue, though I hadn't seen > > it before when I either was able to complete a compile, or at least get > > much farther along. > > > > This is trunk 757 with gcc 4.2 snapshot 20070307 , compile with ObjC, > > shared libraries, and libtool. This is on an older svn checkout where I > > had wiped out toolchain, toolchain_buildi586, and build_i586 and did a svn > > update to rebuild those parts cleanly while retaining the dl directory. > > > > asterisk, > > Wow, that is bleeding edge! > > Anyways, I can't really make much sense out of that error. Can you > turn the number of parallel builds down and run make after script? > > -- > Kristian Kielhofner But the concurrent jobs setting was at the default 2... I've reinstalled a fair amount of stuff, and ran with only 1 concurrent job and that seemed to make it go away, I think. Using gcc 4.2 20070307, binutils 2.16.91.0.7 , and a hand fix in the makefile for syslinux so it doesn't call the windows stuff (changing the $(BOBJECTS) to $(BTARGET), I got a compile to finish properly. Too bad I don't have an easy way to test the image itself. I'm really tempted to make a proper VMware development environment so I can say definitively that my environment isn't at fault. Maybe stick that into the SVN repository as well? |
From: Kristian K. <kri...@gm...> - 2007-03-16 14:37:04
|
On 3/16/07, ast...@el... <ast...@el...> wrote: > > But the concurrent jobs setting was at the default 2... The only reason I asked was because the more concurrent jobs, the harder it is to read errors from make, gcc, etc, as they are printed on the screen. It doesn't have anything to do with the actual number of jobs. It just makes it easier to read. > I've reinstalled a fair amount of stuff, and ran with only 1 > concurrent job and that seemed to make it go away, I think. That could very well be the case. Can you find out for sure? > Using gcc 4.2 20070307, binutils 2.16.91.0.7 , and a hand fix in the > makefile for syslinux so it doesn't call the windows stuff (changing > the $(BOBJECTS) to $(BTARGET), I got a compile to finish properly. Hmmm... That hand fix is not required with the standard configuration. Do you know why you needed to make this change? > Too bad I don't have an easy way to test the image itself. "make iso" - load astlinux.iso into VmWare. It will boot and run as a live cd. > I'm really tempted to make a proper VMware development environment so > I can say definitively that my environment isn't at fault. Maybe stick > that into the SVN repository as well? While many people use VMWare, that really defeats the purpose of buildroot. trunk now includes a basic dependency check. I plan on extending that to check the host system for all of the various required software components, versions, etc. That should drastically cut down on some of the problems people are having. I should also point out that a lot of your problems come from using non-standard options. Your GCC snapshot and bleeding-edge binutils, for instance. Quite frankly, there is no way that we could possibly support all of the various options and packages in buildroot. Just because an option for a given version of binutils (for instance), is available it doesn't mean that it has been tested with all of the various packages and other options. As Darrick pointed out a while ago, we only have the resources to support the default config. Most of the time you can change a few options, packages, etc without causing too many problems. How would we stick a VmWare image into SVN? -- Kristian Kielhofner |
From: Darrick H. <dha...@dj...> - 2007-03-16 15:02:09
|
Kristian Kielhofner wrote: >> I'm really tempted to make a proper VMware development environment so >> I can say definitively that my environment isn't at fault. Maybe stick >> that into the SVN repository as well? >> > > While many people use VMWare, that really defeats the purpose of > buildroot. trunk now includes a basic dependency check. I plan on > extending that to check the host system for all of the various > required software components, versions, etc. That should drastically > cut down on some of the problems people are having. > There are several reasons to use VMWare. The best reason is to prevent accidental mishaps. This will go away with some of the changes in trunk, but currently, the makeimage.sh script must be run as root. I know I've accidentially stayed as root and done some actions which could potentially cause problems to the host system. Again, most of that risk will go away with trunk shortly. Kristian has started using a method of making the images that doesn't require the user to be root. (I believe this involves fakeroot). The other reason for using vmware is to have a consistent build environment. Most, if not all, of the development work has been done on CentOS4. If someone doesn't have a spare PC around, they could setup a VMware image with CentOS in it (just for consistency). However, it shouldn't matter which version of Linux you compile this on, as long as the basic system requirements are met. > How would we stick a VmWare image into SVN? > That would be absurd. The host system requirements don't change often enough to require revision control. (let along the size of the creature). I don't think we need to distribute a virtual machine for the build environment. It would be good to have the build system requirements documented better. Perhaps we need a wiki other than voip-info--the information gets too muddled there. Darrick -- Darrick Hartman DJH Solutions, LLC http://www.djhsolutions.com |