[Buildroot] Analysis of build failures

Mon Nov 11 10:01:17 UTC 2013

On Sat, Nov 9, 2013 at 6:15 PM, Thomas Petazzoni
<thomas.petazzoni at free-electrons.com> wrote:
> Hello all,
>
> If you're Cc'ed on this e-mail, it means that there is a question for
> you below. Here is a summary of who should look at what problems (note
> that some problems are not affected to anyone, either because the
> reason is unknown or because there is no clear "owner" for the problem):
>
> Markos Chandras                 aespipe, fdk-aac, gmp, libiscsi
> Sonic Zhang, Aaron Wu           boost, icu, openpgm
> Chris Zankel                    cdrkit
> Peter Korsgaard                 gst1-plugins-bad, pulseaudio
> Gustavo Zacarias                protobuf-c, mplayer, nettle (ARM), php
> Simon Dawson                    libcap-ng, util-linux, zeromq, zmqpp
> Spenser Gilliland               libglib2, nettle (Microblaze), openssl
> Ezequiel Garcia                 libnspr
> Mischa Jonker                   pulseaudio, ti-utils
> Fatih Aşıcı                     qt5base
> Tzu-Jung Lee                    tstools
>
> Thanks!
>
> Thomas
>
> On Sat,  9 Nov 2013 08:30:28 +0100 (CET), Thomas Petazzoni wrote:
>
>> Detail of failures
>> ===================
>>
>>       mips |                   aespipe-2.4c | NOK | http://autobuild.buildroot.net/results/da0268df99b9bf920d88121e7c3d33d41a57d951/
>
> Unknown problem (error: C compiler cannot create executables). Markos,
> as our MIPS guy, can you have a look at this one?
>
>>        arm |                alsa-lib-1.0.26 | NOK | http://autobuild.buildroot.net/results/983828bd27a521020e42e58cce9393e6b8d23502/
>
> It was a wrong toolchain configuration, with a weird toolchain.
> I've removed the toolchain from the autobuilders.
>
>>    powerpc |              alsa-utils-1.0.26 | NOK | http://autobuild.buildroot.net/results/1c6a33e75fe0617e01d7413cdfd5fbd466f822d4/
>
> Missing link against -lrt for static linking.
>
>>        arm |              alsa-utils-1.0.26 | NOK | http://autobuild.buildroot.net/results/6123186a66bbea833ab2fa980346f16825b22d21/
>
> checking for snd_ctl_open in -lasound... no
> configure: error: No linkable libasound was found.
> make: *** [/home/test/test/3/output/build/alsa-utils-1.0.26/.stamp_configured] Error 1
>
> Weird, don't know what's happening here.
>
>>        arm |                audiofile-0.3.6 | NOK | http://autobuild.buildroot.net/results/6aff0a6b9c7b2890cc069868b041ed74e549933b/
>
> This was my funky toolchain (same as alsa-lib problem above). Toolchain
> removed from the autobuilders.
>
>>     xtensa |                  binutils-2.22 | NOK | http://autobuild.buildroot.net/results/11de6f161585c12148c1b2b4fc8ab830183edfe9/
>>     xtensa |                  binutils-2.22 | NOK | http://autobuild.buildroot.net/results/2d3d255b2d3ae1456a377dbc03f695b7adf010e8/
>
> This is fixed by http://patchwork.ozlabs.org/patch/289532/, which
> hasn't been committed yet. It will also require a rebuild of the
> toolchain in the autobuilders.
>
> Peter, can you commit this patch?
>
>>       bfin |                   boost-1.54.0 | NOK | http://autobuild.buildroot.net/results/84f9d9d2b6b5356984c6d59aad5e28b3e7847651/
>
> Don't know what's going on. Sonic, Aaron, can you have a look?
>
>>        arm |                 busybox-1.21.1 | NOK | http://autobuild.buildroot.net/results/116d13f7489981c39c3759eaf8302be4d795f339/
>
> This was my weird toolchain (same problem as alsa-lib). Toolchain
> removed from the autobuilders.
>
>>     xtensa |                  cdrkit-1.1.11 | NOK | http://autobuild.buildroot.net/results/0315c117986f3aac8c0719d744cb47ee03738ee9/
>
> [ 58%] Building C object libhfs_iso/CMakeFiles/hfs_iso.dir/record.o
> ../libusal/libusal.a(scsi-remote.o):(.literal+0xe0): undefined reference to `valloc'
>
> Chris, can you have a look at this one, since it's apparently a
> Xtensa-specific problem?
>
>>     mipsel |                  fdk-aac-0.1.2 | NOK | http://autobuild.buildroot.net/results/cb87635832f4a1128f174a25c264f716399c52d4/
>
> MIPS specific problem. We're building for mips1, but fdk-aac uses some
> instructions not available on mips1. Maybe we should only enable
> fdk-aac for some more capable variant of MIPS instruction sets.
>
> Markos, can you have a look at this one?
>
>>       bfin |               fontconfig-2.6.0 | NOK | http://autobuild.buildroot.net/results/8579cb481e86dcae0988379c5781bd00f4e70809/
>
> fontconfig forgets to link against libpng. Anyone willing to fix this?
>
>>       mips |                      gmp-5.1.3 | NOK | http://autobuild.buildroot.net/results/69b0804cbfbb24c70f90435a37fb56a118247a57/
>
> configure: error: could not find a working compiler, see config.log for details
>
> Markos, can you have a look?
>
>>        arm |         gst1-plugins-bad-1.2.0 | NOK | http://autobuild.buildroot.net/results/b99eb6a9067cee885da88bff64ee447c37e31e0c/
>>        arm |         gst1-plugins-bad-1.2.0 | NOK | http://autobuild.buildroot.net/results/b320bb61650c58e9c855e3136f5f6d7bf5e4ef56/
>
> Same problem in both cases, related to the DirectFB plugin. Peter, you
> are the one taking care of Gstreamer most of the time, can you have a
> look into this?
>
>>     x86_64 |           host-protobuf-c-0.15 | NOK | http://autobuild.buildroot.net/results/5d163e750808f2fdd8098cb58240bada12b770a9/
>>   mips64el |           host-protobuf-c-0.15 | NOK | http://autobuild.buildroot.net/results/038340dfbbbedbcfcb98ad5dbf828bc03fe26af6/
>
> Weird compilation problem, seems really caused by an issue in
> protobuf-c itself. Gustavo, you're the one who added protobuf-c, can
> you have a look at this?
>
>>       bfin |                       icu-51.2 | NOK | http://autobuild.buildroot.net/results/e0a1130feb1784b3d4fefe1224943d93409a3494/
>>       bfin |                       icu-51.2 | NOK | http://autobuild.buildroot.net/results/85c082a693ef795cecf0ef9204b7c9c0850d4b74/
>
> Problem building icu on Blackfin:
>
> *** ERROR - configure could not detect your platform
>
> Sonic, Zhang, can you have a look into this?
>
>>      avr32 |                libcap-ng-0.7.3 | NOK | http://autobuild.buildroot.net/results/3df16d60f6bbb679129cd568df4440be6f35cd9f/
>
> The missing TLS problem on AVR32. We still haven't found the solution
> to handle this one. As I've suggested previously, I would simply make
> TLS support unconditional in Buildroot as soon as thread support is
> enabled, and then mark libcap-ng as not available on AVR32. Of course,
> if you have an external toolchain without TLS support, you're on your
> own.
>
>> microblaze |                libglib2-2.36.3 | NOK | http://autobuild.buildroot.net/results/6f0066f56f10d3f17023e698fd3ba1bd2d00c4c1/
>> microblaze |                libglib2-2.36.3 | NOK | http://autobuild.buildroot.net/results/79d2df836063167f12f5bba177297a041bb7c16f/
>
> Missing compiler intrinsics and fallocate in the Microblaze toolchain.
> We're waiting for Spenser Gilliland to finish the Microblaze internal
> toolchain support, so that we can ditch the crappy Microblaze external
> toolchains.
>
>>   mips64el |                 libiscsi-1.6.0 | NOK | http://autobuild.buildroot.net/results/39e8e0ebbeb196276492b928b41cc0b0fe1e9ec3/
>>   mips64el |                 libiscsi-1.6.0 | NOK | http://autobuild.buildroot.net/results/efb87b17c49170f6377631371438911b18029cea/
>
> The usual problem of ld being used instead of gcc, causing problem on
> mips64el. However, you can't do partial linking with gcc, so a libtool
> fix is needed.
>
> Markos, I know you've identified the libtool fix for this. Could you
> backport it on top of libtool 2.4.2 (that we have in Buildroot), add it
> as a patch in package/libtool/, and then mark LIBISCSI_AUTORECONF =
> YES so that the libtool stuff gets regenerated with a known-working
> version?
>
>>      nios2 |                  libnspr-4.9.6 | NOK | http://autobuild.buildroot.net/results/8c01fdaac7afa00bd1117c79ae047344242d1463/
>>      nios2 |                  libnspr-4.9.6 | NOK | http://autobuild.buildroot.net/results/a418145d72e92f64956518bb2786e33dadd08a86/
>
> NIOS2 support missing in libnspr. Ezequiel, can you look into this?
> Normally, adding an architecture support in libnspr is really easy, so
> adding a patch to libnspr should be feasible.
>
>>    aarch64 |                   libv4l-0.8.9 | NOK | http://autobuild.buildroot.net/results/4de7dc58e826d1dd60add2821f774c9c5f5a2a71/
>
> Package needs to be bumped, I have already started working on this.
>
>>        arm |                  mplayer-1.1.1 | NOK | http://autobuild.buildroot.net/results/8f946e28b11b2ab1ba0d9688286c665488de0486/
>
> libmpeg2/motion_comp_arm_s.S: Assembler messages:
> libmpeg2/motion_comp_arm_s.S:29: Error: selected processor does not support ARM mode `pld [r1]'
> libmpeg2/motion_comp_arm_s.S:39: Error: selected processor does not support ARM mode `pld [r1]'
>
> That's an ARMv4t configuration. Maybe building mplayer on ARMv4t is not
> possible. Adding a bunch of depends on !BR2_arm<something> is probably
> the solution here. Gustavo, maybe?
>
>>        arm | ne10-88c18f02199947b2c8b577... | NOK | http://autobuild.buildroot.net/results/3631f7e9dad9fab356d31160ac95b7d0fec70ce2/
>
> Needs NEON support apparently. Should be something for me.
>
>> microblaze |                   nettle-2.7.1 | NOK | http://autobuild.buildroot.net/results/82acbea862ee0bb5288ab498d86772bc475d3a53/
>>        arm |                   nettle-2.7.1 | NOK | http://autobuild.buildroot.net/results/88ea0cc0815f865e1a144d39231fdc9b47e689b0/
>
> On Microblaze, a compiler error. Should be fixed by the internal
> toolchain support for Microblaze. Waiting for Spenser on this :)
>
> On ARM, the failure is due to nettle tests requiring C++ support. It
> would be good to disable the compilation of those tests. Gustavo,
> you're the one looking after nettle in general, can you look into this?
>
>>       bfin |           openpgm-5.1.118~dfsg | NOK | http://autobuild.buildroot.net/results/120b7de960a04def5c53b59644eefc598bd0a1d4/
>
> Blackfin does not have getifaddrs():
>
> getifaddrs.c:853:3: error: #error "Unsupported interface enumeration on this platform."
>
> Not sure how to handle this uClibc configuration difference, except by
> adding "depends on !" to exclude the specific Blackfin toolchains
> causing problems.
>
>> microblaze |                 openssl-1.0.1e | NOK | http://autobuild.buildroot.net/results/f313ac3233a8ed5b3087936c1cd8408a00fdf2bb/
>
> Compiler error. Spenser, we need your patches!
>
>>    aarch64 |                     php-5.3.27 | NOK | http://autobuild.buildroot.net/results/08d3255303f7ceda8a1ecfcf56ad665f00829632/
>
> /usr/include/bits/predefs.h:20:3: error: #error "Never use <bits/predefs.h> directly; include <features.h> instead."
>  # error "Never use <bits/predefs.h> directly; include <features.h> instead."
>
> Gustavo, you like PHP, no ? :-)
>
>>        arc |                 pulseaudio-4.0 | NOK | http://autobuild.buildroot.net/results/281ff1c8046b7e93759c5f51995f34788e717fad/
>
> checking for atomic_ops.h... no
> configure: error: *** libatomic-ops headers not found
> make: *** [/home/test/test/1/output/build/pulseaudio-4.0/.stamp_configured] Error 1
>
> For some reason, on ARC, atomic_ops is needed. Why not on other
> architectures? Peter, Mischa, any idea?
>
>>    powerpc |                       qt-4.8.5 | NOK | http://autobuild.buildroot.net/results/0ee61f71e61f2215a15e4655a4e22d0521334c35/
>
> Will be fixed by my patch that prevents building parts of Qt on some
> unsupported architectures.
>
>>       i686 |                  qt5base-5.1.1 | NOK | http://autobuild.buildroot.net/results/bbe7d79da059fdf67fb29880b8539ca2320ba86d/
>>        arm |                  qt5base-5.1.1 | NOK | http://autobuild.buildroot.net/results/37613281f21895716ceaf64654e175148d7f543c/
>>       i686 |                  qt5base-5.1.1 | NOK | http://autobuild.buildroot.net/results/555278227680e90737e6dbe0a6dcc035507150aa/
>
> For
> http://autobuild.buildroot.net/results/37613281f21895716ceaf64654e175148d7f543c/build-end.log,
> Samuel has already sent a patch
> (http://patchwork.ozlabs.org/patch/289999/). The other two ones:
>
>   http://autobuild.buildroot.net/results/555278227680e90737e6dbe0a6dcc035507150aa/build-end.log
>   (sqlite problem)
>
>   http://autobuild.buildroot.net/results/bbe7d79da059fdf67fb29880b8539ca2320ba86d/build-end.log
>   (inlining problem)
>
> are not fixed. Fatih, would you mind having a look?
>
>>        arc | ti-utils-06dbdb2727354b5f3a... | NOK | http://autobuild.buildroot.net/results/63d294574b053aa4e2148fbc3185c5594d925ae9/
>
> {standard input}: Assembler messages:
> {standard input}:1680: Error: missing operand
> {standard input}:1680: Error: junk at end of line: `@plt_tx_bip.part.8'
>
> ARC toolchain problem, for Mischa.
>
>>     x86_64 |                   tstools-1_11 | NOK | http://autobuild.buildroot.net/results/ce02d7fb9ae3aa06eaf53db39cbef27c051b525d/
>
> Should use gcc for linking and not ld + missing -lm. Tzu-Jung Lee,
> you're the one who added tstools, can you have a look at this problem?
>
>>      avr32 |              util-linux-2.23.2 | NOK | http://autobuild.buildroot.net/results/2ba1bd78f4303ef2f543a4e4fa65d1b68ba3a765/
>
> Missing fallocate on avr32. Simon, what is the proposal to solve this?
> Add fallocate support to 0.9.31 ?
>
>>        arm |                 valgrind-3.8.1 | NOK | http://autobuild.buildroot.net/results/f174541ac06aeeac82b8399686a8d4371532760d/
>
> /tmp/cceJNtCh.s: Assembler messages:
> /tmp/cceJNtCh.s:200: Error: r15 not allowed here -- `str r15,[r3,#+0]'
> make[4]: *** [libcoregrind_arm_linux_a-m_libcassert.o] Error 1
> make[4]: *** Waiting for unfinished jobs....
>
> Not sure what's going on here. Anyone to have a look?

valgrind uses inline assembly that is not Thumb compatible. It seems
it isn't expected to build it in Thumb mode (although it emulates
Thumb mode). It might be quite simple to fix but would need
coordination with upstream etc.