[Buildroot] [autobuild.buildroot.net] Your build results for 2017-02-15

[Buildroot] [autobuild.buildroot.net] Your build results for 2017-02-15

[Thomas Petazzoni]

Hello,

Adding the Buildroot mailing list in Cc, as well as Waldemar (uClibc-ng
maintainer).

On Thu, 16 Feb 2017 09:44:39 +0200, Olivier Schonken wrote:
> Hi Thomas
> 
> This looks like it might be a uclibc toolchain issue. Maybe the sequence of
> including/linking? Looks like a function declaration and definition mixup...
> 
> /home/buildroot/buildroot-test/instance-0/output/host/opt/ext-toolchain/bin/../lib/gcc/arm-buildroot-linux-uclibcgnueabi/5.4.0/libgcc.a(libunwind.o):
> In function `_Unwind_Resume':
> /opt/toolchain-build/build/host-gcc-final-5.4.0/build/arm-buildroot-linux-uclibcgnueabi/libgcc/../../../libgcc/config/arm/libunwind.S:357:
> multiple definition of `_Unwind_Resume'
> /home/buildroot/buildroot-test/instance-0/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/lib/libc.a(pt-arm-unwind-resume.os):pt-arm-unwind-resume.c:(.text+0x0):
> first defined here
> /home/buildroot/buildroot-test/instance-0/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/lib/libc.a(pt-arm-unwind-resume.os):
> In function `_Unwind_Resume':
> pt-arm-unwind-resume.c:(.text+0x2c): undefined reference to
> `pthread_cancel_init'
> pt-arm-unwind-resume.c:(.text+0x3c): undefined reference to `__libgcc_s_resume'
> collect2: error: ld returned 1 exit status
> make[4]: *** [pdftocairo] Error 1
> make[3]: *** [all-recursive] Error 1
> make[2]: *** [all] Error 2
> make[1]: *** [/home/buildroot/buildroot-test/instance-0/output/build/poppler-0.51.0/.stamp_built]
> Error 2
> make: *** [_all] Error 2
> make: Leaving directory `/home/buildroot/buildroot-test/instance-0/buildroot'

The exact same build error occurs for czmq:

  http://autobuild.buildroot.net/results/517/5179aade1ddfe3975b17f49b534e84ba1a55bdab/build-end.log

/opt/toolchain-build/build/host-gcc-final-5.4.0/build/arm-buildroot-linux-uclibcgnueabi/libgcc/../../../libgcc/config/arm/libunwind.S:357: multiple definition of `_Unwind_Resume'
/home/test/autobuild/run/instance-3/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/lib/libc.a(pt-arm-unwind-resume.os):pt-arm-unwind-resume.c:(.text+0x0): first defined here
/home/test/autobuild/run/instance-3/output/host/usr/arm-buildroot-linux-uclibcgnueabi/sysroot/usr/lib/libc.a(pt-arm-unwind-resume.os): In function `_Unwind_Resume':
pt-arm-unwind-resume.c:(.text+0x2c): undefined reference to `pthread_cancel_init'
pt-arm-unwind-resume.c:(.text+0x3c): undefined reference to `__libgcc_s_resume'

We have the same issue for glibmm:

  http://autobuild.buildroot.net/results/fb1/fb14e37aaf453874d1c33e5ed73b9c751ace5ae3/build-end.log

For enchant:

  http://autobuild.buildroot.net/results/70d/70d72b5e767965726a1e5eca437f3efc4af92c86/build-end.log

so that would indeed suggest a toolchain issue rather than a package
issue. Waldemar, do you understand what's going on here?

Looking at the glibmm failures:

  http://autobuild.buildroot.net/?reason=glibmm-2.50.0

it happens only with the fully static ARM toolchain, and since December
25, 2016. Enchant started failing on December, 10, and czmq on December
6. Looking more precisely at the build failures, we can see  that:

 * The earlier build failures from December/January all take place with
   an internal toolchain configuration, for which BR2_STATIC_LIBS was
   added.

 * Some of the newer build failures are using a pre-built ARM
   static-only external toolchain. It's only since my rebuild a few
   days ago that we started seeing those.

Bottom line: a toolchain change around early December started to cause
those build failures, and we are now seeing more of them because of the
recent rebuild of the Buildroot toolchains.

With all those informations at end, I would suspect the following
commit to be the culprit:

commit a44d7f2dbb19a54a8f2beb5c0f0adb0680999319
Author: Waldemar Brodkorb <wbx@openadk.org>
Date:   Sun Dec 4 12:20:27 2016 +0100

    uclibc: update to 1.0.20
    
    The test suite is removed from the package and is already a separate
    package in buildroot. All patches are upstream, so remove them.
    The UCLIBC_HAS_LFS option is removed upstream.
    
    Signed-off-by: Waldemar Brodkorb <wbx@openadk.org>
    [Thomas: remove BR2_UCLIBC_INSTALL_TEST_SUITE option, add it to
    Config.in.legacy.]
    Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

Thanks,

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

[Buildroot] [autobuild.buildroot.net] Your build results for 2017-02-15

[Thomas Petazzoni]

Hello,

On Thu, 16 Feb 2017 10:21:54 +0100, Thomas Petazzoni wrote:

> commit a44d7f2dbb19a54a8f2beb5c0f0adb0680999319
> Author: Waldemar Brodkorb <wbx@openadk.org>
> Date:   Sun Dec 4 12:20:27 2016 +0100
> 
>     uclibc: update to 1.0.20
>     
>     The test suite is removed from the package and is already a separate
>     package in buildroot. All patches are upstream, so remove them.
>     The UCLIBC_HAS_LFS option is removed upstream.
>     
>     Signed-off-by: Waldemar Brodkorb <wbx@openadk.org>
>     [Thomas: remove BR2_UCLIBC_INSTALL_TEST_SUITE option, add it to
>     Config.in.legacy.]
>     Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>

Seems like the story is more complicated. I did a build of czmq right
before the uClibc bump to 1.0.20, and right after, and both builds were
successful.

Two options:

 - This commit is not the culprit.

 - My simple configuration with just czmq doesn't trigger the problem,
   and it's only czmq with some optional dependencies that triggers the
   problem.

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

[Buildroot] [autobuild.buildroot.net] Your build results for 2017-02-15

[Waldemar Brodkorb]

Hi Thomas,
Thomas Petazzoni wrote,

> Hello,
> 
> On Thu, 16 Feb 2017 10:21:54 +0100, Thomas Petazzoni wrote:
> 
> > commit a44d7f2dbb19a54a8f2beb5c0f0adb0680999319
> > Author: Waldemar Brodkorb <wbx@openadk.org>
> > Date:   Sun Dec 4 12:20:27 2016 +0100
> > 
> >     uclibc: update to 1.0.20
> >     
> >     The test suite is removed from the package and is already a separate
> >     package in buildroot. All patches are upstream, so remove them.
> >     The UCLIBC_HAS_LFS option is removed upstream.
> >     
> >     Signed-off-by: Waldemar Brodkorb <wbx@openadk.org>
> >     [Thomas: remove BR2_UCLIBC_INSTALL_TEST_SUITE option, add it to
> >     Config.in.legacy.]
> >     Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
> 
> Seems like the story is more complicated. I did a build of czmq right
> before the uClibc bump to 1.0.20, and right after, and both builds were
> successful.
> 
> Two options:
> 
>  - This commit is not the culprit.
> 
>  - My simple configuration with just czmq doesn't trigger the problem,
>    and it's only czmq with some optional dependencies that triggers the
>    problem.

I have a clear idea what is the problem in uClibc-ng. It is arm
specific. But I have trouble to reproduce the czmq issue with the
internal static arm toolchain. (simple config, not the autobuild
config).

Is there any difference between internal and external static
toolchains, when both created with buildroot?

best regards
 Waldemar  

[Buildroot] [autobuild.buildroot.net] Your build results for 2017-02-15

[Thomas Petazzoni]

Hello,

On Thu, 16 Feb 2017 20:14:45 +0100, Waldemar Brodkorb wrote:

> I have a clear idea what is the problem in uClibc-ng. It is arm
> specific. But I have trouble to reproduce the czmq issue with the
> internal static arm toolchain. (simple config, not the autobuild
> config).

You can try to reproduce this build failure:

  http://autobuild.buildroot.net/results/704/7041b34327b0752d3f4294a5541c785dd9989d51/

It occurs with the internal toolchain. Steps to reproduce:

 mkdir foobar
 cd foobar
 wget http://git.buildroot.net/buildroot-test/plain/utils/br-reproduce-build
 chmod +x br-reproduce-build
 BASE_GIT=/path/to/local/buildroot/git/repository ./br-reproduce-build 7041b34327b0752d3f4294a5541c785dd9989d51

> Is there any difference between internal and external static
> toolchains, when both created with buildroot?

The only differences are:

 1. host-mpc, host-mpfr and host-gmp are linked statically inside gcc,
    so that we don't have rpath issues

 2. the toolchain wrapper is tweak to support the moving of files
    outside of the usr/ subdir that I'm doing as part of the toolchain
    "packaging"

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

[Buildroot] [autobuild.buildroot.net] Your build results for 2017-02-15

[Waldemar Brodkorb]

Hi,

> Am 16.02.2017 um 22:15 schrieb Thomas Petazzoni <thomas.petazzoni@free-electrons.com>:
> 
> Hello,
> 
>> On Thu, 16 Feb 2017 20:14:45 +0100, Waldemar Brodkorb wrote:
>> 
>> I have a clear idea what is the problem in uClibc-ng. It is arm
>> specific. But I have trouble to reproduce the czmq issue with the
>> internal static arm toolchain. (simple config, not the autobuild
>> config).
> 
> You can try to reproduce this build failure:
> 
>  http://autobuild.buildroot.net/results/704/7041b34327b0752d3f4294a5541c785dd9989d51/
> 
> It occurs with the internal toolchain. Steps to reproduce:
> 
> mkdir foobar
> cd foobar
> wget http://git.buildroot.net/buildroot-test/plain/utils/br-reproduce-build
> chmod +x br-reproduce-build
> BASE_GIT=/path/to/local/buildroot/git/repository ./br-reproduce-build 7041b34327b0752d3f4294a5541c785dd9989d51
> 
>> Is there any difference between internal and external static
>> toolchains, when both created with buildroot?
> 
> The only differences are:
> 
> 1. host-mpc, host-mpfr and host-gmp are linked statically inside gcc,
>    so that we don't have rpath issues
> 
> 2. the toolchain wrapper is tweak to support the moving of files
>    outside of the usr/ subdir that I'm doing as part of the toolchain
>    "packaging"

I could reproduce it with the glibmm package and have a tested patch ready.
I will sent it in the evening.

best regards
 Waldemar 

[Buildroot] [autobuild.buildroot.net] Your build results for 2017-02-15

[Thomas Petazzoni]

Hello,

On Fri, 17 Feb 2017 08:13:59 +0100, Waldemar Brodkorb wrote:

> I could reproduce it with the glibmm package and have a tested patch ready.
> I will sent it in the evening.

Excellent thanks!

I think the next big toolchain related issue is all the or1k failures
due to missing structure definitions in ucontext.h :-)

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com