undefined Symbol

[MarkusEich]

Running the script results in and error

Traceback (most recent call last):
File "src/SingleHandTracking.py", line 8, in
import PyMBVCore as Core
ImportError: /home/eich/devel/MBV_PythonAPI_Linux_1.1/libs/libCore.so: undefined symbol: ZN2cv9ExceptionC1EiRKSsS2_S2

System: CUDA 7.5, NVIDIA 352, Ubuntu 15.10

[padeler]

Hello,

From the error message it looks like the cv::Exception is undefined. Please make sure that all opencv packages are installed on your system. (libopencv-dev, python-opencv)

[severin-lemaignan]

Hello, same issue here on Ubuntu 15.10. OpenCV 2.4.9 and all Python bindings are installed.

Here the output of ldd on libCore:

$ ldd libs/libCore.so 
    linux-vdso.so.1 =>  (0x00007fffb2117000)
    libOpenMeshCore.so.3.1 => not found
    libcudart.so.7.5 => not found
    libboost_system.so.1.57.0 => not found
    libboost_filesystem.so.1.57.0 => not found
    libboost_thread.so.1.57.0 => not found
    libboost_serialization.so.1.57.0 => not found
    libboost_program_options.so.1.57.0 => not found
    libboost_log.so.1.57.0 => not found
    libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007ffad74b4000)
    libtbb.so.2 => /usr/lib/libtbb.so.2 (0x00007ffad7275000)
    libopencv_core.so.2.4 => /usr/lib/x86_64-linux-gnu/libopencv_core.so.2.4 (0x00007ffad6e4a000)
    libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007ffad6ac8000)
    libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007ffad67bf000)
    libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007ffad65a8000)
    libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007ffad61de000)
    /lib64/ld-linux-x86-64.so.2 (0x0000560399802000)
    libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007ffad5fd9000)
    libz.so.1 => /lib/x86_64-linux-gnu/libz.so.1 (0x00007ffad5dbf000)
    librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007ffad5bb7000)
    libGL.so.1 => /usr/lib/nvidia-352/libGL.so.1 (0x00007ffad5886000)
    libnvidia-tls.so.352.41 => /usr/lib/nvidia-352/tls/libnvidia-tls.so.352.41 (0x00007ffad5683000)
    libnvidia-glcore.so.352.41 => /usr/lib/nvidia-352/libnvidia-glcore.so.352.41 (0x00007ffad2bf0000)
    libX11.so.6 => /usr/lib/x86_64-linux-gnu/libX11.so.6 (0x00007ffad28b6000)
    libXext.so.6 => /usr/lib/x86_64-linux-gnu/libXext.so.6 (0x00007ffad26a4000)
    libxcb.so.1 => /usr/lib/x86_64-linux-gnu/libxcb.so.1 (0x00007ffad2482000)
    libXau.so.6 => /usr/lib/x86_64-linux-gnu/libXau.so.6 (0x00007ffad227e000)
    libXdmcp.so.6 => /usr/lib/x86_64-linux-gnu/libXdmcp.so.6 (0x00007ffad2078000)

Surely enough, OpenCV is correctly found. So I'm not sure why cv::Exception is not found.

There are also other issues here (beyond libcuda not found -- I've not installed it yet):

• It seems that the library was compiled against boost 1.57 (Ubuntu 15.10 ships with boost-1.58).
• I'm not sure what libOpenMeshCore is, but it seems it is not packaged in ubuntu.

[padeler]

Hello,

The libraries you are missing are inside the binaries package in the "libs" folder. You need to set the MBV_LIBS environment variable as explained in the Readme.md. Alternatively you can set the LD_LIBRARY_PATH and PYTHONPATH manually. Check the runme.sh for details.

[severin-lemaignan]

Well, if I got the error with the missing symbol in libCore in the first place, it is because libCore was found, ie, MBV_LIBS was correclty exported ;-)

[severin-lemaignan]

Hum, I see, you manipulate the LD_LIBRARY_PATH in runme.sh, and you ship you own versions of boost and OpenMeshCore. Alright. I'll conduct further tests, then.

[severin-lemaignan]

Could you tell us which version of OpenCV you have been using to compile the library?

[padeler]

The libraries where compiled with Ubuntu 14.04 which comes with OpenCV 2.4.8

[severin-lemaignan]

Alright, I think I've figured out the reason: on Ubuntu 15.10, the default compiler is GCC 5, which effectively changes the ABI of std::string (see here for instance).

One option is to recompile a local OpenCV with gcc < 5, or to recompile OpenCV with gcc 5 and the additional flag -D_GLIBCXX_USE_CXX11_ABI=0.

You may or may not want to provide binaries compiled (well, linked) on a system using gcc 5 as default compiler to avoid the issue (or alternatively, ship the required opencv libraries as well).

[padeler]

Thanks for the find Severin,

To keep things simple we decided to only offer binaries for the LTS release of Ubuntu and tried to avoid packaging libs that are provided by the system. Obviously this leads to issues. Maybe in the next binaries update we will also include the OpenCV libs.

[MarkusEich]

Thank you for the update. Maybe it is easier to have several ubuntu versions for the compiled libs? At least for the LTS and for the current version? Would keep the libs shipped with the tracker to a minimum.

Thanks

Markus

Am 17. November 2015 23:18:06 GMT+10:00, schrieb Panteleris Paschalis notifications@github.com:

Thanks for the find Severin,

To keep things simple we decided to only offer binaries for the LTS
release of Ubuntu and tried to avoid packaging libs that are provided
by the system. Obviously this leads to issues. Maybe in the next
binaries update we will also include the OpenCV libs.

---

Reply to this email directly or view it on GitHub:
#2 (comment)

[MarkusEich]

Confirmed the work around for the issue. After downloading OpenCV 2.4 and compiling it using

export CC=/usr/bin/gcc-4.9
export CXX=/usr/bin/g++-4.9
cmake .. -D_GLIBCXX_USE_CXX11_ABI=0 -DCUDA_HOST_COMPILER=/usr/bin/gcc-4.9

The tracker runs smoothly under Ubuntu 15.10, CUDA 7.5

Thnx guys

[padeler]

Thanks for the confirmation Markus. I am closing this issue.