Package building issues

[georgegu1997]

Hi,

Thanks for the amazing work! The engineering efforts put into building a server to collect 3D scans are fantastic!

However, I have found it very challenging to build the multiscan package. I might have done it the hard way and I am still stuck half way there. Here is my debugging log. Hope I could receive some advise on how I should do it. Overall I am following the instructions shown here.

• Start with a fresh Conda environment and use the python=3.8 as instructed.
• The conda install took some time in solving env but they worked out.
• When conda install -c nvidia cuda-nvcc, make sure the version of cuda-nvcc matches (or smaller than?) the version of cuda shown by nvidia-smi.
  • Otherwise this may lead to complains about CUDA version being 7.x
• Run make -j1 instead of using all CPU cores. Otherwise, you may be overwhelmed by the complaints coming from different compilation threads.
• You can disable building open3d from source as suggested in this [issue](Issue with Build Server #7). Use DMTS_BUILD_OPEN3D=OFF when running cmake and install open3d prebuilt wheel by pip install open3d==0.16.0
  • The full cmake command will be cmake -DCMAKE_PREFIX_PATH="${CONDA_PREFIX}" -DCMAKE_INCLUDE_PATH="${CONDA_PREFIX}/include" -DCMAKE_LIBRARY_PATH="${CONDA_PREFIX}/lib" -DCMAKE_INSTALL_PREFIX="${CONDA_PREFIX}" -DMTS_BUILD_OPEN3D=OFF ..
  • The open3d building process takes a long time and might be buggy
• In my desktop, somehow the glibc is placed at /lib/x86_64-linux-gnu/ rather than /lib64, where the cmake will look into. So you may need to place some soft link to get the compilation working. The commands I used are as follows.

sudo ln -s /lib/x86_64-linux-gnu/libpthread.so.0 /lib64/libpthread.so.0
sudo ln -s /lib/x86_64-linux-gnu/libc.so.6 /lib64/libc.so.6
sudo ln -s /lib/x86_64-linux-gnu/libc_nonshared.a /usr/lib64/libc_nonshared.a
sudo ln -s /lib/x86_64-linux-gnu/librt.so.1 /lib64/librt.so.1

• I also needed to install gcc and gxx using conda. I tried gcc=12.x and it worked.
  • I remember gcc=13.x also did not work.
  • The version of the system-wide gcc is 9.x and it complained extern C errors during compilation.
• After all these, I still get errors like
  /home/guqiao/anaconda3/envs/multiscan/bin/../lib/gcc/x86_64-conda-linux-gnu/12.1.0/../../../../x86_64-conda-linux-gnu/bin/ld: /lib/x86_64-linux-gnu//libgcc_s.so.1: undefined reference to memcpy@GLIBC_2.14, and I don’t know how to solve it.

I will be grateful if the authors can look into the issue of non-matching glibc version (the last bullet point).

Thanks!

[SamMaoYS]

Thank you @georgegu1997 for the hard work you put into building the server and reporting the issues.

Yes, we also notice installing gcc, cuda and some of the libraries with conda probably is not the most convenient way and will potentially run into issues as you reported. We are working on updating the installation doc to add options to use Linux package manage apt install command.

We hope this can be done in a day or two. Thank you for your patience and we will let you know once this update is done.

[georgegu1997]

Hi @SamMaoYS

Thanks for the response. I wonder whether the repo has been updated with the fixes.