Fix issue with libstdc++.so.6 which does not provide the required ABI version (CXXABI_1.3.9) as needed by some python packages (matplotlib and pillow).

libstdc_issue_fix.md

Solution 1: Install GCC and libstdc++ via Conda

Create or Activate a Conda Environment If you don’t already have a Conda environment, create one:

conda create -n gcc_env python=3.8
conda activate gcc_env

Install libstdcxx-ng and gcc via Conda

Conda provides libstdcxx-ng, which includes the necessary libstdc++.so.6 with newer ABI versions. To install the required dependencies:

conda install -c conda-forge libstdcxx-ng
conda install -c conda-forge gcc

Verify the ABI Version After installing, verify that the required ABI version (CXXABI_1.3.9) is available:

strings $(conda info --base)/envs/gcc_env/lib/libstdc++.so.6 | grep CXXABI

This should list the ABI versions available, including CXXABI_1.3.9.

Run Your Python Packages in the Conda Environment Now, you can install and use the Python packages like matplotlib and pillow in this environment:

conda install matplotlib pillow

Since the environment is using the Conda-provided libstdc++.so.6, your Python packages should now work without throwing the ABI version error.

Ensure Conda Environment is Active Whenever you want to run programs that require the updated libstdc++, make sure you activate the Conda environment:

conda activate gcc_env

Summary

Conda offers an easy way to manage and upgrade libraries like libstdc++ and gcc in your environment without affecting the system-wide installation. By installing libstdcxx-ng and gcc from Conda, you get the required ABI version, and your Python packages will be able to run without issues.

This method provides a hassle-free solution since Conda automatically manages dependencies and paths.

Solution 2: Install a Newer Version of libstdc++ Locally

To resolve the issue with the outdated libstdc++.so.6 not providing the required ABI version (CXXABI_1.3.9), you need to update the libstdc++ library in your local environment, as you likely don't have sudo privileges to update system-wide libraries. Here's the step-by-step solution:

1. Download GCC (which provides libstdc++)

   • libstdc++.so.6 is part of the GCC toolchain. You can install a newer version of GCC locally and point your environment to the new libstdc++.so.6 file.

   # Navigate to your software directory
   mkdir -p /data/pnlx/home/rez3/software
   cd /data/pnlx/home/rez3/software
   
   # Download the latest GCC source code (or choose a version compatible with your requirements)
   wget https://ftp.gnu.org/gnu/gcc/gcc-10.3.0/gcc-10.3.0.tar.gz
   tar -xvzf gcc-10.3.0.tar.gz
   cd gcc-10.3.0
   
   # Download prerequisites for GCC
   ./contrib/download_prerequisites
   
   # Create a directory for GCC build
   mkdir build
   cd build
   
   # Configure GCC with a prefix pointing to your local directory
   ../configure --prefix=$HOME/local --disable-multilib --enable-languages=c,c++
   
   # Build and install GCC
   make -j$(nproc)
   make install

2. Update the LD_LIBRARY_PATH to Include the New libstdc++

   After installing GCC, you need to update your environment to use the newly installed libstdc++.so.6.

   # Set environment variables to point to the new libstdc++.so.6
   export LD_LIBRARY_PATH=$HOME/local/lib64:$LD_LIBRARY_PATH

3. Verify the ABI Version Once the new libstdc++.so.6 is installed, you can check its ABI version:

   strings $HOME/local/lib64/libstdc++.so.6 | grep CXXABI

   This should show the supported ABI versions, including CXXABI_1.3.9.

4. Test Your Python Packages (matplotlib, pillow) Now, retry importing matplotlib, pillow, or any other package that was failing due to the missing ABI version. Since the new libstdc++.so.6 should be in your LD_LIBRARY_PATH, the Python packages should now work correctly.

---

This process installs a newer version of libstdc++.so.6 without needing to update the system-wide libraries, allowing you to meet the ABI requirements for your Python packages.