`GLIBCXX_3.4.21' not found

[slfan2013]

The following code give me error:

reticulate::import('tensorflow')
Error in py_module_import(module, convert = convert) :
  ImportError: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.21' not found (required by /home/slfan/.local/share/r-miniconda/envs/r-reticulate/lib/python3.6/site-packages/google/protobuf/pyext/_message.cpython-36m-x86_64-linux-gnu.so)

Please see my config.

py_config()
python:         /home/slfan/.local/share/r-miniconda/envs/r-reticulate/bin/python
libpython:      /home/slfan/.local/share/r-miniconda/envs/r-reticulate/lib/libpython3.6m.so
pythonhome:     /home/slfan/.local/share/r-miniconda/envs/r-reticulate:/home/slfan/.local/share/r-miniconda/envs/r-reticulate
version:        3.6.10 |Anaconda, Inc.| (default, Mar 25 2020, 23:51:54)  [GCC 7.3.0]
numpy:          /home/slfan/.local/share/r-miniconda/envs/r-reticulate/lib/python3.6/site-packages/numpy
numpy_version:  1.19.1

I am using a Linux server system.

[kevinushey]

The problem here is that your Anaconda environment + your system's own library are requesting incompatible versions of the C++ standard library. You likely need to set LD_LIBRARY_PATH to ensure libraries from the Anaconda installation are used instead.

[slfan2013]

Shoud I set LD_LIBRARY_PATH in R or in the system, using sudo...?

[alexvorobiev]

Since reticulate wants R to load libpython.so which in turn needs newer libc than the one R was originally compiled with, LD_LIBRARY_PATH should be set prior to calling R. Unfortunately, due to the peculiar nature of libc, setting LD_LIBRARY_PATH is not going to be enough. The reason is libc should match the so called dynamic linker R is using to load the shared libraries. Basically you need to adjust the location of the dynamic linker stored in the R executable. Here are the step by step instructions https://stackoverflow.com/a/44710599/973603.

I have not tried them with Anaconda but I was facing the same issue with python (but not R) installed by Nix package manager and that approach worked. One issue you may encounter is the R executable doesn't do much - all the actual functionality is in libR.so which may have its own missing dependencies. Fortunately those can be resolved by setting LD_LIBRARY_PATH prior to calling R.

[trangdata]

I ran into a similar issue in a GitHub Action, and this answer finally help me after many trials (including setting LD_LIBRARY_PATH). Essentially, I added the following as a step in my .github/workflows/pythonapp.yml file:

conda install -c conda-forge libstdcxx-ng
sudo rm /usr/lib/x86_64-linux-gnu/libstdc++.so.6
sudo ln -s /home/runner/.local/share/r-miniconda/pkgs/libstdcxx-ng-12.1.0-ha89aaad_16/lib/libstdc++.so.6.0.30 /usr/lib/x86_64-linux-gnu/libstdc++.so.6

Explanation

My initial error was:

Error in py_call_impl(callable, dots$args, dots$keywords) : 
  ImportError: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.30' not found (required by /home/runner/.local/share/r-miniconda/envs/r-reticulate/lib/python3.8/site-packages/scipy/optimize/_highs/_highs_wrapper.cpython-38-x86_64-linux-gnu.so)

so I needed to remove the existing libstdc++.so.6 and make a symlink for it in /usr/lib/x86_64-linux-gnu/.

find / -name "libstdc++.so*" showed me where my libstdc++.so.6.0.30 file was, so then the last step was to point the symlink there.

[carmensandoval]

Is it possible to tell reticulate to look for the library in question in the conda environment lib directory?

[yoshidk6]

I took a shortcut and used an older version of a Python package (in my case xgboost) and that worked fine (detail: #1282 (comment)). Posting it here just in case this info can be helpful for some, especially for those without sudo access

[hhaensel]

conda install -c conda-forge libstdcxx-ng
sudo rm /usr/lib/x86_64-linux-gnu/libstdc++.so.6
sudo ln -s /home/runner/.local/share/r-miniconda/pkgs/libstdcxx-ng-12.1.0-ha89aaad_16/lib/libstdc++.so.6.0.30 /usr/lib/x86_64-linux-gnu/libstdc++.so.6

for my Dockerfile I even shortened that by using the force option

ln -sf /opt/conda/lib/libstdc++.so /opt/julia-1.8.2/lib/julia/libstdc++.so
ln -sf /opt/conda/lib/libstdc++.so.6 /opt/julia-1.8.2/lib/julia/libstdc++.so.6

Your post saved me <3 ❤️

[trangdata]

I ran into a similar issue in a GitHub Action, and this answer finally help me after many trials (including setting LD_LIBRARY_PATH). Essentially, I added the following as a step in my .github/workflows/pythonapp.yml file:

conda install -c conda-forge libstdcxx-ng
sudo rm /usr/lib/x86_64-linux-gnu/libstdc++.so.6
sudo ln -s /home/runner/.local/share/r-miniconda/pkgs/libstdcxx-ng-12.1.0-ha89aaad_16/lib/libstdc++.so.6.0.30 /usr/lib/x86_64-linux-gnu/libstdc++.so.6

Update on 2023-02-19: It looks like this issue has been fixed somewhere upstream (and perhaps thanks to the updated r-lib actions to v2?). Anyhow, I had to now remove this step (called Install gcc step in my action file).

[legel]

conda install -c conda-forge libstdcxx-ng
solved my problem!

[Clear-3d]

conda install -c conda-forge libstdcxx-ng solved my problem!

Thanks a lot!!!!