Reticulate not looking for shared libraries in conda env lib directory

[carmensandoval]

library(reticulate)
reticulate::use_condaenv("/path/to/conda/envs/pegasus",
                         conda = "/path/to/conda/condabin/conda",
                         required = TRUE)

python:         /path/to/conda/envs/pegasus/bin/python
libpython:      /path/to/conda/envs/pegasus/lib/libpython3.9.so
pythonhome:     /path/to/conda/envs/pegasus
version:        3.9.13 | packaged by conda-forge | (main, May 27 2022, 16:56:21)  [GCC 10.3.0]
numpy:          /path/to/conda/envs/pegasus/lib/python3.9/site-packages/numpy
numpy_version:  1.23.3

NOTE: Python version was forced by RETICULATE_PYTHON

import pegasus as pg

ImportError: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.26' not found (required by /path/to/conda/envs/pegasus/lib/python3.9/site-packages/scipy/linalg/_matfuncs_sqrtm_triu.cpython-39-x86_64-linux-gnu.so)

The file where it's looking naturally does not have the requited GLIBXX version:

ls /usr/lib/x86_64-linux-gnu/libstdc++.so.6
/usr/lib/x86_64-linux-gnu/libstdc++.so.6 -> libstdc++.so.6.0.25

strings /usr/lib/x86_64-linux-gnu/libstdc++.so.6 | grep 

GLIBCXX_3.4
GLIBCXX_3.4
GLIBCXX_3.4.1
GLIBCXX_3.4.2
GLIBCXX_3.4.3
GLIBCXX_3.4.4
GLIBCXX_3.4.5
GLIBCXX_3.4.6
GLIBCXX_3.4.7
GLIBCXX_3.4.8
GLIBCXX_3.4.9
GLIBCXX_3.4.10
GLIBCXX_3.4.11
GLIBCXX_3.4.12
GLIBCXX_3.4.13
GLIBCXX_3.4.14
GLIBCXX_3.4.15
GLIBCXX_3.4.16
GLIBCXX_3.4.17
GLIBCXX_3.4.18
GLIBCXX_3.4.19
GLIBCXX_3.4.20
GLIBCXX_3.4.21
GLIBCXX_3.4.22
GLIBCXX_3.4.23
GLIBCXX_3.4.24
GLIBCXX_3.4.25

But libstdc++.so.6.0.30 in /path/to/conda/env/ does have the required version:

ls -lha /path/to/conda/lib | grep libstdc

lrwxrwxrwx   1 2020911 1279   19 Jul  7 06:31 libstdc++.so -> libstdc++.so.6.0.30
lrwxrwxrwx   1 2020911 1279   19 Jul  7 06:31 libstdc++.so.6 -> libstdc++.so.6.0.30
-rwxrwxr-x+  4 2020911 1279  12M May 11 20:30 libstdc++.so.6.0.30

strings /path/to/conda/lib/libstdc++.so.6.0.30 | grep GLIBCXX_3.4

GLIBCXX_3.4
GLIBCXX_3.4.1
GLIBCXX_3.4.2
GLIBCXX_3.4.3
GLIBCXX_3.4.4
GLIBCXX_3.4.5
GLIBCXX_3.4.6
GLIBCXX_3.4.7
GLIBCXX_3.4.8
GLIBCXX_3.4.9
GLIBCXX_3.4.10
GLIBCXX_3.4.11
GLIBCXX_3.4.12
GLIBCXX_3.4.13
GLIBCXX_3.4.14
GLIBCXX_3.4.15
GLIBCXX_3.4.16
GLIBCXX_3.4.17
GLIBCXX_3.4.18
GLIBCXX_3.4.19
GLIBCXX_3.4.20
GLIBCXX_3.4.21
GLIBCXX_3.4.22
GLIBCXX_3.4.23
GLIBCXX_3.4.24
GLIBCXX_3.4.25
GLIBCXX_3.4.26
GLIBCXX_3.4.27
GLIBCXX_3.4.28
GLIBCXX_3.4.29
GLIBCXX_3.4.30

Trying to add the conda env library directory (/path/to/conda/lib) to $LD_LIBRARY_PATH doesn't effect any change:

system("export LD_LIBRARY_PATH=/path/to/conda/lib:$LD_LIBRARY_PATH")
system("echo LD_LIBRARY_PATH") 

Returns an unchanged path, without including /path/to/conda/lib

How can I specify that it needs to look in /path/to/conda/env/lib/ for that GLIBCXX_3.4.26 library?

I can load import other modules from this same env without any issues.

[jan-glx]

My understanding is that you cannot do stuff like that (i.e. almost anything slightly advanced) in general with reticulate as the reticulate python runs in the same process as R and thus are bound to have the same environment. If there requirements for the environment are conflicting there is nothing you can do.

One (painful) solution is t use R from within your conda environment (installing it there).

Another, probably better, solution is to use python in a separate process with basilisk (which, i believe, combines calleR with reticulate to achieve that) .

[kevinushey]

In some cases you can, but you need to set your LD_LIBRARY_PATH before R is launched so that Conda libraries are preferred to the default system libraries.

Or just use a virtual environment...

[davidzgli]

If you do not mind running R standalone, for example, starting R session in a terminal, reticulate package does honor the GLIBCXX in a conda envrionment. Here is the procedure that works for me.

1. Create a cond env
2. Install appropriate version of gcc and g++ in the environment
   conda install -c conda-forge gcc
conda install -c conda-forge gxx
3. Export LD_LIBRARY_PATH environment variable in a terminal.
   export LD_LIBRARY_PATH=/paths/envs/env_name/lib:$LD_LIBRARY_PATH
4. Start R in this terminal
5. Check LD_LIBRARY_PATH via the command Sys.getenv("LD_LIBRARY_PATH")

One nasty way to make reticulate honors conda env in Rstudio server is to modify R system file ldpaths , i.e., put export LD_LIBRARY_PATH=/paths/envs/env_name/lib:$LD_LIBRARY_PATH at the bottom of a file like /opt/R/4.3.2/lib/R/etc/ldpaths on CentOS.