Error Compiling C++ Code with Dijitso in FEniCS Environment FEniCS2019.1

installation
1

Dear all,

For a new project on a compute cluster I am attempting to install legacy FEniCS using conda.

I use the following environment.yml file to create my conda environment:

name: newenv

channels:
  - conda-forge

dependencies:
  - fenics
  - pip
  - git
  - python >=3.8

Installation does not generate any errors.

Then, I try the following:

from dolfin import *

Exp = Expression('5', degree=0)

Upon which I receive the following error message:

dijitso failed to load existing file:
/home/user/.conda/envs/testenv/.cache/dijitso/lib/libdijitso-dolfin_expression_c14489d6e2980e09728dad365e75ccf0.so
error is:
libpetsc.so.3.15: cannot open shared object file: No such file or directory
dijitso failed to load existing file:
/home/user/.conda/envs/testenv/.cache/dijitso/lib/libdijitso-dolfin_expression_c14489d6e2980e09728dad365e75ccf0.so
error is:
libpetsc.so.3.15: cannot open shared object file: No such file or directory
Traceback (most recent call last):
  File "/home/user/.conda/envs/testenv/lib/python3.12/site-packages/dolfin/jit/jit.py", line 168, in compile_class
    submodule = dijitso.extract_factory_function(module, "create_" + module_name)()
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/user/.conda/envs/testenv/lib/python3.12/site-packages/dijitso/jit.py", line 47, in extract_factory_function
    function = getattr(lib, name)
               ^^^^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'create_dolfin_expression_c14489d6e2980e09728dad365e75ccf0'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/user/.conda/envs/testenv/lib/python3.12/site-packages/dolfin/function/expression.py", line 400, in __init__
    self._cpp_object = jit.compile_expression(cpp_code, params)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/user/.conda/envs/testenv/lib/python3.12/site-packages/dolfin/function/jit.py", line 158, in compile_expression
    expression = compile_class(cpp_data, mpi_comm=mpi_comm)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/user/.conda/envs/testenv/lib/python3.12/site-packages/dolfin/jit/jit.py", line 170, in compile_class
    raise RuntimeError("Unable to compile C++ code with dijitso")
RuntimeError: Unable to compile C++ code with dijitso

My thoughts:
It appears that the issue may be related to a missing shared object file (libpetsc.so3.15) which is needed by dijitso during the compilation process. I think it is not necessary missing, but I think the environment variables are not correctly set up such that they point to the wrong location for this file.

What I tried:

  • “conda list” tells me that petsc version 3.20 is installed, and actually there is a file /home/user/.conda/envs/testenv/lib/libpetsc.so.3.20

  • I tried installing fenics 2018.1 in the same way, and I get a similar error using the same example code. However, this time, it concerns a different .so-file, “libmpi.so.40”

  • Setting the LD_LIBRARY_PATH variable manually:

export LD_LIBRARY_PATH=/home/user/.conda/envs/testenv/lib:$LD_LIBRARY_PATH

Trying the above steps, the issue persists.
I did not manage to find a solution to the problem yet. I wanted to share the issue with you all in hopes of getting some suggestions to approach it.

Thank you in advance for your ideas.

2

Is it possible that you already had an existing installation, and recently upgraded it. If so, you may want to clear out the existing dijitso cache (typically located under $HOME/.cache), because there may be files there which were compiled for the old installation, which may well have been using libpetsc.so.3.15 instead of PETSc 3.20.

3

Thank you for your immediate response. I believe your answer may well be the solution, as I did indeed have an existing installation. However, this is an installation that is compiled from source, using the modules available on the cluster. When installing a new version via conda, I expected that the existing installation would not interfere with the conda environment in any way, but it seems I was wrong.

In $HOME/.cache I cannot find the existing dijitso cache. I suppose it could be found in another location, since I installed the existing version from source.

Thank you for your suggestion, this should help me resolve the problem. I will send an update once I got the new installation running.

4

Your output seems to suggest that it’s in

/home/user/.conda/envs/testenv/.cache/dijitso

I’d give cleaning that up a try. Furthermore, if when building from source you had to export environment variables like LIBRARY_PATH or LD_LIBRARY_PATH in your .bashrc file (or similar), I’d undo those exports, just to make sure that they don’t interfere with the paths from conda.

5

dijitso also has its own housecleaning

dijitso clean
6

Fantastic, with your suggestions I was able to get a conda installation working. Following the instructions of @francesco-ballarin I undid the exports of LIBRARY_PATH and LD_LIBRARY_PATH. Subsequently I created a new conda environment and installed everything anew. Now, the above code snippet is working as expected.

Thank you all for your help.