New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Wrong libffi being selected on WSL for dynamically-linked python extensions #12287
Comments
Hi @erip, Thanks for filing this issue. We have tagged it as
os::wsl
We cannot give an estimate currently regarding when this will be resolved, but we encourage you to keep an eye on this issue or update us here when anything changes. |
So I'm seeing a very similar issue on Azure ML with an ubuntu image based on this image when I use ffmpeg. Whatever changed has happened since Feb 2nd on AML though I'm struggling to track anything down. The only difference I see is libffi went from 3.3 to 3.4.1 and python from 3.8.13 to 3.8.15/16 between working and broken envs. Pinning libffi to 3.3 seems to fix things for me, although above looks like you're already on that version - so maybe unrelated |
I am having the same issue. This only happens with Conda environments when using OpenCV+ROS for some reason. Eventually, I ended up using virtualenv instead of conda |
I hit the same issue and the cause seems to be due to libffi.7.so being incorrectly linked to libffi.8.so. See the files below which got installed by conda - "conda install libffi==3.4.2" :/opt/conda/.../lib# ll | grep ffi Why would it point an older version (libffi7) to newer version (libffi8)? Seems wrong and defeats the purpose of versioning. If we remove this link then the correct libffi.7.so installed in the system gets used. |
Have the same issue. |
@Starrah You save my day. |
That issue is none existence in python 3.10. |
@Vulum that's not true. My reproducer explicitly uses 3.10 |
@erip: Ok, probably the version in his conda env must be 3.10 but the python version in his conda base env must be < 3.10. |
I solve with |
thanks this worked for me |
I occur this error when use psycopg2, and @Tlaloc-Es way can fix this bug. But on the time, my another compute with same config can run normally. So I compare two Ubuntu's libffi version use
And I complie docker image with
But on the error computer, os version:
So, every psycopg2 run properly with machine libffi version 3.4, but error on 3.3. |
Checklist
What happened?
tl;dr: more details at pytorch/data#961
When importing a python extension library on WSL through conda, I find that libffi is being installed and linked against inappropriately resulting in
ImportError: /usr/lib/x86_64-linux-gnu/libp11-kit.so.0: undefined symbol: ffi_type_pointer, version LIBFFI_BASE_7.0
at import time.Conda Info
Conda Config
No response
Conda list
Additional Context
Reproducer on WSL:
The text was updated successfully, but these errors were encountered: