I recently installed a bunch of dotfiles on my Mac along with some other applications (I changed to iTerm instead of Terminal, and Sublime as my default text editor) but ever since, all my virtual environments have stopped working, although their folders inside .virtualenvs are still there and they give the following error whenever I try to run anything in them:
dyld: Library not loaded: @executable_path/../.Python
Referenced from: /Users/[user]/.virtualenvs/modclass/bin/python
Reason: image not found
Trace/BPT trap: 5
I have removed all the files related to dotfiles and have restored my .bash_profile to what it was before, but the problem persists. Is there any way to diagnose the problem or solve it in an easy way (e.g. not requiring to create all the virtualenvs all over again)?
I found the solution to the problem here, so all credit goes to the author.
The gist is that when you create a virtualenv, many symlinks are created to the Homebrew installed Python.
Here is one example:
$ ls -la ~/.virtualenvs/my-virtual-env
...
lrwxr-xr-x 1 ryan staff 78 Jun 25 13:21 .Python -> /usr/local/Cellar/python/2.7.7/Frameworks/Python.framework/Versions/2.7/Python
...
When you upgrade Python using Homebrew and then run brew cleanup, the symlinks in the virtualenv point to paths that no longer exist (because Homebrew deleted them).
The symlinks needs to point to the newly installed Python:
lrwxr-xr-x 1 ryan staff 78 Jun 25 13:21 .Python -> /usr/local/Cellar/python/2.7.8_1/Frameworks/Python.framework/Versions/2.7/Python
The solution is to remove the symlinks in the virtualenv and then recreate them:
find ~/.virtualenvs/my-virtual-env/ -type l -delete
virtualenv ~/.virtualenvs/my-virtual-env
It's probably best to check what links will be deleted first before deleting them:
find ~/.virtualenvs/my-virtual-env/ -type l
In my opinion, it's even better to only delete broken symlinks. You can do this using GNU find:
gfind ~/.virtualenvs/my-virtual-env/ -type l -xtype l -delete
You can install GNU find with Homebrew if you don't already have it:
brew install findutils
Notice that by default, GNU programs installed with Homebrew tend to be prefixed with the letter g. This is to avoid shadowing the find binary that ships with OS X.
After trying a few things, this worked for me:
go to your virtualenv directory (but don't run workon):
cd ~/.virtualenv/name_of_broken_venv
Now delete these files:
rm -rf .Python bin/python* lib/python2.7/* include/python2.7
Then to rebuild your venv, run:
virtualenv .
workon name_of_broken_venv
pip freeze
You should now see a list of your installed packages again.
This occurred when I updated to Mac OS X Mavericks from Snow Leopard. I had to re-install brew beforehand too. Hopefully you ran the freeze command for your project with pip.
To resolve, you have to update the paths that the virtual environment points to.
- Install a version of python with brew:
brew install python
- Re-install virtualenvwrapper.
pip install --upgrade virtualenvwrapper
- Removed the old virtual environment:
rmvirtualenv old_project
- Create a new virtual environment:
mkvirtualenv new_project
- Work on new virtual environment
workon new_project
- Use pip to install the requirements for the new project.
pip install -r requirements.txt
This should leave the project as it was before.
It appears the proper way to resolve this issue is to run
pip install --upgrade virtualenv
after you have upgraded python with Homebrew.
This should be a general procedure for any formula that installs something like python, which has it's own package management system. When you install brew install python, you install python and pip and easy_install and virtualenv and so on. So, if those tools can be self-updated, it's best to try to do so before looking to Homebrew as the source of problems.
If this was caused by a brew upgrade that upgraded its Python, and you're ok with downgrading to the previous version, try brew switch python [previous version], eg brew switch python 3.6.5. From here.
A update version @Chris Wedgwood's answer for keeping site-packages (keeping packages installed)
cd ~/.virtualenv/name_of_broken_venv
mv lib/python2.7/site-packages ./
rm -rf .Python bin lib include
virtualenv .
rm -rf lib/python2.7/site-packages
mv ./site-packages lib/python2.7/
If you've busted python3 just try brew upgrade python3 that fixed it for me.
I recently faced this. None of the above solutions worked for me. Seems it wasn't actually Python's problem. When I was running aws s3 ls
I was getting following error: dyld: Library not loaded: @executable_path/../.Python
This means, the library aws executable is pointing towards is either doesn't exist or is corrupted, thus I uninstalled and reinstalled aws-cli following instructions from this link and it worked!!
virtualenvwrapper instructions
As indicated in the accepted answer, the root cause is likely a homebrew update that means your virtualenv symlinks are pointing at broken python paths - see details here.
For each virtual env, you need to reassign the symlinks to point at the correct python path (in brew cellar). Here is how to do it with virtualenvwrapper. Here I am updating a virtual env called "my-example-env".
cd ~/PYTHON_ENVS
find ./my-example-env -type l -delete
mkvirtualenv my-example-env
All done.
Using Python 2.7.10.
A single command virtualenv path-to-env does it. documentation
$ virtualenv path-to-env
Overwriting path-to-env/lib/python2.7/orig-prefix.txt with new content
New python executable in path-to-env/bin/python2.7
Also creating executable in path-to-env/bin/python
Installing setuptools, pip, wheel...done.
The problem for me(a MacOS user) is that brew updated the Python and virtualenvs links to the old version which was deleted.
We can check and fix it by
>> ls -al ~/.virtualenvs/<your-virtual-env>/.Python
.Python -> /usr/local/Cellar/python/<old-version>/Frameworks/Python.framework/Versions/3.7/Python
>> rm ~/.virtualenvs/<your-virtual-env>/.Python
>> ln -s /usr/local/Cellar/python/<new-version>/Frameworks/Python.framework/Versions/3.7/Python ~/.virtualenvs/<your-virtual-env>/.Python
Anyone who is using pipenv (and you should!) can simply use these two commands — without having the venv activated:
rm -rf `pipenv --venv` # remove the broken venv
pipenv install --dev # reinstall the venv from pipfile
The accepted answer does not work for me: the file $WORKON_HOME/*/bin/python2.7 is no longer a symlink, it is a full-fledged executable:
$ file $WORKON_HOME/*/bin/python2.7
/Users/sds/.virtualenvs/.../bin/python2.7: Mach-O 64-bit executable x86_64
...
The solution is, alas, to completely remove and re-create from scratch all the virtual environments.
For the reference:
deactivate
pip install --user virtualenv virtualenvwrapper
pip install --user --upgrade virtualenv virtualenvwrapper
for ve in $(lsvirtualenv -b); do
# assume that each VE is associated with a project
# and the project has the requirements.txt file
project=$(cat $WORKON_HOME/$ve/.project)
rmvirtualenv $ve
mkvirtualenv -a $project -r requirements.txt $ve
done
Simply upgrading python3 worked for me:
brew upgrade python3
I tried the top few methods, but they didn't work, for me, which were trying to make tox work. What eventually worked was:
sudo pip install tox
even if tox was already installed. The output terminated with:
Successfully built filelock
Installing collected packages: py, pluggy, toml, filelock, tox
Successfully installed filelock-3.0.10 pluggy-0.11.0 py-1.8.0 toml-0.10.0 tox-3.9.0
What fixed it for me was just uninstalling python3 and pipenv then reinstalling them.
brew uninstall pipenv
brew uninstall python3
brew install python3
brew install pipenv
I had a broken virtual env due to a Homebrew reinstall of python (thereby broken symlinks) and also a few "sudo pip install"s I had done earlier. Weizhong's tips were very helpful in fixing the issues without having to reinstall packages. I also had to do the following for the mixed permissions problem.
sudo chown -R my_username lib/python2.7/site-packages
来源：https://stackoverflow.com/questions/23233252/broken-references-in-virtualenvs