Python/pip process are killed in virtualenv (Apple M1 chip)

Question

When I execute pip or python in MacBook M1 chip virtualenv, the process always terminates immediately with message like:

[1]    29116 killed     pip

or

[1]    29141 killed     python

Python and pip in M1 MacBook native environment run fine.

Only in the virtualenv they don't work.

Answer 1

After some research, I find out it is related with "x86_64" and "arm 64" architecture. To solve the problem, you have to run virtualenv python/pip in x86_64 mode.

Steps:

1. Disable SIP: reboot MacBook in Recover Mode. Open Utilities > Terminal enter csrutil disable. ref: Disabling and Enabling System Integrity Protection
2. Run SHELL as x86_64 mode: enter arch -x86_64 $SHELL in Terminal.
3. Execute python/pip successfully

After that, I enable SIP again and keep using arch -x86_64 $SHELL every time. It will execute python/pip in virtualenv succesfully.

To save my time, I add an alias in my .zshrc:

alias x86="arch -x86_64 $SHELL"

Now if I need execute some process in x86_64 mode, I just type x86 to open the x86_64 shell.

Note: MxJ24 on github provides this solution and lists some useful scripts for python/django/brew: https://gist.github.com/MxJ24/e1386c9012f533cfbedfed5114da3e60

Answer 2

Python 2 virtualenvs are just not working on Macs with Apple Silicon chip as of now. See https://github.com/pypa/virtualenv/issues/2024 for more information.

The workarounds involving disabling System Integrity Protection and using x86_64 emulation may be unsafe and inefficient and I suggest against following them.

The best solution would obviously be to switch/migrate/upgrade to Python 3, for which virtualenv works like a charm.

If that's not a viable option, then I would consider using Docker instead of virtualenv. python:2 Docker image has the latest Python 2.7.18 version and is available on aarch64 architecture, so it will run without x86_64 emulation. I don't know your use case, but I am using IntelliJ IDEA/PyCharm and it supports using Docker as the SDK, so after a one-time setup it's seamless.