DEV Community

Stop using sudo pip install

eLabFTW on October 17, 2018

We've all done it: pip install numpy # run into permissions issues sudo pip install numpy # or "sudo !!" for the power users ;) ~~~{% endraw %} ...
Collapse
 
codesandbox profile image
code-sandbox

sudo pip is for global packages, virtualenv for literally everything else. Neither venv or pipenv will prevent you getting pwnd if you're careless enough to install a malicious package. They are not effective security measures.

Collapse
 
blubberdiblub profile image
Niels Böhm • Edited

Using sudo pip on distros that provide native Python packages (apt-get install python-numpy, etc.) is calling for trouble.

Collapse
 
codesandbox profile image
code-sandbox

Fair enough. I do prefer the native package manager route when it's an option, but in those cases it's generally handled automatically as a dependency anyways. Really, 99.99% of all manual pip interactions should be happening in some sandbox env anyways. Regardless it's better practice to understand and respect root ops rather than fear them, because sometimes they are necessary.

Collapse
 
tmr232 profile image
Tamir Bahar

When would a pip install --user not suffice instead of a global install?

Collapse
 
matthutchison profile image
Matt Hutchison

We see it when we're installing an application that will be used by another user account, since ~/.local/lib (or the equivalent) isn't shared. This is pretty rare in a development environment, but it comes up frequently in an administered multi-user setup (say a shared workstation or batch cluster). Sometimes service accounts as well depending on what they're doing.

Collapse
 
rupankarghosh profile image
RupankarGhosh

After running pip with sudo in my arch Linux I messed-up some of my system files. And now the whole system is broken.

Collapse
 
ijstokes profile image
Ian Stokes-Rees

Or use conda which solves this problem for Python plus many other languages and arbitrary binaries.

Collapse
 
elabftw profile image
eLabFTW

but conda is not per project, is it?

Collapse
 
rodolfoferro profile image
Rodolfo Ferro

You can create a specific conda environment for each proyect, and even specify the conda channel from which it is installed.

I usually create an environment.yml file for this, so I only run

$ conda env create -f environment.yml

to create the conda env.

After this I only do:

# Activate conda env:
$ conda activate (env-name)

# Deactivate conda env:
$ conda deactivate

(Depending on your Anaconda version, to activate/deactivate the env, you can use source instead of conda.)

Thread Thread
 
elabftw profile image
eLabFTW

Good to know thx.

Collapse
 
phydroxide profile image
phydroxide

What about for pip install --upgrade pip

"ERROR: Could not install packages due to an EnvironmentError: [Errno 13] Permission denied: '/usr/bin/pip3'
"

How do you get pip into /usr/bin without running it privileged?

Collapse
 
elabftw profile image
eLabFTW

You don't, that's the whole point. Unless you give your user permissions to write in /usr/bin, which is not a good idea.

Collapse
 
vikrantsingh47 profile image
vikrant singh

should we install pip itself with sudo?
eg: sudo apt install python3-pip
and then ,
pip install --upgrade pip

is that corrent process?

Collapse
 
dvershinin profile image
Danila Vershinin

Here's a simple virtualenv manager that you can use to install Python apps in a safe way: github.com/dvershinin/pip-safe You can then just type pip-safe install <pypy-name> and it will take care of installing stuff and making it available on PATH. (you can use it instead of typing "pip install" and breaking things :)

Collapse
 
mandarvaze profile image
Mandar Vaze

I also prefer pipenv where I can.
But it hasn't worked 100% of the times :(
So had to "fall back to" manually creating virtualenv followed by pip install

Collapse
 
errietta profile image
Erry Kostala

I use pipenv, works great!

Collapse
 
architekco profile image
Evan

Yes yes but how to reset the nightmare that is my current python environment(s)? Honestly I'd be happy resetting my entire Dev ecosystem (but would like to avoid reformatting). MacOS of course.

Collapse
 
elabftw profile image
eLabFTW

Simple: install a real OS like GNU/Linux or Free/OpenBSD on your mac and you're good to go :p

No seriously macos is fucked up in various ways… But if you use pipenv or similar, you don't really have to worry about what is installed on the rest of the computer.

Collapse
 
architekco profile image
Evan

Right, I do use virtual environments these days, kinda... but ... my mac is a barren wasteland of python installs and duplicate packages and all sorts of other mind-boggling environment fuckery that I have been putting off dealing with! Ill get to it one of these days, I guess.

Collapse
 
bradleygrant profile image
Bradley Grant

But venv.

Collapse
 
xiaodaigh profile image
evalparse

how do yu specify which environment to use?