[python] What's the proper way to install pip, virtualenv, and distribute for Python?

Short Question

Background

In my answer to SO question 4314376, I recommended using ez_setup so that you could then install pip and virtualenv as follows:

curl -O http://peak.telecommunity.com/dist/ez_setup.py
sudo python ez_setup.py
sudo easy_install pip
sudo pip install virtualenv

I originally pulled these instructions from Jesse Noller's blog post So you want to use Python on the Mac?. I like the idea of keeping a clean global site-packages directory, so the only other packages I install there are virtualenvwrapper and distribute. (I recently added distribute to my toolbox because of this Python public service announcement. To install these two packages, I used:

sudo pip install virtualenvwrapper
curl -O http://python-distribute.org/distribute_setup.py
sudo python distribute_setup.py

No more setuptools and easy_install

To really follow that Python public service announcement, on a fresh Python install, I would do the following:

curl -O http://python-distribute.org/distribute_setup.py
sudo python distribute_setup.py
sudo easy_install pip
sudo pip install virtualenv
sudo pip install virtualenvwrapper

Glyph's Rebuke

In a comment to my answer to SO question 4314376, SO user Glyph stated:

NO. NEVER EVER do sudo python setup.py install whatever. Write a ~/.pydistutils.cfg that puts your pip installation into ~/.local or something. Especially files named ez_setup.py tend to suck down newer versions of things like setuptools and easy_install, which can potentially break other things on your operating system.

Back to the short question

So Glyph's response leads me to my original question:

This question is related to python virtualenv setuptools distribute

The answer is


Here is a nice way to install virtualenvwrapper(update of this).

Download virtualenv-1.11.4 (you can find latest at here), Unzip it, open terminal

# Create a bootstrapenv and activate it:
$ cd ~
$ python <path to unzipped folder>/virtualenv.py bootstrapenv
$ source bootstrapenv/bin/activate

# Install virtualenvwrapper:
$ pip install virtualenvwrapper
$ mkdir -p ~/bootstrapenv/Envs

# append it to file `.bashrc`
$ vi ~/.bashrc
  source ~/bootstrapenv/bin/activate
  export WORKON_HOME=~/bootstrapenv/Envs
  source ~/bootstrapenv/bin/virtualenvwrapper.sh

# run it now.
$ source ~/.bashrc

That is it, now you can use mkvirtualenv env1, lsvirtualenv ..etc

Note: you can delete virtualenv-1.11.4 and virtualenv-1.11.4.zip from Downloads folders.


Update: As of July 2013 this project is no longer maintained. The author suggests using pyenv. (pyenv does not have built-in support for virtualenv, but plays nice with it.)

Pythonbrew is a version manager for python and comes with support for virtualenv.

After installing pythonbrew and a python-version using venvs is really easy:

# Initializes the virtualenv 
pythonbrew venv init

# Create a virtual/sandboxed environment 
pythonbrew venv create mycoolbundle  

# Use it 
pythonbrew venv use mycoolbundle

Install ActivePython. It includes pip, virtualenv and Distribute.


I came across the same problem recently. I’m becoming more partial to the “always use a virtualenv” mindset, so my problem was to install virtualenv with pip without installing distribute to my global or user site-packages directory. To do this, I manually downloaded distribute, pip and virtualenv, and for each one I ran “python setup.py install --prefix ~/.local/python-private” (with a temporary setting of PYTHONPATH=~/.local/python-private) so that setup scripts were able to find distribute). I’ve moved the virtualenv script to another directory I have on my PATH and edited it so that the distribute and virtualenv modules can be found on sys.path. Tada: I did not install anything to /usr, /usr/local or my user site-packages dir, but I can run virtualenv anywhere, and in that virtualenv I get pip.


On Ubuntu:

sudo apt-get install python-virtualenv

The package python-pip is a dependency, so it will be installed as well.


I made this procedure for us to use at work.

cd ~
curl -s https://pypi.python.org/packages/source/p/pip/pip-1.3.1.tar.gz | tar xvz
cd pip-1.3.1
python setup.py install --user
cd ~
rm -rf pip-1.3.1

$HOME/.local/bin/pip install --user --upgrade pip distribute virtualenvwrapper

# Might want these three in your .bashrc
export PATH=$PATH:$HOME/.local/bin
export VIRTUALENVWRAPPER_VIRTUALENV_ARGS="--distribute"
source $HOME/.local/bin/virtualenvwrapper.sh

mkvirtualenv mypy
workon mypy
pip install --upgrade distribute
pip install pudb # Or whatever other nice package you might want.

Key points for the security minded:

  1. curl does ssl validation. wget doesn't.
  2. Starting from pip 1.3.1, pip also does ssl validation.
  3. Fewer users can upload the pypi tarball than a github tarball.

There is no problem to do sudo python setup.py install, if you're sure it's what you want to do.

The difference is that it will use the site-packages directory of your OS as a destination for .py files to be copied.

so, if you want pip to be accessible os wide, that's probably the way to go. I do not say that others way are bad, but this is probably fair enough.


You can do this without installing anything into python itself.

You don't need sudo or any privileges.

You don't need to edit any files.

Install virtualenv into a bootstrap virtual environment. Use the that virtual environment to create more. Since virtualenv ships with pip and distribute, you get everything from one install.

  1. Download virtualenv:
  2. Unpack the source tarball
  3. Use the unpacked tarball to create a clean virtual environment. This virtual environment will be used to "bootstrap" others. All of your virtual environments will automatically contain pip and distribute.
  4. Using pip, install virtualenv into that bootstrap environment.
  5. Use that bootstrap environment to create more!

Here is an example in bash:

# Select current version of virtualenv:
VERSION=12.0.7
# Name your first "bootstrap" environment:
INITIAL_ENV=bootstrap
# Set to whatever python interpreter you want for your first environment:
PYTHON=$(which python)
URL_BASE=https://pypi.python.org/packages/source/v/virtualenv

# --- Real work starts here ---
curl -O $URL_BASE/virtualenv-$VERSION.tar.gz
tar xzf virtualenv-$VERSION.tar.gz
# Create the first "bootstrap" environment.
$PYTHON virtualenv-$VERSION/virtualenv.py $INITIAL_ENV
# Don't need this anymore.
rm -rf virtualenv-$VERSION
# Install virtualenv into the environment.
$INITIAL_ENV/bin/pip install virtualenv-$VERSION.tar.gz

Now you can use your "bootstrap" environment to create more:

# Create a second environment from the first:
$INITIAL_ENV/bin/virtualenv py-env1
# Create more:
$INITIAL_ENV/bin/virtualenv py-env2

Go nuts!

Note

This assumes you are not using a really old version of virtualenv. Old versions required the flags --no-site-packges (and depending on the version of Python, --distribute). Now you can create your bootstrap environment with just python virtualenv.py path-to-bootstrap or python3 virtualenv.py path-to-bootstrap.


I think Glyph means do something like this:

  1. Create a directory ~/.local, if it doesn't already exist.
  2. In your ~/.bashrc, ensure that ~/.local/bin is on PATH and that ~/.local is on PYTHONPATH.
  3. Create a file ~/.pydistutils.cfg which contains

    [install]
    prefix=~/.local
    

    It's a standard ConfigParser-format file.

  4. Download distribute_setup.py and run python distribute_setup.py (no sudo). If it complains about a non-existing site-packages directory, create it manually:

    mkdir -p ~/.local/lib/python2.7/site-packages/

  5. Run which easy_install to verify that it's coming from ~/.local/bin

  6. Run pip install virtualenv
  7. Run pip install virtualenvwrapper
  8. Create a virtual env containing folder, say ~/.virtualenvs
  9. In ~/.bashrc add

    export WORKON_HOME
    source ~/.local/bin/virtualenvwrapper.sh
    

That's it, no use of sudo at all and your Python environment is in ~/.local, completely separate from the OS's Python. Disclaimer: Not sure how compatible virtualenvwrapper is in this scenario - I couldn't test it on my system :-)


Python 3.4 onward

Python 3.3 adds the venv module, and Python 3.4 adds the ensurepip module. This makes bootstrapping pip as easy as:

python -m ensurepip

Perhaps preceded by a call to venv to do so inside a virtual environment.

Guaranteed pip is described in PEP 453.


  • You can do this without installing anything into python itself.

  • You don't need sudo or any privileges.

  • You don't need to find the latest version of a virtualenv tar file

  • You don't need to edit version info in a bash script to keep things up-to-date.

  • You don't need curl/wget or tar installed, nor pip or easy_install

  • this works for 2.7 as well as for 3.X

Save the following to /tmp/initvenv.py:

from future import print_function

import os, sys, shutil, tempfile, subprocess, tarfile, hashlib

try:
    from urllib2 import urlopen
except ImportError:
    from urllib.request import urlopen

tmp_dir = tempfile.mkdtemp(prefix='initvenv_')
try:
    # read the latest version from PyPI
    f = urlopen("https://pypi.python.org/pypi/virtualenv/")
    # retrieve the .tar.gz file
    tar_found = False
    url = None
    sha256 = None
    for line in f.read().splitlines():
        if isinstance(line, bytes):
            line = line.decode('utf-8')
        if tar_found:
            if 'sha256' in line:
                sha256 = line.split('data-clipboard-text')[1].split('"')[1]
                break
            continue
        if not tar_found and 'tar.gz">' not in line:
            continue
        tar_found = True
        for url in line.split('"'):
            if url.startswith('https'):
                break
    else:
        print('tar.gz not found')
        sys.exit(1)
    file_name = url.rsplit('/', 1)[1]
    print(file_name)
    os.chdir(tmp_dir)
    data = urlopen(url).read()
    data_sha256 = hashlib.sha256(data).hexdigest()
    if sha256 != data_sha256:
        print('sha256 not correct')
        print(sha256)
        print(data_sha256)
        sys.exit(1)
    with open(file_name, 'wb') as fp:
        fp.write(data)
    tar = tarfile.open(file_name)
    tar.extractall()
    tar.close()
    os.chdir(file_name.replace('.tar.gz', ''))
    print(subprocess.check_output([sys.executable, 'virtualenv.py'] +
                                  [sys.argv[1]]).decode('utf-8'), end='')
    if len(sys.argv) > 2:
        print(subprocess.check_output([
            os.path.join(sys.argv[1], 'bin', 'pip'), 'install', 'virtualenv'] +

            sys.argv[2:]).decode('utf-8'), end='')
except:
    raise
finally:
    shutil.rmtree(tmp_dir)  # always clean up

and use it as

python_binary_to_use_in_venv /tmp/initvenv.py your_venv_name [optional packages]

e.g. (if you really need the distribute compatibility layer for setuptools)

python /tmp/initvenv.py venv distribute

Please note that, with older python versions, this might give you InsecurePlatformWarnings¹.

Once you have your virtualenv (name e.g. venv) you can setup another virtualenv by using the virtualenv just installed:

venv/bin/virtualenv venv2

virtualenvwrapper

I recommend taking a look at virtualenvwrapper as well, after a one time setup:

% /opt/python/2.7.10/bin/python /tmp/initvenv.py venv virtualenvwrapper

and activation (can be done from your login script):

% source venv/bin/virtualenvwrapper.sh

you can do things like:

% mktmpenv 
New python executable in tmp-17bdc3054a46b2b/bin/python
Installing setuptools, pip, wheel...done.
This is a temporary environment. It will be deleted when you run 'deactivate'.
(tmp-17bdc3054a46b2b)% 

¹ I have not found a way to suppress the warning. It could be solved in pip and/or request, but the developers point to each other as the cause. I got the, often non-realistic, recommendation to upgrade the python version I was using to the latest version. I am sure this would break e.g my Linux Mint 17 install. Fortunately pip caches packages, so the Warning is made only once per package install.


There are good instructions on the Virtualenv official site. https://pypi.python.org/pypi/virtualenv

Basically what I did, is install pip with sudo easy_install pip, then used sudo pip install virtualenv then created an environment with: virtualenv my_env (name it what you want), following that I did: virtualenv --distribute my_env; which installed distribute and pip in my virtualenv.

Again, follow the instruction on the virtualenv page.

Kind of a hassle, coming from Ruby ;P


The good news is if you have installed python3.4, pyvenv is already been installed. So, Just

pyvenv project_dir
source project_dir/bin/activate
python --version   
python 3.4.*

Now in this virtual env, you can use pip to install modules for this project.

Leave this virtual env , just

deactivate

If you follow the steps advised in several tutorials I linked in this answer, you can get the desired effect without the somewhat complicated "manual" steps in Walker's and Vinay's answers. If you're on Ubuntu:

sudo apt-get install python-pip python-dev

The equivalent is achieved in OS X by using homebrew to install python (more details here).

brew install python

With pip installed, you can use it to get the remaining packages (you can omit sudo in OS X, as you're using your local python installation).

sudo pip install virtualenvwrapper

(these are the only packages you need installed globally and I doubt that it will clash with anything system-level from the OS. If you want to be super-safe, you can keep the distro's versions sudo apt-get install virtualenvwrapper)

Note: in Ubuntu 14.04 I receive some errors with pip install, so I use pip3 install virtualenv virtualenvwrapper and add VIRTUALENVWRAPPER_PYTHON=/usr/bin/python3 to my .bashrc/.zshrc file.

You then append to your .bashrc file

export WORKON_HOME
source /usr/local/bin/virtualenvwrapper.sh

and source it

. ~/.bashrc

This is basically it. Now the only decision is whether you want to create a virtualenv to include system-level packages

mkvirtualenv --system-site-packages foo

where your existing system packages don't have to be reinstalled, they are symlinked to the system interpreter's versions. Note: you can still install new packages and upgrade existing included-from-system packages without sudo - I tested it and it works without any disruptions of the system interpreter.

kermit@hocus-pocus:~$ sudo apt-get install python-pandas
kermit@hocus-pocus:~$ mkvirtualenv --system-site-packages s
(s)kermit@hocus-pocus:~$ pip install --upgrade pandas
(s)kermit@hocus-pocus:~$ python -c "import pandas; print(pandas.__version__)"
0.10.1
(s)kermit@hocus-pocus:~$ deactivate
kermit@hocus-pocus:~$ python -c "import pandas; print(pandas.__version__)"
0.8.0

The alternative, if you want a completely separated environment, is

mkvirtualenv --no-site-packages bar

or given that this is the default option, simply

mkvirtualenv bar

The result is that you have a new virtualenv where you can freely and sudolessly install your favourite packages

pip install flask

I've had various problems (see below) installing upgraded SSL modules, even inside a virtualenv, on top of older OS-provided Python versions, so I now use pyenv.

pyenv makes it very easy to install new Python versions and supports virtualenvs. Getting started is much easier than the recipes for virtualenv listed in other answers:

  • On Mac, type brew install pyenv and on Linux, use pyenv-installer
  • this gets you built-in virtualenv support as well as Python version switching (if required)
  • works well with Python 2 or 3, can have many versions installed at once

This works very well to insulate the "new Python" version and virtualenv from system Python. Because you can easily use a more recent Python (post 2.7.9), the SSL modules are already upgraded, and of course like any modern virtualenv setup you are insulated from the system Python modules.

A couple of nice tutorials:

The pyenv-virtualenv plugin is now built in - type pyenv commands | grep virtualenv to check. I wouldn't use the pyenv-virtualenvwrapper plugin to start with - see how you get on with pyenv-virtualenv which is more integrated into pyenv, as this covers most of what virtualenvwrapper does.

pyenv is modelled on rbenv (a good tool for Ruby version switching) and its only dependency is bash.

  • pyenv is unrelated to the very similarly named pyvenv - that is a virtualenv equivalent that's part of recent Python 3 versions, and doesn't handle Python version switching

Caveats

Two warnings about pyenv:

  1. It only works from a bash or similar shell - or more specifically, the pyenv-virtualenv plugin doesn't like dash, which is /bin/sh on Ubuntu or Debian.
  2. It must be run from an interactive login shell (e.g. bash --login using a terminal), which is not always easy to achieve with automation tools such as Ansible.

Hence pyenv is best for interactive use, and less good for scripting servers.

Older distributions - SSL module problems

One reason to use pyenv was that there were often problems with upgrading Python SSL modules when using older system-provided Python versions. This may be less of a problem now that current Linux distributions support Python 3.x.


Examples related to python

programming a servo thru a barometer Is there a way to view two blocks of code from the same file simultaneously in Sublime Text? python variable NameError Why my regexp for hyphenated words doesn't work? Comparing a variable with a string python not working when redirecting from bash script is it possible to add colors to python output? Get Public URL for File - Google Cloud Storage - App Engine (Python) Real time face detection OpenCV, Python xlrd.biffh.XLRDError: Excel xlsx file; not supported Could not load dynamic library 'cudart64_101.dll' on tensorflow CPU-only installation

Examples related to virtualenv

How to setup virtual environment for Python in VS Code? Conda version pip install -r requirements.txt --target ./lib What is the purpose of "pip install --user ..."? What is the difference between venv, pyvenv, pyenv, virtualenv, virtualenvwrapper, pipenv, etc? ImportError: No module named 'encodings' how to specify new environment location for conda create Use virtualenv with Python with Visual Studio Code in Ubuntu Is it ok having both Anacondas 2.7 and 3.5 installed in the same time? How to uninstall a package installed with pip install --user Unable to install boto3

Examples related to setuptools

Why can I not create a wheel in python? Python setup.py develop vs install Reference requirements.txt for the install_requires kwarg in setuptools setup.py file No module named pkg_resources How to install pip with Python 3? What is the official "preferred" way to install pip and virtualenv systemwide? What's the proper way to install pip, virtualenv, and distribute for Python? Installing SetupTools on 64-bit Windows Why use pip over easy_install? How do I remove packages installed with Python's easy_install?

Examples related to distribute

What's the proper way to install pip, virtualenv, and distribute for Python?