How do I automatically install missing python modules?

Question

I would like to be able to write:

try:
    import foo
except ImportError:
    install_the_module("foo")

What is the recommended/idiomatic way to handle this scenario?

I've seen a lot of scripts simply print an error or warning notifying the user about the missing module and (sometimes) providing instructions on how to install. However, if I know the module is available on PyPI, then I could surely take this a step further an initiate the installation process. No?

Answer 1

Risking negative votes, I would like to suggest a quick hack. Please note that I'm completely on board with accepted answer that dependencies should be managed externally.

But for situations where you absolutely need to hack something that acts like self contained, you can try something like below:

import os

try:
  import requests
except ImportError:
  print "Trying to Install required module: requests\n"
  os.system('python -m pip install requests')
# -- above lines try to install requests module if not present
# -- if all went well, import required module again ( for global access)
import requests

Answer 2

Installation issues are not subject of the source code!

You define your dependencies properly inside the setup.py of your package using the install_requires configuration.

That's the way to go...installing something as a result of an ImportError is kind of weird and scary. Don't do it.

Answer 3

try:
    import foo
except ImportError:
    sys.exit("""You need foo!
                install it from http://pypi.python.org/pypi/foo
                or run pip install foo.""")

Don't touch user's installation.

Answer 4

Here's the solution I put together which I call pyInstall.py. It actually checks whether the module is installed rather than relying on ImportError (it just looks cleaner, in my opinion, to handle this with an if rather than a try/except).

I've used it under version 2.6 and 2.7... it would probably work in older versions if I didn't want to handle print as a function... and I think it'll work in version 3.0+ but I've never tried it.

Also, as I note in the comments of my getPip function, I don't think that particular function will work under OS X.

from __future__ import print_function
from subprocess import call

def installPip(log=print):
    """
    Pip is the standard package manager for Python. Starting with Python 3.4
    it's included in the default installation, but older versions may need to
    download and install it. This code should pretty cleanly do just that.
    """
    log("Installing pip, the standard Python Package Manager, first")
    from os     import remove
    from urllib import urlretrieve
    urlretrieve("https://bootstrap.pypa.io/get-pip.py", "get-pip.py")
    call(["python", "get-pip.py"])

    # Clean up now...
    remove("get-pip.py")

def getPip(log=print):
    """
    Pip is the standard package manager for Python.
    This returns the path to the pip executable, installing it if necessary.
    """
    from os.path import isfile, join
    from sys     import prefix
    # Generate the path to where pip is or will be installed... this has been
    # tested and works on Windows, but will likely need tweaking for other OS's.
    # On OS X, I seem to have pip at /usr/local/bin/pip?
    pipPath = join(prefix, 'Scripts', 'pip.exe')

    # Check if pip is installed, and install it if it isn't.
    if not isfile(pipPath):
        installPip(log)
        if not isfile(pipPath):
            raise("Failed to find or install pip!")
    return pipPath

def installIfNeeded(moduleName, nameOnPip=None, notes="", log=print):
    """ Installs a Python library using pip, if it isn't already installed. """
    from pkgutil import iter_modules

    # Check if the module is installed
    if moduleName not in [tuple_[1] for tuple_ in iter_modules()]:
        log("Installing " + moduleName + notes + " Library for Python")
        call([getPip(log), "install", nameOnPip if nameOnPip else moduleName])

Here are some usage examples:

from datetime  import datetime
from pyInstall import installIfNeeded

# I like to have my messages timestamped so I can get an idea of how long they take.
def log(message):
    print(datetime.now().strftime("%a %b %d %H:%M:%S") + " - " + str(message))

# The name fabric doesn't really convey to the end user why the module is needed,
# so I include a very quick note that it's used for SSH.
installIfNeeded("fabric", notes = " (ssh)", log = log)

# SoftLayer is actually named softlayer on pip.
installIfNeeded("SoftLayer", "softlayer", log = log)

Edit: A more cross-platform way of getting pipPath is:

from subprocess import Popen, PIPE
finder = Popen(['where' if isWindows() else 'which', 'pip'], stdout = PIPE, stderr = PIPE)
pipPath = finder.communicate()[0].strip()

This makes the assumption that pip is/will be installed on the system path. It tends to be pretty reliable on non-Windows platforms, but on Windows it may be better to use the code in my original answer.