pip installs packages. Python packages. An easy_install replacement
pip is a tool for installing and managing Python packages, such as those found in the Python Package Index.
pip is a replacement for easy_install. It mostly uses the same techniques for finding packages, so packages that are easy_installable should be pip-installable as well. This means that you can use pip install SomePackage instead of easy_install SomePackage.
In order to use pip, you must first install setuptools or distribute. If you use virtualenv, a copy of pip will be automatically be installed in each virtual environment you create.
The main website for pip is www.pip-installer.org. You can also install the in-development version of pip with easy_install pip==dev.
Once you have pip, you can use it like this:
$ pip install SomePackage
SomePackage is some package you’ll find on PyPI. This installs the package and all its dependencies.
pip does other stuff too, with packages, but install is the biggest one. You can pip uninstall too.
You can also install from a URL (that points to a tar or zip file), install from some version control system (use URLs like hg+http://domain/repo – or prefix git+, svn+ etc). pip knows a bunch of stuff about revisions and stuff, so if you need to do things like install a very specific revision from a repository pip can do that too.
If you’ve ever used python setup.py develop, you can do something like that with pip install -e ./ – this works with packages that use distutils too (usually this only works with Setuptools projects).
You can use pip install --upgrade SomePackage to upgrade to a newer version, or pip install SomePackage==1.0.4 to install a very specific version.
Pip Compared To easy_install
pip is meant to improve on easy_install. Some of the improvements:
All packages are downloaded before installation. Partially-completed installation doesn’t occur as a result.
Care is taken to present useful output on the console.
The reasons for actions are kept track of. For instance, if a package is being installed, pip keeps track of why that package was required.
Error messages should be useful.
The code is relatively concise and cohesive, making it easier to use programmatically.
Packages don’t have to be installed as egg archives, they can be installed flat (while keeping the egg metadata).
Native support for other version control systems (Git, Mercurial and Bazaar)
Uninstallation of packages.
Simple to define fixed sets of requirements and reliably reproduce a set of packages.
pip doesn’t do everything that easy_install does. Specifically:
It cannot install from eggs. It only installs from source. (In the future it would be good if it could install binaries from Windows .exe or .msi – binary install on other platforms is not a priority.)
It doesn’t understand Setuptools extras (like package[test]). This should be added eventually.
It is incompatible with some packages that extensively customize distutils or setuptools in their setup.py files.
pip is complementary with virtualenv, and it is encouraged that you use virtualenv to isolate your installation.
The homepage for pip is at pip-installer.org. Bugs can be filed in the pip issue tracker. Discussion happens on the virtualenv email group.
pip is able to uninstall most installed packages with pip uninstall package-name.
Known exceptions include pure-distutils packages installed with python setup.py install (such packages leave behind no metadata allowing determination of what files were installed), and script wrappers installed by develop-installs (python setup.py develop).
pip also performs an automatic uninstall of an old version of a package before upgrading to a newer version, so outdated files (and egg-info data) from conflicting versions aren’t left hanging around to cause trouble. The old version of the package is automatically restored if the new version fails to download or install.
When installing software, and Python packages in particular, it’s common that you get a lot of libraries installed. You just did easy_install MyPackage and you get a dozen packages. Each of these packages has its own version.
Maybe you ran that installation and it works. Great! Will it keep working? Did you have to provide special options to get it to find everything? Did you have to install a bunch of other optional pieces? Most of all, will you be able to do it again? Requirements files give you a way to create an environment: a set of packages that work together.
If you’ve ever tried to setup an application on a new system, or with slightly updated pieces, and had it fail, pip requirements are for you. If you haven’t had this problem then you will eventually, so pip requirements are for you too – requirements make explicit, repeatable installation of packages.
So what are requirements files? They are very simple: lists of packages to install. Instead of running something like pip MyApp and getting whatever libraries come along, you can create a requirements file something like:
MyApp Framework==0.9.4 Library>=0.2
Then, regardless of what MyApp lists in setup.py, you’ll get a specific version of Framework (0.9.4) and at least the 0.2 version of Library. (You might think you could list these specific versions in MyApp’s setup.py – but if you do that you’ll have to edit MyApp if you want to try a new version of Framework, or release a new version of MyApp if you determine that Library 0.3 doesn’t work with your application.) You can also add optional libraries and support tools that MyApp doesn’t strictly require, giving people a set of recommended libraries.
You can also include “editable” packages – packages that are checked out from Subversion, Git, Mercurial and Bazaar. These are just like using the -e option to pip. They look like:
You have to start the URL with svn+ (git+, hg+ or bzr+), and you have to include #egg=Package so pip knows what to expect at that URL. You can also include @rev in the URL, e.g., @275 to check out revision 275.
Requirement files are mostly flat. Maybe MyApp requires Framework, and Framework requires Library. I encourage you to still list all these in a single requirement file; it is the nature of Python programs that there are implicit bindings directly between MyApp and Library. For instance, Framework might expose one of Library’s objects, and so if Library is updated it might directly break MyApp. If that happens you can update the requirements file to force an earlier version of Library, and you can do that without having to re-release MyApp at all.
Read the requirements file format to learn about other features.
So you have a working set of packages, and you want to be able to install them elsewhere. Requirements files let you install exact versions, but it won’t tell you what all the exact versions are.
To create a new requirements file from a known working environment, use:
$ pip freeze > stable-req.txt
This will write a listing of all installed libraries to stable-req.txt with exact versions for every library. You may want to edit the file down after generating (e.g., to eliminate unnecessary libraries), but it’ll give you a stable starting point for constructing your requirements file.
You can also give it an existing requirements file, and it will use that as a sort of template for the new file. So if you do:
$ pip freeze -r devel-req.txt > stable-req.txt
it will keep the packages listed in devel-req.txt in order and preserve comments.
Another way to distribute a set of libraries is a bundle format (specific to pip). This format is not stable at this time (there simply hasn’t been any feedback, nor a great deal of thought). A bundle file contains all the source for your package, and you can have pip install them all together. Once you have the bundle file further network access won’t be necessary. To build a bundle file, do:
$ pip bundle MyApp.pybundle MyApp
(Using a requirements file would be wise.) Then someone else can get the file MyApp.pybundle and run:
$ pip install MyApp.pybundle
This is not a binary format. This only packages source. If you have binary packages, then the person who installs the files will have to have a compiler, any necessary headers installed, etc. Binary packages are hard, this is relatively easy.
Using pip with virtualenv
pip is most nutritious when used with virtualenv. One of the reasons pip doesn’t install “multi-version” eggs is that virtualenv removes much of the need for it. Because pip is installed by virtualenv, just use path/to/my/environment/bin/pip to install things into that specific environment.
To tell pip to only run if there is a virtualenv currently activated, and to bail if not, use:
To tell pip to automatically use the currently active virtualenv:
Providing an environment with -E will be ignored.
Using pip with virtualenvwrapper
If you are using virtualenvwrapper, you might want pip to automatically create its virtualenvs in your $WORKON_HOME.
You can tell pip to do so by defining PIP_VIRTUALENV_BASE in your environment and setting it to the same value as that of $WORKON_HOME.
Do so by adding the line:
in your .bashrc under the line starting with export WORKON_HOME.
Using pip with buildout
If you are using zc.buildout you should look at gp.recipe.pip as an option to use pip and virtualenv in your buildouts.
Command line completion
pip comes with support for command line completion in bash and zsh and allows you tab complete commands and options. To enable it you simply need copy the required shell script to the your shell startup file (e.g. .profile or .zprofile) by running the special completion command, e.g. for bash:
$ pip completion --bash >> ~/.profile
And for zsh:
$ pip completion --zsh >> ~/.zprofile
Alternatively, you can use the result of the completion command directly with the eval function of you shell, e.g. by adding:
eval "`pip completion --bash`"
to your startup file.
Searching for packages
pip can search the Python Package Index (PyPI) for packages using the pip search command. To search, run:
$ pip search "query"
The query will be used to search the names and summaries of all packages indexed.
pip searches http://pypi.python.org/pypi by default but alternative indexes can be searched by using the --index flag.
The PyPI mirroring infrastructure as described in PEP 381 can be used by passing the --use-mirrors option to the install command. Alternatively, you can use the other ways to configure pip, e.g.:
$ export PIP_USE_MIRRORS=true
If enabled, pip will automatically query the DNS entry of the mirror index URL to find the list of mirrors to use. In case you want to override this list, please use the --mirrors option of the install command, or add to your pip configuration file:
[install] use-mirrors = true mirrors = http://d.pypi.python.org http://b.pypi.python.org
News / Changelog
Next release (1.1) schedule
Beta release mid-July 2011, final release early August.
Fixed docs issues.
Fixed issue #295 - Reinstall a package when using the install -I option
Fixed issue #283 - Finds a Git tag pointing to same commit as origin/master
Fixed issue #279 - Use absolute path for path to docs in setup.py
Fixed issue #320 - Correctly handle exceptions on Python3.
Fixed issue #314 - Correctly parse --editable lines in requirements files
Start to use git-flow.
Fixed issue #274 - find_command should not raise AttributeError
Fixed issue #273 - respect Content-Disposition header. Thanks Bradley Ayers.
Fixed issue #233 - pathext handling on Windows.
Fixed issue #252 - svn+svn protocol.
Fixed issue #44 - multiple CLI searches.
Fixed issue #266 - current working directory when running setup.py clean.
Added Python 3 support! Huge thanks to Vinay Sajip, Vitaly Babiy, Kelsey Hightower, and Alex Gronholm, among others.
Download progress only shown on a real TTY. Thanks Alex Morega.
Fixed finding of VCS binaries to not be fooled by same-named directories. Thanks Alex Morega.
Fixed uninstall of packages from system Python for users of Debian/Ubuntu python-setuptools package (workaround until fixed in Debian and Ubuntu).
Added get-pip.py installer. Simply download and execute it, using the Python interpreter of your choice:
$ curl -O https://raw.github.com/pypa/pip/master/contrib/get-pip.py $ python get-pip.py
This may have to be run as root.
Moved main repository to Github: https://github.com/pypa/pip
Transferred primary maintenance from Ian to Jannis Leidel, Carl Meyer, Brian Rosner
Fixed issue #14 - No uninstall-on-upgrade with URL package. Thanks Oliver Tonnhofer
Fixed issue #163 - Egg name not properly resolved. Thanks Igor Sobreira
Fixed issue #178 - Non-alphabetical installation of requirements. Thanks Igor Sobreira
Fixed issue #199 - Documentation mentions –index instead of –index-url. Thanks Kelsey Hightower
Fixed issue #204 - rmtree undefined in mercurial.py. Thanks Kelsey Hightower
Fixed bug in Git vcs backend that would break during reinstallation.
Fixed bug in Mercurial vcs backend related to pip freeze and branch/tag resolution.
Fixed bug in version string parsing related to the suffix “-dev”.
Avoid redundant unpacking of bundles (from pwaller)
Fixed issue #32, #150, #161 - Fixed checking out the correct tag/branch/commit when updating an editable Git requirement.
Fixed issue #49 - Added ability to install version control requirements without making them editable, e.g.:
pip install git+https://github.com/pypa/pip/
Fixed issue #175 - Correctly locate build and source directory on Mac OS X.
Added git+https:// scheme to Git VCS backend.
Added global –user flag as shortcut for –install-option=”–user”. From Ronny Pfannschmidt.
Added support for PyPI mirrors as defined in PEP 381, from Jannis Leidel.
Fixed issue #138 - Git revisions ignored. Thanks John-Scott Atlakson.
Fixed issue #95 - Initial editable install of github package from a tag fails. Thanks John-Scott Atlakson.
Fixed issue #107 - Can’t install if a directory in cwd has the same name as the package you’re installing.
Fixed issue #39 - –install-option=”–prefix=~/.local” ignored with -e. Thanks Ronny Pfannschmidt and Wil Tan.
Track which build/ directories pip creates, never remove directories it doesn’t create. From Hugo Lopes Tavares.
Pip now accepts file:// index URLs. Thanks Dave Abrahams.
Various cleanup to make test-running more consistent and less fragile. Thanks Dave Abrahams.
Real Windows support (with passing tests). Thanks Dave Abrahams.
pip-2.7 etc. scripts are created (Python-version specific scripts)
contrib/build-standalone script creates a runnable .zip form of pip, from Jannis Leidel
Editable git repos are updated when reinstalled
Fix problem with --editable when multiple .egg-info/ directories are found.
A number of VCS-related fixes for pip freeze, from Hugo Lopes Tavares.
Significant test framework changes, from Hugo Lopes Tavares.
Set zip_safe=False to avoid problems some people are encountering where pip is installed as a zip file.
Fixed opening of logfile with no directory name. Thanks Alexandre Conrad.
Temporary files are consistently cleaned up, especially after installing bundles, also from Alex Conrad.
Tests now require at least ScriptTest 1.0.3.
Fixed uninstallation on Windows
Added pip search command.
Tab-complete names of installed distributions for pip uninstall.
Support tab-completion when there is a global-option before the subcommand.
Install header files in standard (scheme-default) location when installing outside a virtualenv. Install them to a slightly more consistent non-standard location inside a virtualenv (since the standard location is a non-writable symlink to the global location).
pip now logs to a central location by default (instead of creating pip-log.txt all over the place) and constantly overwrites the file in question. On Unix and Mac OS X this is '$HOME/.pip/pip.log' and on Windows it’s '%HOME%\\pip\\pip.log'. You are still able to override this location with the $PIP_LOG_FILE environment variable. For a complete (appended) logfile use the separate '--log' command line option.
Fixed an issue with Git that left an editable packge as a checkout of a remote branch, even if the default behaviour would have been fine, too.
Fixed installing from a Git tag with older versions of Git.
Expand “~” in logfile and download cache paths.
Speed up installing from Mercurial repositories by cloning without updating the working copy multiple times.
Fixed installing directly from directories (e.g. pip install path/to/dir/).
Fixed installing editable packages with svn+ssh URLs.
Don’t print unwanted debug information when running the freeze command.
Create log file directory automatically. Thanks Alexandre Conrad.
Make test suite easier to run successfully. Thanks Dave Abrahams.
Fixed “pip install .” and “pip install ..”; better error for directory without setup.py. Thanks Alexandre Conrad.
Support Debian/Ubuntu “dist-packages” in zip command. Thanks duckx.
Fix relative –src folder. Thanks Simon Cross.
Handle missing VCS with an error message. Thanks Alexandre Conrad.
Added –no-download option to install; pairs with –no-install to separate download and installation into two steps. Thanks Simon Cross.
Fix uninstalling from requirements file containing -f, -i, or –extra-index-url.
Leftover build directories are now removed. Thanks Alexandre Conrad.
Fixed import error on Windows with regard to the backwards compatibility package
Fixed uninstall when /tmp is on a different filesystem.
Fixed uninstallation of distributions with namespace packages.
Added support for the https and http-static schemes to the Mercurial and ftp scheme to the Bazaar backend.
Fixed uninstallation of scripts installed with easy_install.
Fixed an issue in the package finder that could result in an infinite loop while looking for links.
Fixed issue with pip bundle and local files (which weren’t being copied into the bundle), from Whit Morriss.
Add pip uninstall and uninstall-before upgrade (from Carl Meyer).
Extended configurability with config files and environment variables.
Allow packages to be upgraded, e.g., pip install Package==0.1 then pip install Package==0.2.
Allow installing/upgrading to Package==dev (fix “Source version does not match target version” errors).
Added command and option completion for bash and zsh.
Extended integration with virtualenv by providing an option to automatically use an active virtualenv and an option to warn if no active virtualenv is found.
Fixed a bug with pip install –download and editable packages, where directories were being set with 0000 permissions, now defaults to 755.
Fixed uninstallation of easy_installed console_scripts.
Fixed uninstallation on Mac OS X Framework layout installs
Fixed bug preventing uninstall of editables with source outside venv.
Creates download cache directory if not existing.
Fixed a couple little bugs, with git and with extensions.
Added ability to override the default log file name (pip-log.txt) with the environmental variable $PIP_LOG_FILE.
Made the freeze command print installed packages to stdout instead of writing them to a file. Use simple redirection (e.g. pip freeze > stable-req.txt) to get a file with requirements.
Fixed problem with freezing editable packages from a Git repository.
Added support for base URLs using <base href='...'> when parsing HTML pages.
Fixed installing of non-editable packages from version control systems.
Fixed issue with Bazaar’s bzr+ssh scheme.
Added –download-dir option to the install command to retrieve package archives. If given an editable package it will create an archive of it.
Added ability to pass local file and directory paths to --find-links, e.g. --find-links=file:///path/to/my/private/archive
Reduced the amount of console log messages when fetching a page to find a distribution was problematic. The full messages can be found in pip-log.txt.
Added --no-deps option to install ignore package dependencies
Added --no-index option to ignore the package index (PyPI) temporarily
Fixed installing editable packages from Git branches.
Fixes freezing of editable packages from Mercurial repositories.
Fixed handling read-only attributes of build files, e.g. of Subversion and Bazaar on Windows.
When downloading a file from a redirect, use the redirected location’s extension to guess the compression (happens specifically when redirecting to a bitbucket.org tip.gz file).
Editable freeze URLs now always use revision hash/id rather than tip or branch names which could move.
Fixed comparison of repo URLs so incidental differences such as presence/absence of final slashes or quoted/unquoted special characters don’t trigger “ignore/switch/wipe/backup” choice.
Fixed handling of attempt to checkout editable install to a non-empty, non-repo directory.
Make -e work better with local hg repositories
Construct PyPI URLs the exact way easy_install constructs URLs (you might notice this if you use a custom index that is slash-sensitive).
Improvements on Windows (from Ionel Maries Cristian).
Fixed problem with not being able to install private git repositories.
Make pip zip zip all its arguments, not just the first.
Fix some filename issues on Windows.
Allow the -i and --extra-index-url options in requirements files.
Fix the way bundle components are unpacked and moved around, to make bundles work.
Adds -s option to allow the access to the global site-packages if a virtualenv is to be created.
Fixed support for Subversion 1.6.
Improved virtualenv restart and various path/cleanup problems on win32.
Fixed a regression with installing from svn repositories (when not using -e).
Fixes when installing editable packages that put their source in a subdirectory (like src/).
Improve pip -h
Added support for editable packages created from Git, Mercurial and Bazaar repositories and ability to freeze them. Refactored support for version control systems.
Do not use sys.exit() from inside the code, instead use a return. This will make it easier to invoke programmatically.
Put the install record in Package.egg-info/installed-files.txt (previously they went in site-packages/install-record-Package.txt).
Fix a problem with pip freeze not including -e svn+ when an svn structure is peculiar.
Allow pip -E to work with a virtualenv that uses a different version of Python than the parent environment.
Fixed Win32 virtualenv (-E) option.
Search the links passed in with -f for packages.
Detect zip files, even when the file doesn’t have a .zip extension and it is served with the wrong Content-Type.
Installing editable from existing source now works, like pip install -e some/path/ will install the package in some/path/. Most importantly, anything that package requires will also be installed by pip.
Add a --path option to pip un/zip, so you can avoid zipping files that are outside of where you expect.
Add --simulate option to pip zip.
Fixed small problem that prevented using pip.py without actually installing pip.
Fixed --upgrade, which would download and appear to install upgraded packages, but actually just reinstall the existing package.
Fixed Windows problem with putting the install record in the right place, and generating the pip script with Setuptools.
Download links that include embedded spaces or other unsafe characters (those characters get %-encoded).
Fixed use of URLs in requirement files, and problems with some blank lines.
Turn some tar file errors into warnings.
Renamed to pip, and to install you now do pip install PACKAGE
Added command pip zip PACKAGE and pip unzip PACKAGE. This is particularly intended for Google App Engine to manage libraries to stay under the 1000-file limit.
Some fixes to bundles, especially editable packages and when creating a bundle using unnamed packages (like just an svn repository without #egg=Package).
Added an option --install-option to pass options to pass arguments to setup.py install
.svn/ directories are no longer included in bundles, as these directories are specific to a version of svn – if you build a bundle on a system with svn 1.5, you can’t use the checkout on a system with svn 1.4. Instead a file svn-checkout.txt is included that notes the original location and revision, and the command you can use to turn it back into an svn checkout. (Probably unpacking the bundle should, maybe optionally, recreate this information – but that is not currently implemented, and it would require network access.)
Avoid ambiguities over project name case, where for instance MyPackage and mypackage would be considered different packages. This in particular caused problems on Macs, where MyPackage/ and mypackage/ are the same directory.
Added support for an environmental variable $PIP_DOWNLOAD_CACHE which will cache package downloads, so future installations won’t require large downloads. Network access is still required, but just some downloads will be avoided when using this.
Always use svn checkout (not export) so that tag_svn_revision settings give the revision of the package.
Don’t update checkouts that came from .pybundle files.
Improve error text when there are errors fetching HTML pages when seeking packages.
Improve bundles: include empty directories, make them work with editable packages.
If you use -E env and the environment env/ doesn’t exist, a new virtual environment will be created.
Fix dependency_links for finding packages.
Fixed a NameError exception when running pip outside of a virtualenv environment.
Added HTTP proxy support (from Prabhu Ramachandran)
Fixed use of hashlib.md5 on python2.5+ (also from Prabhu Ramachandran)
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.