Many, pure python, packages are not yet available on the Python Package Index as wheels but still install fine. However, some packages on Windows give the dreaded vcvarsall.bat not found error.
The problem is that the package that you are trying to install contains a C or C++ extension and is not currently available as a pre-built wheel from the python package index, pypi, and on windows you do not have the tool chain needed to build such items.
The simplest answer is to go to Christoph Gohlke's excellent site and locate the appropriate version of the libraries that you need. By appropriate in the package name a -cpNN- has to match your version of python, i.e. if you are using windows 32 bit python even on win64 the name must include -win32- and if using the 64 bit python it must include -win_amd64- and then the python version must match, i.e. for Python 34 the filename must include -cp34-, etc. this is basically the magic that pip does for you on the pypi site.
Alternatively, you need to get the appropriate windows development kit for the version of python that you are using, the headers for any library that the package you are trying to build interfaces to, possibly the python headers for the version of python, etc.
Python 2.7 used Visual Studio 2008, Python 3.3 and 3.4 used Visual Studio 2010, and Python 3.5+ uses Visual Studio 2015.
Then you may need to locate the header files, at the matching revision for any libraries that your desired package links to and download those to an appropriate locations.
Finally you can let pip do your build - of course if the package has dependencies that you don't yet have you may also need to find the header files for them as well.
Alternatives: It is also worth looking out, both on pypi or Christop's site, for any slightly earlier version of the package that you are looking for that is either pure python or pre-built for your platform and python version and possibly using those, if found, until your package does become available. Likewise if you are using the very latest version of python you may find that it takes the package maintainers a little time to catch up so for projects that really need a specific package you may have to use a slightly older python for the moment. You can also check the packages source site to see if there is a forked version that is available pre-built or as pure python and searching for alternative packages that provide the functionality that you require but are available - one example that springs to mind is the Pillow, actively maintained, drop in replacement for PIL currently not updated in 6 years and not available for python 3.
Afterword, I would encourage anybody who is having this problem to go to the bug tracker for the package and add to, or raise if there isn't one already, a ticket politely requesting that the package maintainers provide a wheel on pypi for your specific combination of platform and python, if this is done then normally things will get better with time, some package maintainers don't realise that they have missed a given combination that people may be using.
Pip follows the rules of Semantic Versioning and by default prefers released packages over pre-releases. So if a given package has been released as
V0.98 and there is also a release candidate
V1.0-rc1 the default behaviour of
pip install will be to install
V0.98 - if you wish to install the release candidate, you are advised to test in a virtual environment first, you can enable do so with
--pip install --pre package-name or
--pip install --pre --upgrade package-name. In many cases pre-releases or release candidates may not have wheels built for all platform & version combinations so you are more likely to encounter the issues above.
You can also use pip to install development versions of packages from github and other locations, since such code is in flux it is very unlikely to have wheels built for it, so any impure packages will require the presence of the build tools, and they may be broken at any time so the user is strongly encouraged to only install such packages in a virtual environment.
Three options exist for such installations:
pip installpath/to/downloaded/file note that for most compression formats pip will handle unpacking to a cache area, etc.
pip installURL/of/package/repository - you may also need to use the
--proxyflags for this to work correctly, especially in a corporate environment. e.g:
> py -3.5-32 -m venv demo-pip > demo-pip\Scripts\activate.bat > python -m pip install -U pip Collecting pip Using cached pip-9.0.1-py2.py3-none-any.whl Installing collected packages: pip Found existing installation: pip 8.1.1 Uninstalling pip-8.1.1: Successfully uninstalled pip-8.1.1 Successfully installed pip-9.0.1 > pip install git+https://github.com/sphinx-doc/sphinx/ Collecting git+https://github.com/sphinx-doc/sphinx/ Cloning https://github.com/sphinx-doc/sphinx/ to c:\users\steve-~1\appdata\local\temp\pip-04yn9hpp-build Collecting six>=1.5 (from Sphinx==1.7.dev20170506) Using cached six-1.10.0-py2.py3-none-any.whl Collecting Jinja2>=2.3 (from Sphinx==1.7.dev20170506) Using cached Jinja2-2.9.6-py2.py3-none-any.whl Collecting Pygments>=2.0 (from Sphinx==1.7.dev20170506) Using cached Pygments-2.2.0-py2.py3-none-any.whl Collecting docutils>=0.11 (from Sphinx==1.7.dev20170506) Using cached docutils-0.13.1-py3-none-any.whl Collecting snowballstemmer>=1.1 (from Sphinx==1.7.dev20170506) Using cached snowballstemmer-1.2.1-py2.py3-none-any.whl Collecting babel!=2.0,>=1.3 (from Sphinx==1.7.dev20170506) Using cached Babel-2.4.0-py2.py3-none-any.whl Collecting alabaster<0.8,>=0.7 (from Sphinx==1.7.dev20170506) Using cached alabaster-0.7.10-py2.py3-none-any.whl Collecting imagesize (from Sphinx==1.7.dev20170506) Using cached imagesize-0.7.1-py2.py3-none-any.whl Collecting requests>=2.0.0 (from Sphinx==1.7.dev20170506) Using cached requests-2.13.0-py2.py3-none-any.whl Collecting typing (from Sphinx==1.7.dev20170506) Using cached typing-3.6.1.tar.gz Requirement already satisfied: setuptools in f:\toolbuild\temp\demo-pip\lib\site-packages (from Sphinx==1.7.dev20170506) Collecting sphinxcontrib-websupport (from Sphinx==1.7.dev20170506) Downloading sphinxcontrib_websupport-1.0.0-py2.py3-none-any.whl Collecting colorama>=0.3.5 (from Sphinx==1.7.dev20170506) Using cached colorama-0.3.9-py2.py3-none-any.whl Collecting MarkupSafe>=0.23 (from Jinja2>=2.3->Sphinx==1.7.dev20170506) Using cached MarkupSafe-1.0.tar.gz Collecting pytz>=0a (from babel!=2.0,>=1.3->Sphinx==1.7.dev20170506) Using cached pytz-2017.2-py2.py3-none-any.whl Collecting sqlalchemy>=0.9 (from sphinxcontrib-websupport->Sphinx==1.7.dev20170506) Downloading SQLAlchemy-1.1.9.tar.gz (5.2MB) 100% |################################| 5.2MB 220kB/s Collecting whoosh>=2.0 (from sphinxcontrib-websupport->Sphinx==1.7.dev20170506) Downloading Whoosh-2.7.4-py2.py3-none-any.whl (468kB) 100% |################################| 471kB 1.1MB/s Installing collected packages: six, MarkupSafe, Jinja2, Pygments, docutils, snowballstemmer, pytz, babel, alabaster, imagesize, requests, typing, sqlalchemy, whoosh, sphinxcontrib-websupport, colorama, Sphinx Running setup.py install for MarkupSafe ... done Running setup.py install for typing ... done Running setup.py install for sqlalchemy ... done Running setup.py install for Sphinx ... done Successfully installed Jinja2-2.9.6 MarkupSafe-1.0 Pygments-2.2.0 Sphinx-1.7.dev20170506 alabaster-0.7.10 babel-2.4.0 colorama-0.3.9 docutils-0.13.1 imagesize-0.7.1 pytz-2017.2 requests-2.13.0 six-1.10.0 snowballstemmer-1.2.1 sphinxcontrib-websupport-1.0.0 sqlalchemy-1.1.9 typing-3.6.1 whoosh-2.7.4
git+ prefix to the URL.
mercurialor other acceptable tool, preferably a DVCS tool, and use
pip installpath/to/cloned/repo - this will both process any requires.text file and perform the build and setup steps, you can manually change directory to your cloned repository and run
pip install -r requires.txtand then
python setup.py installto get the same effect. The big advantages of this approach is that while the initial clone operation may take longer than the snapshot download you can update to the latest with, in the case of git:
git pull origin masterand if the current version contains errors you can use
pip uninstallpackage-name then use
git checkoutcommands to move back through the repository history to earlier version(s) and re-try.