setuptools |
UserPreferences |
The PEAK Developers' Center | FrontPage | RecentChanges | TitleIndex | WordIndex | SiteNavigation | HelpContents |
setuptools is a collection of enhancements to the Python distutils (for Python 2.3 and up) that allow you to more easily build and distribute Python packages, especially ones that have dependencies on other packages.
Packages built and distributed using setuptools look to the user like ordinary Python packages based on the distutils. Your users don't need to install or even know about setuptools in order to use them, and you don't have to include the entire setuptools package in your distributions. By including just a single bootstrap module (a 7K .py file), your package will automatically download and install setuptools if the user is building your package from source and doesn't have a suitable version already installed.
Feature Highlights:
If you are behind an NTLM-based firewall that prevents Python programs from accessing the net directly, you may wish to first install and use the APS proxy server, which lets you get past such firewalls in the same way that your web browser(s) do.
If you do not have write access to your computer's site-packages directory, please also see the EasyInstall documentation on Non-Root Installation for more detailed instructions on pre-configuring your system for the best usability with setuptools and EasyInstall, then return here for the remaining steps.
To install setuptools, first download ez_setup.py and run it; this will automatically download and install the appropriate egg for your Python version.
You may receive a message telling you about an obsolete version of setuptools being present; if so, you must be sure to delete it entirely, along with the old pkg_resources module if it's present on sys.path.
To get the in-development version of setuptools, run:
cvs -d:pserver:anonymous@cvs.sourceforge.net:/cvsroot/python login cvs -z3 -d:pserver:anonymous@cvs.sourceforge.net:/cvsroot/python \ co -d setuptools python/nondist/sandbox/setuptools
You can then install it using the usual "setup.py install" incantation.
(Note that setuptools must be installed as an egg directory; it will not operate correctly otherwise. If you are unable to install to a valid site-packages directory (e.g. a "non-root install" that doesn't conform to the Non-Root Installation procedure), you will therefore need to manually add the setuptools egg to your PYTHONPATH. You won't need to do this for every egg you install, because the pkg_resources module can automatically find eggs and add them to sys.path at runtime. It's just that the setuptools egg contains the pkg_resources runtime, and therefore has to be manually bootstrapped if you can't install it to a valid site-packages directory. However, if you are installing as root or you followed the Non-Root Installation procedure, you shouldn't have to worry about any of this.)
For basic use of setuptools, just import things from setuptools instead of the distutils. Here's a minimal setup script using setuptools:
from setuptools import setup, find_packages setup( name = "HelloWorld", version = "0.1", packages = find_packages(), )
As you can see, it doesn't take much to use setuptools in a project. Just by doing the above, this project will be able to produce eggs, upload to PyPI, and automatically include all packages in the directory where the setup.py lives. See the Command Reference section below to see what commands you can give to this setup script.
Of course, before you release your project to PyPI, you'll want to add a bit more information to your setup script to help people find or learn about your project. And maybe your project will have grown by then to include a few dependencies, and perhaps some data files and scripts:
from setuptools import setup, find_packages setup( name = "HelloWorld", version = "0.1", packages = find_packages(), scripts = ['say_hello.py'], # Project uses reStructuredText, so ensure that the docutils get # installed or upgraded on the target machine install_requires = ['docutils>=0.3'], package_data = { # If any package contains *.txt or *.rst files, include them: '': ['*.txt', '*.rst'], # And include any *.msg files found in the 'hello' package, too: 'hello': ['*.msg'], } # metadata for upload to PyPI author = "Me", author_email = "me@example.com", description = "This is an Example Package", license = "PSF", keywords = "hello world example examples", url = "http://example.com/HelloWorld/", # project home page, if any # could also include long_description, download_url, classifiers, etc. )
In the sections that follow, we'll explain what most of these setup() arguments do (except for the metadata ones), and the various ways you might use them in your own project(s).
The following keyword arguments to setup() are added or changed by setuptools. All of them are optional; you do not have to supply them unless you need the associated setuptools feature.
A list of strings naming resources that should be extracted together, if any of them is needed, or if any C extensions included in the project are imported. This argument is only useful if the project will be installed as a zipfile, and there is a need to have all of the listed resources be extracted to the filesystem as a unit. Resources listed here should be '/'-separated paths, relative to the source root, so to list a resource foo.png in package bar.baz, you would include the string bar/baz/foo.png in this argument.
If you only need to obtain resources one at a time, or you don't have any C extensions that access other files in the project (such as data files or shared libraries), you probably do NOT need this argument and shouldn't mess with it. For more details on how this argument works, see the section below on Automatic Resource Extraction.
For simple projects, it's usually easy enough to manually add packages to the packages argument of setup(). However, for very large projects (Twisted, PEAK, Zope, Chandler, etc.), it can be a big burden to keep the package list updated. That's what setuptools.find_packages() is for.
find_packages() takes a source directory, and a list of package names or patterns to exclude. If omitted, the source directory defaults to the same directory as the setup script. Some projects use a src or lib directory as the root of their source tree, and those projects would of course use "src" or "lib" as the first argument to find_packages(). (And such projects also need something like package_dir = {'':'src'} in their setup() arguments, but that's just a normal distutils thing.)
Anyway, find_packages() walks the target directory, and finds Python packages by looking for __init__.py files. It then filters the list of packages using the exclusion patterns.
Exclusion patterns are package names, optionally including wildcards. For example, find_packages(exclude=["*.tests"]) will exclude all packages whose last name part is tests. Or, find_packages(exclude=["*.tests", "*.tests.*"]) will also exclude any subpackages of packages named tests, but it still won't exclude a top-level tests package or the children thereof. In fact, if you really want no tests packages at all, you'll need something like this:
find_packages(exclude=["*.tests", "*.tests.*", "tests.*", "tests"])
in order to cover all the bases. Really, the exclusion patterns are intended to cover simpler use cases than this, like excluding a single, specified package and its subpackages.
Regardless of the target directory or exclusions, the find_packages() function returns a list of package names suitable for use as the packages argument to setup(), and so is usually the easiest way to set that argument in your setup script. Especially since it frees you from having to remember to modify your setup script whenever your project grows additional top-level packages or subpackages.
setuptools supports automatically installing dependencies when a package is installed, and including information about dependencies in Python Eggs (so that package management tools like EasyInstall can use the information).
setuptools and pkg_resources use a common syntax for specifying a project's required dependencies. This syntax consists of a project's PyPI name, optionally followed by a comma-separated list of "extras" in square brackets, optionally followed by a comma-separated list of version specifiers. A version specifier is one of the operators <, >, <=, >=, == or !=, followed by a version identifier. Tokens may be separated by whitespace, but any whitespace or nonstandard characters within a project name or version identifier must be replaced with -.
Here are some example requirement specifiers:
docutils >= 0.3 # comment lines and \ continuations are allowed in requirement strings BazSpam ==1.1, ==1.2, ==1.3, ==1.4, ==1.5, \ ==1.6, ==1.7 # and so are line-end comments PEAK[FastCGI, reST]>=0.5a4 setuptools==0.5a7
The simplest way to include requirement specifiers is to use the install_requires argument to setup(). It takes a string or list of strings containing requirement specifiers. If you include more than one requirement in a string, each requirement must begin on a new line.
This has three effects:
Note, by the way, that if you declare your dependencies in setup.py, you do not need to use the require() function in your scripts or modules, as long as you either install the project or use setup.py develop to do development work on it. (See "Development Mode" below for more details on using setup.py develop.)
Sometimes a project has "recommended" dependencies, that are not required for all uses of the project. For example, a project might offer optional PDF output if ReportLab is installed, and reStructuredText support if docutils is installed. These optional features are called "extras", and setuptools allows you to define their requirements as well. In this way, other projects that require these optional features can force the additional requirements to be installed, by naming the desired extras in their install_requires.
For example, let's say that Project A offers optional PDF and reST support:
setup( name="Project-A", ... extras_require = { 'PDF': ["ReportLab>=1.2", "RXP"], 'reST': ["docutils>=0.3"], } )
And that project B needs project A, with PDF support:
setup( name="Project-B", install_requires = ["Project-A[PDF]"], ... )
This will cause ReportLab to be installed along with project A, if project B is installed -- even if project A was already installed. In this way, a project can encapsulate groups of optional "downstream dependencies" under a feature name, so that packages that depend on it don't have to know what the downstream dependencies are. If a later version of Project A builds in PDF support and no longer needs ReportLab, or if it ends up needing other dependencies besides ReportLab in order to provide PDF support, Project B's setup information does not need to change, but the right packages will still be installed if needed.
As you can see, the extras_require argument takes a dictionary mapping names of "extra" features, to strings or lists of strings describing those features' requirements. These requirements will not be automatically installed unless another package depends on them (directly or indirectly) by including the desired "extras" in square brackets after the associated project name. (Or if the extras were listed in a requirement spec on the EasyInstall command line.)
Note, by the way, that if a project ends up not needing any other packages to support a feature, it should keep an empty requirements list for that feature in its extras_require argument, so that packages depending on that feature don't break (due to an invalid feature name). For example, if Project A above builds in PDF support and no longer needs ReportLab, it should change its setup to this:
setup( name="Project-A", ... extras_require = { 'PDF': [], 'reST': ["docutils>=0.3"], } )
so that Package B doesn't have to remove the [PDF] from its requirement specifier.
The distutils have traditionally allowed installation of "data files", which are placed in a platform-specific location. However, the most common use case for data files distributed with a package is for use by the package, usually by including the data files in the package directory. Setuptools supports this by allowing a package_data argument to setup(), e.g.:
from setuptools import setup, find_packages setup( ... package_data = { # If any package contains *.txt or *.rst files, include them: '': ['*.txt', '*.rst'], # And include any *.msg files found in the 'hello' package, too: 'hello': ['*.msg'], } )
The package_data argument is a dictionary that maps from package names to lists of glob patterns. The globs may include subdirectory names, if the data files are contained in a subdirectory of the package. For example, if the package tree looks like this:
setup.py src/ mypkg/ __init__.py mypkg.txt data/ somefile.dat otherdata.dat
The setuptools setup file might look like this:
from setuptools import setup, find_packages setup( ... packages = find_packages('src'), # include all packages under src package_dir = {'':'src'}, # tell distutils packages are under src package_data = { # If any package contains *.txt files, include them: '': ['*.txt'], # And include any *.dat files found in the 'data' subdirectory # of the 'mypkg' package, also: 'mypkg': ['data/*.dat'], } )
Notice that if you list patterns in package_data under the empty string, these patterns are used to find files in every package, even ones that also have their own patterns listed. Thus, in the above example, the mypkg.txt file gets included even though it's not listed in the patterns for mypkg.
Also notice that if you use paths, you must use a forward slash (/) as the path separator, even if you are on Windows. Setuptools automatically converts slashes to appropriate platform-specific separators at build time.
(Note: although the package_data argument was previously only available in setuptools, it was also added to the Python distutils package as of Python 2.4; there is some documentation for the feature available on the python.org website.)
Typically, existing programs manipulate a package's __file__ attribute in order to find the location of data files. However, this manipulation isn't compatible with PEP 302-based import hooks, including importing from zip files and Python Eggs. It is strongly recommended that, if you are using data files, you should use the Resource Management API of pkg_resources to access them. The pkg_resources module is distributed as part of setuptools, so if you're using setuptools to distribute your package, there is no reason not to use its resource management API. See also Accessing Package Resources for a quick example of converting code that uses __file__ to use pkg_resources instead.
The distutils normally install general "data files" to a platform-specific location (e.g. /usr/share). This feature intended to be used for things like documentation, example configuration files, and the like. setuptools does not install these data files in a separate location, however. They are bundled inside the egg file or directory, alongside the Python modules and packages. The data files can also be accessed using the Resource Management API, by specifying a Requirement instead of a package name:
from pkg_resources import Requirement, resource_filename filename = resource_filename(Requirement.parse("MyProject"),"sample.conf")
The above code will obtain the filename of the "sample.conf" file in the data root of the "MyProject" distribution.
Note, by the way, that this encapsulation of data files means that you can't actually install data files to some arbitrary location on a user's machine; this is a feature, not a bug. You can always include a script in your distribution that extracts and copies your the documentation or data files to a user-specified location, at their discretion. If you put related data files in a single directory, you can use resource_filename() with the directory name to get a filesystem directory that then can be copied with the shutil module. (Even if your package is installed as a zipfile, calling resource_filename() on a directory will return an actual filesystem directory, whose contents will be that entire subtree of your distribution.)
(Of course, if you're writing a new package, you can just as easily place your data files or directories inside one of your packages, rather than using the distutils' approach. However, if you're updating an existing application, it may be simpler not to change the way it currently specifies these data files.)
If you are using tools that expect your resources to be "real" files, or your project includes non-extension native libraries or other files that your C extensions expect to be able to access, you may need to list those files in the eager_resources argument to setup(), so that the files will be extracted together, whenever a C extension in the project is imported. This is especially important if your project includes shared libraries other than distutils-built C extensions. Those shared libraries should be listed as eager_resources, because they need to be present in the filesystem when the C extensions that link to them are used.
The pkg_resources runtime for compressed packages will automatically extract all C extensions and eager_resources at the same time, whenever any C extension or eager resource is requested via the resource_filename() API. (C extensions are imported using resource_filename() internally.) This ensures that C extensions will see all of the "real" files that they expect to see.
Note also that you can list directory resource names in eager_resources as well, in which case the directory's contents (including subdirectories) will be extracted whenever any C extension or eager resource is requested.
Please note that if you're not sure whether you need to use this argument, you don't! It's really intended to support projects with lots of non-Python dependencies and as a last resort for crufty projects that can't otherwise handle being compressed. If your package is pure Python, Python plus data files, or Python plus C, you really don't need this. You've got to be using either C or an external program that needs "real" files in your project before there's any possibility of eager_resources being relevant to your project.
setuptools supports creating libraries that "plug in" to extensible applications and frameworks, by letting you register "entry points" in your project that can be imported by the application or framework.
For example, suppose that a blogging tool wants to support plugins that provide translation for various file types to the blog's output format. The framework might define an "entry point group" called blogtool.parsers, and then allow plugins to register entry points for the file extensions they support.
This would allow people to create distributions that contain one or more parsers for different file types, and then the blogging tool would be able to find the parsers at runtime by looking up an entry point for the file extension (or mime type, or however it wants to).
Note that if the blogging tool includes parsers for certain file formats, it can register these as entry points in its own setup script, which means it doesn't have to special-case its built-in formats. They can just be treated the same as any other plugin's entry points would be.
If you're creating a project that plugs in to an existing application or framework, you'll need to know what entry points or entry point groups are defined by that application or framework. Then, you can register entry points in your setup script. Here are a few examples of ways you might register an .rst file parser entry point in the blogtool.parsers entry point group, for our hypothetical blogging tool:
setup( # ... entry_points = {'blogtool.parsers': '.rst = some_module:SomeClass'} ) setup( # ... entry_points = {'blogtool.parsers': ['.rst = some_module:a_func']} ) setup( # ... entry_points = """ [blogtool.parsers] .rst = some.nested.module:SomeClass.some_classmethod [reST] """, extras_require = dict(reST = "Docutils>=0.3.5") )
The entry_points argument to setup() accepts either a string with .ini-style sections, or a dictionary mapping entry point group names to either strings or lists of strings containing entry point specifiers. An entry point specifier consists of a name and value, separated by an = sign. The value consists of a dotted module name, optionally followed by a : and a dotted identifier naming an object within the module. It can also include a bracketed list of "extras" that are required for the entry point to be used. When the invoking application or framework requests loading of an entry point, any requirements implied by the associated extras will be passed to pkg_resources.require(), so that an appropriate error message can be displayed if the needed package(s) are missing. (Of course, the invoking app or framework can ignore such errors if it wants to make an entry point optional if a requirement isn't installed.)
Some extensible applications and frameworks may need to define their own kinds of metadata to include in eggs, which they can then access using the pkg_resources metadata APIs. Ordinarily, this is done by having plugin developers include additional files in their ProjectName.egg-info directory. However, since it can be tedious to create such files by hand, you may want to create a distutils extension that will create the necessary files from arguments to setup(), in much the same way that setuptools does for many of the setup() arguments it adds. See the section below on Creating distutils Extensions for more details, especially the subsection on Adding new EGG-INFO Files.
Under normal circumstances, the distutils assume that you are going to build a distribution of your project, not use it in its "raw" or "unbuilt" form. If you were to use the distutils that way, you would have to rebuild and reinstall your project every time you made a change to it during development.
Another problem that sometimes comes up with the distutils is that you may need to do development on two related projects at the same time. You may need to put both projects' packages in the same directory to run them, but need to keep them separate for revision control purposes. How can you do this?
Setuptools allows you to deploy your projects for use in a common directory or staging area, but without copying any files. Thus, you can edit each project's code in its checkout directory, and only need to run build commands when you change a project's C extensions or similarly compiled files. You can even deploy a project into another project's checkout directory, if that's your preferred way of working (as opposed to using a common independent staging area or the site-packages directory).
To do this, use the setup.py develop command. It works very similarly to setup.py install or the EasyInstall tool, except that it doesn't actually install anything. Instead, it creates a special .egg-link file in the deployment directory, that links to your project's source code. And, if your deployment directory is Python's site-packages directory, it will also update the easy-install.pth file to include your project's source code, thereby making it available on sys.path for all programs using that Python installation.
In addition, the develop command creates wrapper scripts in the target script directory that will run your in-development scripts after ensuring that all your install_requires packages are available on sys.path.
You can deploy the same project to multiple staging areas, e.g. if you have multiple projects on the same machine that are sharing the same project you're doing development work.
When you're done with a given development task, you can remove the project source from a staging area using setup.py develop --uninstall, specifying the desired staging area if it's not the default.
There are several options to control the precise behavior of the develop command; see the section on the develop command below for more details.
Your users might not have setuptools installed on their machines, or even if they do, it might not be the right version. Fixing this is easy; just download ez_setup.py, and put it in the same directory as your setup.py script. (Be sure to add it to your revision control system, too.) Then add these two lines to the very top of your setup script, before the script imports anything from setuptools:
import ez_setup ez_setup.use_setuptools()
That's it. The ez_setup module will automatically download a matching version of setuptools from PyPI, if it isn't present on the target system. Whenever you install an updated version of setuptools, you should also update your projects' ez_setup.py files, so that a matching version gets installed on the target machine(s).
By the way, setuptools supports the new PyPI "upload" command, so you can use setup.py sdist upload or setup.py bdist_egg upload to upload your source or egg distributions respectively. Your project's current version must be registered with PyPI first, of course; you can use setup.py register to do that. Or you can do it all in one step, e.g. setup.py register sdist bdist_egg upload will register the package, build source and egg distributions, and then upload them both to PyPI, where they'll be easily found by other projects that depend on them.
If you're managing several projects that need to use ez_setup, and you are using Subversion as your revision control system, you can use the "svn:externals" property to share a single copy of ez_setup between projects, so that it will always be up-to-date whenever you check out or update an individual project, without having to manually update each project to use a new version.
However, because Subversion only supports using directories as externals, you have to turn ez_setup.py into ez_setup/__init__.py in order to do this, then create "externals" definitions that map the ez_setup directory into each project. Also, if any of your projects use find_packages() on their setup directory, you will need to exclude the resulting ez_setup package, to keep it from being included in your distributions, e.g.:
setup( ... packages = find_packages(exclude=['ez_setup']), )
Of course, the ez_setup package will still be included in your packages' source distributions, as it needs to be.
For your convenience, you may use the following external definition, which will track the latest version of setuptools:
ez_setup svn://svn.eby-sarna.com/svnroot/ez_setup
You can set this by executing this command in your project directory:
svn propedit svn:externals .
And then adding the line shown above to the file that comes up for editing.
For maximum performance, Python packages are best installed as zip files. Not all packages, however, are capable of running in compressed form, because they may expect to be able to access either source code or data files as normal operating system files. So, setuptools can install your project as a zipfile or a directory, and its default choice is determined by the project's zip_safe flag.
You can pass a True or False value for the zip_safe argument to the setup() function, or you can omit it. If you omit it, the bdist_egg command will analyze your project's contents to see if it can detect any conditions that would prevent it from working in a zipfile. It will output notices to the console about any such conditions that it finds.
Currently, this analysis is extremely conservative: it will consider the project unsafe if it contains any C extensions or datafiles whatsoever. This does not mean that the project can't or won't work as a zipfile! It just means that the bdist_egg authors aren't yet comfortable asserting that the project will work. If the project contains no C or data files, and does no __file__ or __path__ introspection or source code manipulation, then there is an extremely solid chance the project will work when installed as a zipfile. (And if the project uses pkg_resources for all its data file access, then C extensions and other data files shouldn't be a problem at all. See the Accessing Data Files at Runtime section above for more information.)
However, if bdist_egg can't be sure that your package will work, but you've checked over all the warnings it issued, and you are either satisfied it will work (or if you want to try it for yourself), then you should set zip_safe to True in your setup() call. If it turns out that it doesn't work, you can always change it to False, which will force setuptools to install your project as a directory rather than as a zipfile.
Of course, the end-user can still override either decision, if they are using EasyInstall to install your package. And, if you want to override for testing purposes, you can just run setup.py easy_install --zip-ok . or setup.py easy_install --always-unzip . in your project directory. to install the package as a zipfile or directory, respectively.
In the future, as we gain more experience with different packages and become more satisfied with the robustness of the pkg_resources runtime, the "zip safety" analysis may become less conservative. However, we strongly recommend that you determine for yourself whether your project functions correctly when installed as a zipfile, correct any problems if you can, and then make an explicit declaration of True or False for the zip_safe flag, so that it will not be necessary for bdist_egg or EasyInstall to try to guess whether your project can work as a zipfile.
Sometimes, a large package is more useful if distributed as a collection of smaller eggs. However, Python does not normally allow the contents of a package to be retrieved from more than one location. "Namespace packages" are a solution for this problem. When you declare a package to be a namespace package, it means that the package has no meaningful contents in its __init__.py, and that it is merely a container for modules and subpackages.
The pkg_resources runtime will automatically ensure that the contents of namespace packages that are spread over multiple eggs or directories are combined into a single virtual package.
The namespace_packages argument to setup() lets you declare your project's namespace packages, so that they will be included in your project's metadata. Then, the runtime will automatically detect this when it adds the distribution to sys.path, and ensure that the packages are properly merged.
The argument should list the namespace packages that the egg participates in. For example, the ZopeInterface project might do this:
setup( # ... namespace_packages = ['zope'] )
because it contains a zope.interface package that lives in the zope namespace package. Similarly, a project for a standalone zope.publisher would also declare the zope namespace package.
Namespace packages don't have to be top-level packages. For example, Zope 3's zope.app package is a namespace package, and in the future PEAK's peak.util package will be too.
Note, by the way, that your project's source tree must include the namespace packages' __init__.py files (and the __init__.py of any parent packages), in a normal Python package layout. These __init__.py files should not contain any code or data, because only one egg's __init__.py files will be used to construct the parent packages in memory at runtime, and there is no guarantee which egg will be used.
For example, if both zope.interface and zope.publisher have been installed from separate distributions, it is unspecified which of the two distributions' zope/__init__.py files will be used to create the zope package in memory. Therefore, it is better not to have any code or data in a namespace package's __init__ module, so as to prevent any complications.
(This is one reason the concept is called a "namespace package": it is a package that exists only to provide a namespace under which other modules or packages are gathered. In Java, for example, namespace packages are often used just to avoid naming collisions between different projects, using package names like org.apache as a namespace for packages that are part of apache.org projects.)
When a set of related projects are under development, it may be important to track finer-grained version increments than you would normally use for e.g. "stable" releases. While stable releases might be measured in dotted numbers with alpha/beta/etc. status codes, development versions of a project often need to be tracked by revision or build number or even build date. This is especially true when projects in development need to refer to one another, and therefore may literally need an up-to-the-minute version of something!
To support these scenarios, setuptools allows you to "tag" your source and egg distributions by adding one or more of the following to the project's "official" version identifier:
You can add these tags by adding egg_info and the desired options to the command line ahead of the sdist or bdist commands that you want to generate a daily build or snapshot for. See the section below on the egg_info command for more details.
Also, if you are creating builds frequently, and either building them in a downloadable location or are copying them to a distribution server, you should probably also check out the rotate command, which lets you automatically delete all but the N most-recently-modified distributions matching a glob pattern. So, you can use a command line like:
setup.py egg_info -rbDEV bdist_egg rotate -m.egg -k3
to build an egg whose version info includes 'DEV-rNNNN' (where NNNN is the most recent Subversion revision that affected the source tree), and then delete any egg files from the distribution directory except for the three that were built most recently.
If you have to manage automated builds for multiple packages, each with different tagging and rotation policies, you may also want to check out the alias command, which would let each package define an alias like daily that would perform the necessary tag, build, and rotate commands. Then, a simpler script or cron job could just run setup.py daily in each project directory. (And, you could also define sitewide or per-user default versions of the daily alias, so that projects that didn't define their own would use the appropriate defaults.)
setuptools enhances the distutils' default algorithm for source file selection, so that all files managed by CVS or Subversion in your project tree are included in any source distribution you build. This is a big improvement over having to manually write a MANIFEST.in file and try to keep it in sync with your project. So, if you are using CVS or Subversion, and your source distributions only need to include files that you're tracking in revision control, don't create a a MANIFEST.in file for your project. (And, if you already have one, you might consider deleting it the next time you would otherwise have to change it.)
Unlike the distutils, setuptools regenerates the source distribution MANIFEST file every time you build a source distribution, as long as you don't have a MANIFEST.in file. If you do have a MANIFEST.in (e.g. because you aren't using CVS or Subversion), then you'll have to follow the normal distutils procedures for managing what files get included in a source distribution, and setuptools' enhanced algorithms will not be used.
(Note, by the way, that if you're using some other revision control system, you might consider submitting a patch to the setuptools.command.sdist module so we can include support for it, too.)
If you use the register command (setup.py register) to register your package with PyPI, that's most of the battle right there. (See the docs for the register command for more details.)
If you also use the `upload`_ command to upload actual distributions of your package, that's even better, because EasyInstall will be able to find and download them directly from your project's PyPI page.
However, there may be reasons why you don't want to upload distributions to PyPI, and just want your existing distributions (or perhaps a Subversion checkout) to be used instead.
So here's what you need to do before running the register command. There are three setup() arguments that affect EasyInstall:
A URL is considered a "primary link" if it is a link to a .tar.gz, .tgz, .zip, .egg, .egg.zip, .tar.bz2, or .exe file, or if it has an #egg=project or #egg=project-version fragment identifier attached to it. EasyInstall attempts to determine a project name and optional version number from the text of a primary link without downloading it. When it has found all the primary links, EasyInstall will select the best match based on requested version, platform compatibility, and other criteria.
So, if your url or download_url point either directly to a downloadable source distribution, or to HTML page(s) that have direct links to such, then EasyInstall will be able to locate downloads automatically. If you want to make Subversion checkouts available, then you should create links with either #egg=project or #egg=project-version added to the URL (replacing project and version with appropriate values).
Note that Subversion checkout links are of lower precedence than other kinds of distributions, so EasyInstall will not select a Subversion checkout for downloading unless it has a version included in the #egg= suffix, and it's a higher version than EasyInstall has seen in any other links for your project.
As a result, it's a common practice to use mark checkout URLs with a version of "dev" (i.e., #egg=projectname-dev), so that users can do something like this:
easy_install --editable projectname==dev
in order to check out the in-development version of projectname.
setuptools includes transparent support for building Pyrex extensions, as long as you define your extensions using setuptools.Extension, not distutils.Extension. You must also not import anything from Pyrex in your setup script.
If you follow these rules, you can safely list .pyx files as the source of your Extension objects in the setup script. setuptools will detect at build time whether Pyrex is installed or not. If it is, then setuptools will use it. If not, then setuptools will silently change the Extension objects to refer to the .c counterparts of the .pyx files, so that the normal distutils C compilation process will occur.
Of course, for this to work, your source distributions must include the C code generated by Pyrex, as well as your original .pyx files. This means that you will probably want to include current .c files in your revision control system, rebuilding them whenever you check changes in for the .pyx source files. This will ensure that people tracking your project in CVS or Subversion will be able to build it even if they don't have Pyrex installed, and that your source releases will be similarly usable with or without Pyrex.
Sometimes, you need to use the same commands over and over, but you can't necessarily set them as defaults. For example, if you produce both development snapshot releases and "stable" releases of a project, you may want to put the distributions in different places, or use different egg_info tagging options, etc. In these cases, it doesn't make sense to set the options in a distutils configuration file, because the values of the options changed based on what you're trying to do.
Setuptools therefore allows you to define "aliases" - shortcut names for an arbitrary string of commands and options, using setup.py alias aliasname expansion, where aliasname is the name of the new alias, and the remainder of the command line supplies its expansion. For example, this command defines a sitewide alias called "daily", that sets various egg_info tagging options:
setup.py alias --global-config daily egg_info --tag-svn-revision \ --tag-build=development
Once the alias is defined, it can then be used with other setup commands, e.g.:
setup.py daily bdist_egg # generate a daily-build .egg file setup.py daily sdist # generate a daily-build source distro setup.py daily sdist bdist_egg # generate both
The above commands are interpreted as if the word daily were replaced with egg_info --tag-svn-revision --tag-build=development.
Note that setuptools will expand each alias at most once in a given command line. This serves two purposes. First, if you accidentally create an alias loop, it will have no effect; you'll instead get an error message about an unknown command. Second, it allows you to define an alias for a command, that uses that command. For example, this (project-local) alias:
setup.py alias bdist_egg bdist_egg rotate -k1 -m.egg
redefines the bdist_egg command so that it always runs the rotate command afterwards to delete all but the newest egg file. It doesn't loop indefinitely on bdist_egg because the alias is only expanded once when used.
You can remove a defined alias with the --remove (or -r) option, e.g.:
setup.py alias --global-config --remove daily
would delete the "daily" alias we defined above.
Aliases can be defined on a project-specific, per-user, or sitewide basis. The default is to define or remove a project-specific alias, but you can use any of the configuration file options (listed under the saveopts command, below) to determine which distutils configuration file an aliases will be added to (or removed from).
Note that if you omit the "expansion" argument to the alias command, you'll get output showing that alias' current definition (and what configuration file it's defined in). If you omit the alias name as well, you'll get a listing of all current aliases along with their configuration file locations.
This command generates a Python Egg (.egg file) for the project. Python Eggs are the preferred binary distribution format for EasyInstall, because they are cross-platform (for "pure" packages), directly importable, and contain project metadata including scripts and information about the project's dependencies. They can be simply downloaded and added to sys.path directly, or they can be placed in a directory on sys.path and then automatically discovered by the egg runtime system.
This command runs the egg_info command (if it hasn't already run) to update the project's metadata (.egg-info) directory. If you have added any extra metadata files to the .egg-info directory, those files will be included in the new egg file's metadata directory, for use by the egg runtime system or by any applications or frameworks that use that metadata.
You won't usually need to specify any special options for this command; just use bdist_egg and you're done. But there are a few options that may be occasionally useful:
There are also some options you will probably never need, but which are there because they were copied from similar bdist commands used as an example for creating this one. They may be useful for testing and debugging, however, which is why we kept them:
This command allows you to deploy your project's source for use in one or more "staging areas" where it will be available for importing. This deployment is done in such a way that changes to the project source are immediately available in the staging area(s), without needing to run a build or install step after each change.
The develop command works by creating an .egg-link file (named for the project) in the given staging area. If the staging area is Python's site-packages directory, it also updates an easy-install.pth file so that the project is on sys.path by default for all programs run using that Python installation.
The develop command also installs wrapper scripts in the staging area (or a separate directory, as specified) that will ensure the project's dependencies are available on sys.path before running the project's source scripts. And, it ensures that any missing project dependencies are available in the staging area, by downloading and installing them if necessary.
Last, but not least, the develop command invokes the build_ext -i command to ensure any C extensions in the project have been built and are up-to-date, and the egg_info command to ensure the project's metadata is updated (so that the runtime and wrappers know what the project's dependencies are). If you make any changes to the project's setup script or C extensions, you should rerun the develop command against all relevant staging areas to keep the project's scripts, metadata and extensions up-to-date. Most other kinds of changes to your project should not require any build operations or rerunning develop, but keep in mind that even minor changes to the setup script (e.g. changing an entry point definition) require you to re-run the develop or test commands to keep the distribution updated.
Here are the options that the develop command accepts. Note that they affect the project's dependencies as well as the project itself, so if you have dependencies that need to be installed and you use --exclude-scripts (for example), the dependencies' scripts will not be installed either! For this reason, you may want to use EasyInstall to install the project's dependencies before using the develop command, if you need finer control over the installation options for dependencies.
Un-deploy the current project. You may use the --install-dir or -d option to designate the staging area. The created .egg-link file will be removed, if present and it is still pointing to the project directory. The project directory will be removed from easy-install.pth if the staging area is Python's site-packages directory.
Note that this option currently does not uninstall script wrappers! You must uninstall them yourself, or overwrite them by using EasyInstall to activate a different version of the package. You can also avoid installing script wrappers in the first place, if you use the --exclude-scripts (aka -x) option when you run develop to deploy the project.
"Multi-version" mode. Specifying this option prevents develop from adding an easy-install.pth entry for the project(s) being deployed, and if an entry for any version of a project already exists, the entry will be removed upon successful deployment. In multi-version mode, no specific version of the package is available for importing, unless you use pkg_resources.require() to put it on sys.path, or you are running a wrapper script generated by setuptools or EasyInstall. (In which case the wrapper script calls require() for you.)
Note that if you install to a directory other than site-packages, this option is automatically in effect, because .pth files can only be used in site-packages (at least in Python 2.3 and 2.4). So, if you use the --install-dir or -d option (or they are set via configuration file(s)) your project and its dependencies will be deployed in multi- version mode.
This command performs two operations: it updates a project's .egg-info metadata directory (used by the bdist_egg, develop, and test commands), and it allows you to temporarily change a project's version string, to support "daily builds" or "snapshot" releases. It is run automatically by the sdist, bdist_egg, develop, and test commands in order to update the project's metadata, but you can also specify it explicitly in order to temporarily change the project's version string.
The following options can be used to modify the project's version string for all remaining commands on the setup command line. The options are processed in the order shown, so if you use more than one, the requested tags will be added in the following order:
For advanced uses, there is one other option that can be set, to change the location of the project's .egg-info directory. Commands that need to find the project's source directory or metadata should get it from this setting:
In addition to writing the core egg metadata defined by setuptools and required by pkg_resources, this command can be extended to write other metadata files as well, by defining entry points in the egg_info.writers group. See the section on Adding new EGG-INFO Files below for more details. Note that using additional metadata writers may require you to include a setup_requires argument to setup() in order to ensure that the desired writers are available on sys.path.
As you develop new versions of your project, your distribution (dist) directory will gradually fill up with older source and/or binary distribution files. The rotate command lets you automatically clean these up, keeping only the N most-recently modified files matching a given pattern.
Example 1: Delete all .tar.gz files from the distribution directory, except for the 3 most recently modified ones:
setup.py rotate --match=.tar.gz --keep=3
Example 2: Delete all Python 2.3 or Python 2.4 eggs from the distribution directory, except the most recently modified one for each Python version:
setup.py rotate --match=-py2.3*.egg,-py2.4*.egg --keep=1
Finding and editing distutils configuration files can be a pain, especially since you also have to translate the configuration options from command-line form to the proper configuration file format. You can avoid these hassles by using the saveopts command. Just add it to the command line to save the options you used. For example, this command builds the project using the mingw32 C compiler, then saves the --compiler setting as the default for future builds (even those run implicitly by the install command):
setup.py build --compiler=mingw32 saveopts
The saveopts command saves all options for every commmand specified on the command line to the project's local setup.cfg file, unless you use one of the configuration file options to change where the options are saved. For example, this command does the same as above, but saves the compiler setting to the site-wide (global) distutils configuration:
setup.py build --compiler=mingw32 saveopts -g
Note that it doesn't matter where you place the saveopts command on the command line; it will still save all the options specified for all commands. For example, this is another valid way to spell the last example:
setup.py saveopts -g build --compiler=mingw32
Note, however, that all of the commands specified are always run, regardless of where saveopts is placed on the command line.
Normally, settings such as options and aliases are saved to the project's local setup.cfg file. But you can override this and save them to the global or per-user configuration files, or to a manually-specified filename.
These options are used by other setuptools commands that modify configuration files, such as the alias and setopt commands.
This command is mainly for use by scripts, but it can also be used as a quick and dirty way to change a distutils configuration option without having to remember what file the options are in and then open an editor.
Example 1. Set the default C compiler to mingw32 (using long option names):
setup.py setopt --command=build --option=compiler --set-value=mingw32
Example 2. Remove any setting for the distutils default package installation directory (short option names):
setup.py setopt -c install -o install_lib -r
Options for the setopt command:
In addition to the above options, you may use any of the configuration file options (listed under the saveopts command, above) to determine which distutils configuration file the option will be added to (or removed from).
When doing test-driven development, or running automated builds that need testing before they are deployed for downloading or use, it's often useful to be able to run a project's unit tests without actually deploying the project anywhere, even using the develop command. The test command runs a project's unit tests without actually deploying it, by temporarily putting the project's source on sys.path, after first running build_ext -i and egg_info to ensure that any C extensions and project metadata are up-to-date.
To use this command, your project's tests must be wrapped in a unittest test suite by either a function, a TestCase class or method, or a module containing TestCase classes. Note that many test systems including doctest support wrapping their non-unittest tests in TestSuite objects. So, if you are using a test package that does not support this, we suggest you encourage its developers to implement test suite support, as this is a convenient and standard way to aggregate a collection of tests to be run under a common test harness.
By default, tests will be run in the "verbose" mode of the unittest package's text test runner, but you can get the "quiet" mode (just dots) if you supply the -q or --quiet option, either as a global option to the setup script (e.g. setup.py -q test) or as an option for the test command itself (e.g. setup.py test -q). There is one other option available:
Specify the test suite (or module, class, or method) to be run (e.g. some_module.test_suite). The default for this option can be set by giving a test_suite argument to the setup() function, e.g.:
setup( # ... test_suite = "my_package.tests.test_all" )
If you did not set a test_suite in your setup() call, and do not provide a --test-suite option, an error will occur.
PyPI now supports uploading project files for redistribution; uploaded files are easily found by EasyInstall, even if you don't have download links on your project's home page.
Although Python 2.5 will support uploading all types of distributions to PyPI, setuptools only supports source distributions and eggs. (This is partly because PyPI's upload support is currently broken for various other file types.) To upload files, you must include the upload command after the sdist or bdist_egg commands on the setup command line. For example:
setup.py bdist_egg upload # create an egg and upload it setup.py sdist upload # create a source distro and upload it setup.py sdist bdist_egg upload # create and upload both
Note that to upload files for a project, the corresponding version must already be registered with PyPI, using the distutils register command. It's usually a good idea to include the register command at the start of the command line, so that any registration problems can be found and fixed before building and uploading the distributions, e.g.:
setup.py register sdist bdist_egg upload
This will update PyPI's listing for your project's current version.
Note, by the way, that the metadata in your setup() call determines what will be listed in PyPI for your package. Try to fill out as much of it as possible, as it will save you a lot of trouble manually adding and updating your PyPI listings. Just put it in setup.py and use the register comamnd to keep PyPI up to date.
The upload command has a few options worth noting:
It can be hard to add new commands or setup arguments to the distutils. But the setuptools package makes it a bit easier, by allowing you to distribute a distutils extension as a separate project, and then have projects that need the extension just refer to it in their setup_requires argument.
With setuptools, your distutils extension projects can hook in new commands and setup() arguments just by defining "entry points". These are mappings from command or argument names to a specification of where to import a handler from. (See the section on Dynamic Discovery of Services and Plugins above for some more background on entry points.)
You can add new setup commands by defining entry points in the distutils.commands group. For example, if you wanted to add a foo command, you might add something like this to your distutils extension project's setup script:
setup( # ... entry_points = { "distutils.commands": [ "foo = mypackage.some_module:foo", ], }, )
(Assuming, of course, that the foo class in mypackage.some_module is a setuptools.Command subclass.)
Once a project containing such entry points has been activated on sys.path, (e.g. by running "install" or "develop" with a site-packages installation directory) the command(s) will be available to any setuptools-based setup scripts. It is not necessary to use the --command-packages option or to monkeypatch the distutils.command package to install your commands; setuptools automatically adds a wrapper to the distutils to search for entry points in the active distributions on sys.path. In fact, this is how setuptools' own commands are installed: the setuptools project's setup script defines entry points for them!
Sometimes, your commands may need additional arguments to the setup() script. You can enable this by defining entry points in the distutils.setup_keywords group. For example, if you wanted a setup() argument called bar_baz, you might add something like this to your distutils extension project's setup script:
setup( # ... entry_points = { "distutils.commands": [ "foo = mypackage.some_module:foo", ], "distutils.setup_keywords": [ "bar_baz = mypackage.some_module:validate_bar_baz", ], }, )
The idea here is that the entry point defines a function that will be called to validate the setup() argument, if it's supplied. The Distribution object will have the initial value of the attribute set to None, and the validation function will only be called if the setup() call sets it to a non-None value. Here's an example validation function:
def assert_bool(dist, attr, value): """Verify that value is True, False, 0, or 1""" if bool(value) != value: raise DistutilsSetupError( "%r must be a boolean value (got %r)" % (attr,value) )
Your function should accept three arguments: the Distribution object, the attribute name, and the attribute value. It should raise a DistutilsSetupError (from the distutils.error module) if the argument is invalid. Remember, your function will only be called with non-None values, and the default value of arguments defined this way is always None. So, your commands should always be prepared for the possibility that the attribute will be None when they access it later.
If more than one active distribution defines an entry point for the same setup() argument, all of them will be called. This allows multiple distutils extensions to define a common argument, as long as they agree on what values of that argument are valid.
Also note that as with commands, it is not necessary to subclass or monkeypatch the distutils Distribution class in order to add your arguments; it is sufficient to define the entry points in your extension, as long as the setup script lists your extension in its setup_requires argument.
Some extensible applications or frameworks may want to allow third parties to develop plugins with application or framework-specific metadata included in the plugins' EGG-INFO directory, for easy access via the pkg_resources metadata API. The easiest way to allow this is to create a distutils extension to be used from the plugin projects' setup scripts (via setup_requires) that defines a new setup keyword, and then uses that data to write an EGG-INFO file when the egg_info command is run.
The egg_info command looks for extension points in an egg_info.writers group, and calls them to write the files. Here's a simple example of a distutils extension defining a setup argument foo_bar, which is a list of lines that will be written to foo_bar.txt in the EGG-INFO directory of any project that uses the argument:
setup( # ... entry_points = { "distutils.setup_keywords": [ "foo_bar = setuptools.dist:assert_string_list", ], "egg_info.writers": [ "foo_bar.txt = setuptools.command.egg_info:write_arg", ], }, )
This simple example makes use of two utility functions defined by setuptools for its own use: a routine to validate that a setup keyword is a sequence of strings, and another one that looks up a setup argument and writes it to a file. Here's what the writer utility looks like:
def write_arg(cmd, basename, filename): argname = os.path.splitext(basename)[0] value = getattr(cmd.distribution, argname, None) if value is not None: value = '\n'.join(value)+'\n' cmd.write_or_delete_file(argname, filename, value)
As you can see, egg_info.writers entry points must be a function taking three arguments: a egg_info command instance, the basename of the file to write (e.g. foo_bar.txt), and the actual full filename that should be written to.
In general, writer functions should honor the command object's dry_run setting when writing files, and use the distutils.log object to do any console output. The easiest way to conform to this requirement is to use the cmd object's write_file(), delete_file(), and write_or_delete_file() methods exclusively for your file operations. See those methods' docstrings for more details.
Sorry, this section isn't written yet, and neither is a lot of what's below this point, except for the change log. You might want to subscribe to changes in this page to see when new documentation is added or updated.
XXX
Added support for "self-installation" bootstrapping. Packages can now include ez_setup.py in their source distribution, and add the following to their setup.py, in order to automatically bootstrap installation of setuptools as part of their setup process:
from ez_setup import use_setuptools use_setuptools() from setuptools import setup # etc...