Differences between revisions 5 and 6
Revision 5 as of 2008-01-06 10:40:08
Size: 7976
Editor: 224
Revision 6 as of 2008-01-06 18:11:06
Size: 7981
Deletions are marked like this. Additions are marked like this.
Line 178: Line 178:
in each server, but only under the categories the server has. in each server, but only under the categories known to the server.


This document provides:

  • a list of enhancements to make Distutils packaging system support several index servers;
  • changes to be made in the central server index implementation, a.k.a PyPI;
  • a guideline for implementing a PyPI-compatible server.


The PyPI is playing a very important role in the standardization of Python package distribution since Python 2.4. Together with Distutils and the third-party library Setuptools, most Python programmers are releasing their packages in eggs, and are making them avaible in PyPI.

Web frameworks such as Plone and the underlying framework, are now entirely distributed in eggs, and use PyPI as the main place to publish them. Some tools like zc.buildout automates the build of Python application by looking for eggs in PyPI and downloading them.

This central approach is great for Python advocacy, and for the cohesion of the community of developers, but brings a few issues:

  • when PyPI is unreachable, developers are unable to build their application, unless they have a local PyPI cache;
  • when people from a sub-community, like Plone, want to promote their work, they need to shout it out in several places, to make sure it has reach people focused on the given framework. These places are acting a bit like the PyPI, so a common releasing standard make thing easier to automate.

This document provides a solution to transparently support several index servers. It is a very low risk and very simple to implement. It is also finalizing a feature that was primarily intended by Distutils and PyPI, but not easy to use as-is: making it a central server, but allowing people to register and upload their package elsewhere.

The rest of the document presents the actions to take, and the work to be done for it:

  • making .pypirc supports multiple servers
  • making PyPI permissive for Trove classification
  • providing a base layer for PyPI-like servers implementation

Making .pypirc supports multiple servers

Right now, the .pypirc file is intended to keep username and password for registering a package. The file looks like


The default repository is PyPI, and when another repository has to be used, there are two options.

Either adding the repository url in the command line:

  $ python setup.py register -r http://example.com/repository

Either adding it in the .pypirc file:


In both cases, it is not possible when your username differs from a server to another, to keep a username/password for each server. Furthermore the realm associated with the server is hardcoded to "pypi".

Several sections in .pypirc

A simple way to enhance it, is to be able to add several sections in .pypirc.

For instance:




This will allow backward-compatibility, and make register and upload able to interact with several servers when they are called.

How sections in .pypirc are used

When a user call the register or the upload command with such a file, it will loop through every section, and ask the used at the prompt if he wants to perform the action over the given server.

For example:

$ python setup.py register sdist upload
running mregister
running egg_info
Using PyPI login from /Users/tziade/.pypirc
Do you want to register the package metadata at http://www.python.org/pypi (y/N)? n
Do you want to register the package metadata at http://example.com/repository (y/N)? y
Server at http://example.com/repository response (200): OK
Do you want to register the package metadata at http://example2.com/repository (y/N)? n
running sdist
running upload
Using PyPI login from /Users/tziade/.pypirc
Do you want to upload the package at http://pypi.python.org/pypi (y/N)? n
Do you want to upload the package at http://example.com/repository (y/N)? y
Submitting dist/iw.releaser-0.1.tar.gz to http://example.com/repository
Server response (200): OK
Do you want to upload the package at http://example2.com/repository (y/N)? n

XXX to be completed if if accepted, until then, iw.dist is a working prototype available at the cheeseshop.

see: http://www.nabble.com/enhancing-.pypirc-file-for-multiple-servers-handling-to14504516.html#a14527503

Making PyPI permissive for Trove classification

PyPI is based on a Trove classification, see http://www.python.org/dev/peps/pep-0301/#distutils-trove-classification. Another server may have its own trove classification though, that differs from PyPI. This is intended because the server can provide a "package center" that has its own domain-specific categories.

For example, a package might have this classification:

  Development Status :: 5 - Production/Stable
  Intended Audience :: Developers

But a slightly different one in another server. Let's say, the ACME company:

  ACME :: Visibility :: Public
  Development Status :: 5 - Production/Stable
  Intended Audience :: Developers

A permissive Trove classification would allow the registering of the package in both servers, even if the categories does not exist. For each unkown category a warning is popped at the prompt. The unknown category is not created on the server: each server keeps its own classifiers.

When registering it to several servers, the expected output would be::

        $ python setup.py register
        Do you want to register at PyPI (y/N) ? y
        Warning "ACME :: Visibility :: Public" classifier not found on the server
        200 - OK

        Do you want to register at ACME (y/N) ? y       
        Warning "Development Status :: 5 - Production/Stable" classifier not found on the server
        Warning "Intended Audience :: Developers" classifier not found on the server
        200 - OK

This will allow visual checking when a typo is made. The package will then be available in each server, but only under the categories known to the server.

XXX code the patch for PyPI code if accepted

Providing a base layer for PyPI-like servers implementation

The PyPI current implementation is done in a cgi-like Python application, that works with a Postgres database. For example, when a package is uploaded, the request's form is checked, then the data are pushed into the database through an SQL query.

Another server could have another storage strategy. So extracting an abstract layer from PyPI implementation could facilitate implementing any server.

A module could be added to provide:

  • an abstract class for the storage, that works with a backend.
  • an abstract class for the API the web server has to provide.

This module could be named interfaces.py and contain the exposed API. In this example, zope.interface is used to define these abstract class::

from zope.interface import Interface

class IPyPIServer(Interface):
    """Interface for the web server"""

    def file_upload(response=True):
        """uploads a file"""

    def list_classifiers():
        """returns list of classifiers"""

    ... more details ...        

class IPyPIStorage(Interface):
    """Defines a storage"""

    def get_classifiers():
        """returns the classifiers"""

    ... more details ...

Then each server could implement these interfaces. With such a layer, it would facilitate the standardization of all PyPI-like servers, that could all used the same base.

XXX to be completed if accepted

EnhancedPyPI (last edited 2008-11-15 14:00:12 by localhost)

Unable to edit the page? See the FrontPage for instructions.