Your search query "linkto%3A%22A new pypi module%22" didn't return any results. Please change some terms and refer to HelpOnSearching for more information.
(!) Consider performing a full-text search with your search terms.

Clear message

Create a new package in python called pypi


This PEP describes how the commands that are used to register and upload a package to PyPI can be extracted from distutils and put in a new independant package in Python called pypi, that would also describe the PyPI protocol.


distutils is responsible for too many things, and the register and upload commands are completely standalone. In other words they can be extracted from distutils and placed into a new package that would also describe the protocol used by the PyPI server.

Secondly, there are a lot interactions in those two commands. They are interacting with the user when the .pypirc file is created, and they implement an authentication mecanism that pushes the user login and password values into the HTTP request. This is done by reading a clear text password in .pypirc.

The users have to call a very precise sequence of commands with in order to upload or even upgrade their packages.

Let's improve all of this by:

The last point would let anyone implement this protocol for the client-side or the server-side, by using this package as a base.

Improving password handling

Currently, the password and the user are in clear text in .pypirc. This is a bit insecure. Let's make it an optional and if not provided let the user type it at the prompt when needed.

Improving metadata controls

When a package is uploaded at PyPI, there are several things that can be done on the client-side in order to control it.

A new variable in .pypirc can be added in order to block a registration or an upload if the metadata does not meet the minimum requirements:


From there, the command line utility will be able to decide if it should continue or not. The problem with the reST control is that it would require docutils.

Providing a command line utility, independantly from

Let's drop the command line to register and upload a package. The new pypi package can handle this as long as the package is pointed. A high-level script can be provided in the Scripts/ folder of Python, and a developer can use it to control, register or upload a package. And also to browse PyPI.

Here's a example of such a session:

    $ cd my.package
    $ pypi check
    Checking metadata...
    Warning : The 'url' metadata is missing
    Warning: The long_description seem to be in reStructuredText, 
    but does not compile

    $ pypi register -r tarek
    Registering the package using "tarek" account in .pypirc
    Registered !

    $ pypi upload 
    What kind of release do you want to upload ?
    Available releases:
        sdist   Source release
        bdist   Binary release

    Type the desired released, separated by a space [sdist] : _

    $ pypi passwd tarek
    Changing password for tarek
    What kind of hash (default: sha1) ? sha1
    Type password:
    Type password again:
    Password changed

The pypi command could also let you browse PyPI, like the Yolk project does (see

Browsing capabilities using PyPI XML-RPC features : XXX TBD

Clearly describe the PyPI protocol

XXX TBD : describe the methods inmplemented in the CheeseShop

How ?

A first refactoring was made a few months ago to allow users to handle several PyPI logins and servers in .pypirc ( and the code responsible for .pypirc managment and for handling the registering and the upload were isolated.

The first action would be to create a new package called pypi and to copy the code from distutils, in order to make it work on its own.

The files to use are:

We can add deprecated flags into distutils, just to warn people to use the new module instead.

XXX to be finished

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