Time for a Python "distribution" ? was:Not enough Python library development

[Skip Montanaro]

    Chris> Unfortunately, not having all the major modules available for a
    Chris> single source really is a big limitation. Not only do people like
    Chris> me have to spend a whole lot more time installing stuff,
    Chris> developers are limited as to what they can expect a Python
    Chris> installation to include.

I'm not sure there's too much farther to go from where Python's distutils
sits now to where you need to go.  I've used Perl's CPAN shell to install
packages and recursively install their dependencies.  While it's a bit
chatty, I generally haven't had any problems with it.  With distutils I
believe you get more-or-less platform-independent package installation (at
least that's the plan).  If you can add package dependencies to a package's
distutils spec and adopt a standard set of package download schemes (ftplib
and urllib may be all that's needed) as well as a set of package mirror
sites, then you've got something quite similar to CPAN.  You also need a way
to announce the canonical location of a package so mirrors can get at it.
The Vaults of Parnassus may already contain that information.

Obvious tasks I see:

    * binary vs. source - in particular, Windows users often won't have
      compilers 

    * extend distutils (esp. setup.py) to allow authors to record package
      dependencies

    * establish a site mirror facility

    * solve problems with potential name clashes

-- 
Skip Montanaro (skip at pobox.com)
(847)971-7098