[Catalog-sig] [Distutils] People want CPAN :-)

[Paul Boddie]

James Bennett wrote:
> On Fri, Nov 6, 2009 at 7:52 PM, Chris Withers <chris at simplistix.co.uk> 
wrote:
> > This feels like a post relative to the catalog-sig too, or maybe even
> > moreso than distutils-sig...
>
> The question that's on my mind here, and which I haven't really seen a
> solid answer to in the various threads, is which parts of CPAN people
> seem to want. The original post mainly seemed to be concerned with
> saying "install this" and having it automagically work, dependencies
> and all, which is much less a PyPI issue and much more a "sort out the
> packaging and installation tools" issue than a PyPI issue.

There are tens of messages on the distutils-sig about this now, most of them 
presumably discussing the innards of distutils and conflating distribution 
and installation as usual.

The comment mentioned R packages and CPAN. My colleague who uses R has 
demonstrated the mechanism for R, and it didn't seem completely automatic - 
you had to load the package from a particular address - although maybe it did 
pull in dependencies as well, and maybe you can configure the addresses and 
not think about them too hard. It reminded me of work done by various people 
(including my brother) to import Python packages over HTTP.

As for CPAN, my last experience with it was a multi-hour nightmare involving a 
new installation of Bugzilla and trying to persuade CPAN not to...

1) upgrade the date/time package beyond the version that actually works with 
Bugzilla to precisely the one which explicitly doesn't work with Bugzilla 
(where in the end I dug around on the CPAN site until I found the precise, 
unpublished URL yielding archived versions), and

2) render itself incapable of obtaining and installing packages.

Previous experiences also included tracking down package build failures in the 
overly verbose output. Maybe CPAN packages are tested upon upload or 
something, but it's somewhat short of what stuff like Debian manages to 
provide. If CPAN is the goal for Python packaging then we are aspiring to 
fail our users.

I work in a scientific/academic environment - in bioinformatics, in fact - and 
the biggest obstacle to just using proper package management is the typical 
institution-wide mentality of locking down Unix-like machines, and yet giving 
everyone on Windows machines administrator-like privileges and "free rein" 
[1] to mess up their computers with all sorts of stuff.

Meanwhile, packages like this would help solve the problem of the commenter:

http://packages.debian.org/lenny/python-biopython

Paul

[1] http://www.dailywritingtips.com/free-rein-or-free-reign/