[Zope3-dev] Re: Packaging pre-proposal/notes

[Tres Seaver]

Anthony Baxter wrote:
> Philipp von Weitershausen wrote:
> 
>> Martijn Faassen wrote:
>>
>>> Note that "The Python Community" has had several attempts to improve the
>>> packaging situation for Python (CPAN for Python projects, etc) for
>>> the last 6 years at least.
>>
>>
>>
>> If you want my frank opinion, I think that's very pathetic. We're not 
>> even close to what Perl has...
> 
> 
> "Send Code".

PEPs *do* get rejected, even those with clear-and-present usage in the 
field (interfaces, for instance).

> Sorry, but this stuff is hard. The people who're trying to work on it 
> have to do this in their spare time - for instance, the current pypi 
> system.

Note that PyPI invested effort to add Trove metadata, which I would call 
"nice-to-have", and left out dependencies, which I would call "crucial". 
  I am not trivializing the effort required to add dependencies, either; 
  I just don't consider PyPI very useful without them.

> There's always lots of "why doesn't python have a CPAN" complaints, but 
> very few people who're willing to help fix the problem.
> 
> Anthony, slightly cranky.

The current distutils implementation makes too many assumptions about 
how Python-based solutions are to be distributed / installed;  in 
particular, its sweet spot could be described as "installing a 
standalone module / package into site-packages."  Almost *any* of the 
non-particles in that prhase can be reasonably challenged:

   'installing' -- I need to be able to *package* software for
     later installation;  distutils won't be involved in putting
     the actual bits onto the deployed system.

   'standalone' -- distutils no mechanisms for describing dependencies,
     which may be largely an artifact of how the standard library is
     organized (and the fact that the standard library is so big).

   'package / module' -- distutils is weak in its handling of
     "applications", in particular w.r.t. the "data" part.

   'site-packages' -- I almost *never* want to write into the
     Python library when I install application-related stuff.

We should adopt a mechanism for expressing explicit dependencies, and 
beat out a framework over distutils which uses it;  we will then be in a 
position to nominate that framework for inclusion in 2.5 (I would guess; 
  I don't know how close to a freeze 2.4 is today), and be able to move 
on even if, as with interfaces, Python doesn't adopt our nominee.

Tres.
-- 
===============================================================
Tres Seaver                                tseaver at zope.com
Zope Corporation      "Zope Dealers"       http://www.zope.com