[Python-Dev] Multiple levels of site-packages

[Ronald Oussoren]

Hi,

This issue first surfaced on the MacOS SIG list, but seems to be a more 
cross-platform issue.

The Python distribution, and distutils, knows of two types of 
packages/modules: those in the core distribution and the site-packages. 
This is fine when Python is locally installed anyway, but now that 
Python is distributed as part of the OS on some systems (e.g. Linux and 
MacOS X) there seem to be more levels at which you can install 
packages: between the core python packages and locally installed 
site-packages are packages installed by the system vendor. As Jack 
noticed on the MacOS Sig list there are actually even more levels of 
packages:

1) part of the python core
2) part of the system vendor packages
3) locally installed network-wide packages
4) locally installed system-wide packages
5) locally installed user packages

Items 2 to 5 are currently lumped together in site-packages. IMO it 
would be usefull to define a mechanism to define these levels of 
package collections, and to teach distutils about them. This would make 
it easier for system vendors to keep the packages provided by the 
system seperate from locally installed packages.

A related issue is that the site-packages is usually stored in the same 
part of the OS tree as the python executable, e.g. /usr/lib/python2.3 
if the executable is in /usr/bin. This means that local python packages 
will be installed into "os vendor space" when python is part of the 
system software. I think it would be prudent to define a policy on how 
system vendors should deal with this.

The Apple distribution of Python 2.3 in Panther solves this problem by 
moving site-packages to "system admin" space using a symlink. This is a 
less then perfect option because some Apple provided python packages 
are in the site-packages directory, and therefore in "system admin" 
space.

Is this really a problem worth writing a PEP for, or am I seeing ghosts?

Ronald

P.S. As this is my first post to python-dev, I'd better introduce 
myself :-). My claim to fame is PyObjC, the bridge between Python and 
Cocoa on MacOS X. I've been a (happy) Python user from about release 
1.2 (mostly on various Unix flavors and returning to the Mac with MacOS 
X)