[Pythonmac-SIG] PackageManager philosophy

[Jack Jansen]

As it turns out Bob's ideas and my ideas are pretty similar, see his 
last message.

The only difference seems to be that I prefer to have a small number of 
moderately rigorously (if that isn't a contradictio in terminis:-) 
tested modules, and he wants a more comprehensive set of lightly tested 
modules.

And Kevin points at a solution:

> To make PM behave more in line with your goals (as I understand them, 
> anyways) while still giving Bob and others (myself included) the 
> functionality and flexibility they desire, I was thinking there could 
> be some sort of "stamp of approval" to say that the 'scapegoat' 
> (Package Inspector - PI? =) has tested package X binaries on platforms 
> X, Y and Z. A warning could pop up if such an approval doesn't exist - 
> "Warning: This package has not been confirmed to work on your 
> particular platform or sufficiently tested. If you experience any 
> installation problems, please report any problems to 
> scapegoat at python.org" - sort of like digital signatures. Or, the end 
> user could configure PM to simply not show any untested packages. (it 
> could also be shipped this way by default)

I think the stamp of approval is that a package is in the default 
database. How about the following: per platform there are two 
databases, the stable one and the experimental one. The experimental 
one is easier to select than currently, probably with a checkbox. The 
experimental database is actually only a set of include directives 
pointing off to a small number of databases maintained by Bob, Kevin, 
me and a few others. PM is modified so that when a database is included 
it knows that the maintainer of the included database is the scapegoat 
for those packages (currently the maintainer of the toplevel database 
is responsible for everything).

One problem that we would still need to solve on the user side (there's 
lots of issues to solve on the scapegoat side) is that of finding 
packages, especially in the potentially large experimental database. 
The logical thing to do would be to use the PyPI model, but as of this 
writing it just doesn't cut it. I just tried find a gui package for 
MacOSX (pretending I didn't know any names). Whatever I typed in I 
couldn't find anything. I eventually located pyobjc by typing "cocoa" 
into the *description* field, but that's it.
--
- Jack Jansen        <Jack.Jansen at oratrix.com>        
http://www.cwi.nl/~jack -
- If I can't dance I don't want to be part of your revolution -- Emma 
Goldman -