Stephane Ducasse <ducasse at iam.unibe.ch> wrote:
> Hi
> 
> this would be a pity because we would introduce a lot of errors in 
> external packages.
...
> Now I was thinking that we are doing too much in the same time. 
> Cleaning in a closed world assumption is much easier
> than in presence of removal.

But of course you can't work under a closed world assumption - unless
you load everything in SM (and probably different things in it clash),
you're by definition breaking stuff whenever you change APIs. 

> So now that we are following this path, I think that the best way to 
> continue would be to
> have a script loading all the packages that will be present in the 3.6 
> full image and that the cleaners
> works on this image.

IMO, the solution is not for you to work with a different subset of
stuff than the alpha stream, the solution is along the lines of tests
and tools for external package maintainers. And requires SM 1.1, for
package version support.

> What will be absolutely important for the future is to have package (a 
> class knowing which class and methods it contains), this way (except 
> for class extensions made from one package to another one), we could 
> make that each
> change related to a given package goes into the right package.

This is orthogonal, but yes, I agree we'll need something like that.
PackageInfo can do this, but depends on people adhering to its
standards.

Daniel