Problem with the current model.

In the current system, changes could be in 'merge' which may not end up in 'untested' for a long time. This will disrupt read access to the code that exists beyond merge. An extreme example would be where a change in 'merge' deleted or renamed files.

A solution would be to keep the future stuff in branches above 'merge'. Perhaps in logical sections, a bit like the branching that was done with SVN. (See diagram) ccw 12:46, 12 September 2009