RPM roadmapping

[Owen Taylor]

On Fri, 2007-07-27 at 12:05 -0400, Dimi Paun wrote:
> On Fri, 2007-07-27 at 10:33 -0400, Jeremy Katz wrote:
> > We need to get to where we're doing a lot of small
> > transactions rather than one so that things like power failures aren't
> > catastrophic. 
> 
> That would be a step backwards in all but the power-failure scenario.
> Conceptually, the transactional part of rpm is is a good thing. Now
> I know we're nowhere close to ACID, but maybe we can improve on that
> instead of making it even worse.

The problem is that you need transactions not for the RPM database
but for the *system*; if you are installing 20 packages, and the
first package install overwrites some file, then something fails
as we are running the %post for the 15th package, can we go back and
restore that file we overwrote to it's original state?

To even start at being able to do that, I think you have to have
separate filesystem for the system image that is read-only except
for RPM operations. 

(Yes, RPM could save the files it overwrites in a undo buffer, but
you also have system changes done by scripts.)

						- Owen