[packagekit] Resolve method

[Richard Hughes]

On Wed, 2007-10-03 at 11:35 -0400, Elliot Peele wrote:
> On Wed, 2007-10-03 at 09:02 +0200, Tim Lauridsen wrote:
> > Richard Hughes wrote: 
> > > What is the usecase for this? firefox.i386 wants gnash.i386 on an
> > > otherwise x64 system? Would libflash.x64 even exist in the repos or be
> > > installable?
> > > 
> > > The reason I want an example is that it's pointless discussing
> > > theoretical cases that make the daemon a lot more complex when it's just
> > > theory. Plus, we can get to a solution more quickly without
> > > analogies ;-)
> > >   
> > I agree with Richard the multilib shall be handled in the backend,
> > different backend and distributions handles multilib in different
> > ways, so there is no "This is the right way to do multilib".
> 
> I agree that the backend should handle multilib, however I think that we
> are going to run into compatibility issues when applications start being
> able to install other packages.

Sure, but as soon as you add architecture info into the abstraction
something has to make a choice; packagekit is not clever enough to do
this; and the user isn't going to care.

I would love to see my mum's face when she sees this:

"You wanted to install the borwser plugin. Do you want to use the x86_64
package or the x86 package?"

> A real world example of this case:
> 
> On an x86_64 system I have x86 firefox. Firefox requests mplayerplug-in,
> we have both x86 and x86_64 versions of mplayerplug-in available. How
> does the backend know which arch to install?

If I type in "conary install mplayerplug-in" (or whatever the command
is), what gets installed?

Richard.