Martin Wirblat <sql.mawi at t-link.de> wrote:
[SNIP]
> Compared with the sig-namespace version I would strongly vote for 
> clean variable names and the central name server. Having an 
> infrastructure of a server hierarchy an additional name checking 
> authority should not make much difference. 
> 
> In a way the sig-namespace must be checked too - who assures that no 
> one uses a sig privately that no one else uses? Look at todays 
> situation, there is just a webpage, and it is by no means complete. 

Ok, so in short you are proposing:

- No UUIDs.
- No namespaces. Or rather ONE central flat namespace of packagenames.
:-) Can old names be reused by other packages? I think not. Could be
reused by the same package I guess.
- Move the package name to PackageRelease and make it immutable there.
This means you will need to do a new release to change a name.

This gives us the way to refer to specific releases as
"name+autoversion" or "name+manualversion" (manualversion would need to
be verified to be unique of course within the package).

If you just write "name" then it can resolve to either the last version
of that specific name or the last version of that package regardless if
it has a different name now. Depending on the context.

Ok, sounds fine to me. What about the cache then? Noone bothered to give
replies on that I think... ;-)

> regards,
>  Martin

regards, Göran

PS. As you see I am a guy who can change his mind. You just need to work
hard sometimes. :-)

PPS. I am not sure I want to kill the UUIDs from the model though.
Categories have them too. And other things will need an id. And if we
get into trouble I think it may be wise to have them there "just in
case" to be able to track what the heck happened...