Ned Konz <ned at bike-nomad.com> wrote:
> On Tuesday 19 November 2002 07:01 pm, Daniel Joyce wrote:
> > > My thinking exactly.
> > >
> > > My "corollary" thought is: just why does SqueakMap even need a
> > > "Universally Unique ID", anyway?
> > >
> > > Isn't "locally unique" (to the SqueakMap registration server)
> > > good enough?
> 
> No, because there may be more than one server. And packages may be 
> registered to any of a number of servers.
> 
> > Why not let the server set UUIDs? That way, the server can
> > guarantee uniqueness.
> 
> The server *is* setting UUIDs. Unfortunately, it was doing so without 
> managing to make them unique.

Exactly. The design here is so that we can move SM into a more
distributed topology.
I am planning to add the capability of registering packages locally and
then "publish" them later. They will then be uploaded to the masters.
The simplest way to make them have worldwide unique ides was to use...
he! UUID. What is otherwise the whole point with that class?

If I can not trust it to produce unique ids then I can of course do all
sorts of other things, like handing out "id ranges" to local servers
etc. But I just didn't want to bother with that.

Obviously I can live with the "infinitesimal chance of duplicate ids" -
so we just need to fix the Random problem (Random is bound to the saved
image) - perhaps someone has already posted a fix.

For now I will just add a check and a retry if an id is taken (which I
can get away with for now since it is done centrally).


regards, Göran

PS. Starting SM repair as we speak, the "missing" packages will
reappear. Will post when done.