On Jan 18, 2008 12:20 PM, Vincent Hennebert
<[EMAIL PROTECTED]> wrote:
> Andreas L Delmelle wrote:
>
> > I have always somehow assumed there to be a threshold in the most common
> > cases. In the sense that certain (maybe in fact even the bulk of)
> > feasible breaks can already be excluded quite early, even if the
> > sequence consists of millions of elements. Simply because they would, in
> > intermittent runs, repeatedly be considered to be 'bad' breaks, or
> > because they become 'worse' than the previous run.
>
> That's not possible. The last page of a sequence of 300 pages may
> influence the choice of the very first page break.
(well, first of all, hi! :-) quite a long time since my last message ...)
Actually, I think that even using total-fit there could be situations
where some feasible breaks can be "promoted" to actual breaks before
the end of the sequence is reached.
Surely, if the algorithm needs to restart then the chain of feasible
breaks up to the restarting node can be promoted, as there isn't any
other alternative set of breaks for that part of the sequence: but, as
a restart is definitively a bad thing, this does not help us in the
likely and desirable situation where everything runs ok from start to
end.
I was wondering whether this the promotion can be performed also each
time the active node list contains just a single node ...
Luca
