Thank you for taking the time to answer.

On 18/04/07, Mikael Zayenz Lagerkvist <[EMAIL PROTECTED]> wrote:
When the search-procedure recomputes a node, it will not follow
exactly the same path as the first time - several branching decisions
are added at once and then a fixpoint is computed (this is called
batch recomputation). If this fails, the an error such as the one you
observed might be thrown. This is one potential cause for your
problem.

If i turn off re-computation i no longer suffer this problem, so this
could very well be the source of my problem.

One of the key properties that Gecode uses for enabling batch
recomputation this is that propagators should be monotonic. Typical
cases where propagators are non-monotonic might be: not subscribing to
all events used, using heuristic/random propagation algorithms,
reporting ES_FIX when that is not the case, and of course general
bugs.

I think what i am working on fits nicely into the category of
heuristic propagation. The propagator in question is not generally
monotonic (ie it might detect in-feasibility on one set of domains and
neglect to spot in-feasibility of a stricter set of domains) and in
addition the result of assign, propagate, assign,propagate and
assign,assign, propagate, propagate is not guaranteed to be the same.
Propagation is however completely deterministic.

So i guess this should be the source of my problem, right?

I suppose i will just have to run without re-computation.

Thank your for explaining this. The propagator im working with is a
bit 'special' so i suppose it isn't too surprising that im
encountering this type of issue :)

Peter Tiedemann

_______________________________________________
Gecode users mailing list
[EMAIL PROTECTED]
https://www.gecode.org/mailman/listinfo/gecode-users

Reply via email to