flexible durations for unary and cumulative

[Kish Shen]

Hi,

I have been adding the functionality of posting different propagation
levels for unary and cumulative constraints for our Gecode interface.

For the flexible durations variant of both constraints, the
documentation says that
the constraints s[ + dp == ei is not enforced, and that the user
should post these constraints.

This is indeed what my existing code does. But on rereading the
documentation, I am wondering if this is necessary in our case. The
reason is that in our ECLiPSe level constraint, there are no end
times. These end times variables are added by the interface code, and
is unavailable to the user.
So for us, no useful domain information can come from the end-times,
and we are only interested in the start and duration variables.

So if we don't post the extra constraints, would this have any effect
on the domains of the start and duration?

Thanks and cheers,

Kish

[Mikael Zayenz Lagerkvist]

Hi Kish,

I'm not at the moment sure if the start+duration==end are strictly
needed for correctness and pruning, but I would recommend keeping them
around.

In general, it is quite reasonable to think of a scheduling
propagation algorithm that just prunes the minimum bounds of the start
variables and the maximum bounds of the end variables, and leaves the
rest for the start+druation==end constraints. I think there might be
some cases with reasoning like this is the cumulative and unary code,
but I'm not that familiar with it at the moment.

Cheers,
Mikael

> --
> You received this message because you are subscribed to the Google Groups "Gecode" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to gecode+un...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/gecode/CAHM-QAvYhwqq_8yy_yy%2BgOQySGJR9NF%3D8Vzu2_HSDvGDELe_0A%40mail.gmail.com.



--
Mikael Zayenz Lagerkvist