Calling a method on the solution from a constraint

[hyphz]

Is there a way to call a method on the solution class from a constraint? The ConstraintFactory seems to force you to use one of the ForEach variations, and to identify the list to go through by type rather than by name. Is there not an option to get the current Solution candidate?

Context: I'm experimenting with using Timefold to solve a simple box packing problem (spacefold?) Conveying a rule preventing boxes overlapping isn't too difficult as they are axis-aligned rectangles and can be checked using forEachUniquePair. However, I now need to check that each box is supported by either the bottom of a shelf or another box. The problem is that a box can be supported by any number of other boxes; obviously a loop through all combinations of boxes would be horribly slow. But support can be checked relatively easily by maintaining a grid voxel map of all boxes on each shelf. However this map could not be stored inside a single planning entity because it is a property of all of them. So if I have a method on the solution that checks for support, how can I call it from a constraint?

[triceo]

Hello @hyphz!

I can see three ways of doing this:

• The first and probably the most performant approach would be to pass the map to each entity's constructor, and access it from there. Just make sure solution cloning deals with this map correctly, you may need to deep clone it.
• The second approach would be to create a shadow variable, where each entity will get an automatically updated field with the correct value, backed by the map.
• The third approach would be to create a meta object, sometimes called "parameterization", mark it as a problem fact on the planning solution, and then join with that object in your constraints. That object can then include any context you wish, assuming the context never changes. This approach may not work if your map contains entities, because those change.

As you can see, each approach has its benefits and downsides. Most often, we would use a shadow variable for this, understanding that the maintenance of shadow variables will cost you a little performance.

[hyphz]

Thanks! If the map is deep cloned when the solution is cloned, how do I make sure that any cloned planning entities have their reference updated to point to the cloned map?

I like the idea of the shadow variable but that seems to inherit the same problem as the code to update the shadow variable would have to hold a reference to the map.

[triceo]

The cloning logic is smart enough to know what needs to be cloned, and once it clones something, all future references to that something are replaced by the clone. So, as long as you mark your fields properly for deep cloning, that ought to be enough.

Wrt. shadow variable - yes, you're right. The map in your design can be considered some kind of a context which always needs to be present.