Seamless integration of LongVar (64-bit integer)

[dimitri-justeau]

I believe that integer variables, based on 32-bits integers, are suited for most usages of choco-solver. Operations on 32-bits integers are also apparently faster in Java. However, I know from a recent experience that 64-bit long variables could also be helpful in some situations.

Ideally, having a LongVar extending IntVar for such particular cases would be perfect. However, this involves some design challenges:

• It would need the implementation of backtrackable longs.
• Propagators implemented for int variables should also work seamlessly with long vars.
• Mixing int vars and long vars in propagator should be seamlessly possible.

In summary, I think that if we can implement this feature as a pure extension, without impacting propagators implementation nor solver performances would be a great asset for choco-solver.

If judged interesting by the community, I am ok to get involved in this work.

[cprudhom]

As a first step, which does not fit entirely the need, I propose to add the following API to IntVar:

• removeValue(long value, ICause cause)
• instantiateTo(long value, ICause cause)
• updateLowerBound(long value, ICause cause)
• updateUpperBound(long value, ICause cause)

which automatically deal with int casting (if possible).
This would make possible to deal with large range of values in IntVar, for instance very close to Integer bounds, without unexpected behaviors. This would make easier to prevent from underflow/overflow cases.
For instance, PropTimes can be adapted quite easily.

[dimitri-justeau]

There might be some radical changes about the way Java handles primitives, which should include generics over primitives:

https://www.levi9.com/articles/10-ways-java-is-getting-better/#anchor-7

https://openjdk.org/projects/valhalla/

I think that we should stay tuned about this :-)

[cprudhom]

This argues for upgrading the JDK to a new one