[2.7.4] Reified element constraint

[IgnaceBleukx]

Since the update to v2.7.4 I get an invalid solution to the following model:

% Generated by CPMpy
include "globals.mzn";

var -3..5: x;
var -3..5: y;
var -3..5: z;
var 0..10: idx;
var -3..5: val;
var bool: p;
constraint ((
    let { array[int] of var int: arr=array1d(0..2,[x,y,z]) } in
      arr[idx]) == (val)) -> (p);

Which is a model generated by our CPMpy library. We use the alternative array to circumvent the differnece between 1-indexed Minizinc arrays vs 0-indexed Python arrays.

The solution retured on version 2.7.4 by Gecode is:

x = -3;
y = -3;
z = -3;
idx = 0;
val = -3;
p = false;

While v2.7.3 returns

x = -2;
y = -3;
z = -3;
idx = 0;
val = -3;
p = false;

[Dekker1]

It looks like the result for MiniZinc 2.7.3 would also have been incorrect since the value -2 should not have been allowed according to the domain of x.

[cyderize]

It looks like the result for MiniZinc 2.7.3 would also have been incorrect since the value -2 should not have been allowed according to the domain of x.

-2 is in the range -3..5, so I think the old solution was valid - just the new one is wrong since p should be true

[Dekker1]

Ah, of course. My brain must still be half asleep.

I was indeed investigating the issue with p for the current solution

[PerMildner]

I was about to report another reification-related bug in 2.7.4 (not present in 2.7.2). Do you have nightly builds or some such where the latest fixes can be tried out to see if my problem is gone?

[Dekker1]

We don't create the full packages (at least not signed and available), but you can try using the Docker edge tag: https://hub.docker.com/r/minizinc/minizinc/tags or the edge channel when using snap: https://snapcraft.io/minizinc