HardReset returns the variable to an inconsistent creation state

[gadavidd]

Choco Solver's hardReset method resets the domain of the variables to their original state. However, there is inconsistent behavior for variables of type SetVar and IntVar with an enumerated domain. While real variables ( RealVar ) and bounded IntVar are correctly reset, this is not the case for SetVar and enumerated IntVar variables.

The inconsistency occurs as follows:

SetVar is reset to {} (empty set), regardless of the original domain.
Enumerated IntVar directly assumes the first value of the domain instead of maintaining the expected behavior (full original domain).

Expected behavior:

According to the documentation, hardReset should return all variables to the creation state, with their domains complete and unchanged. This behavior works correctly for RealVar and bounded IntVar variables, but fails for the other configurations mentioned.

Model model = new Model();
model.getEnvironment().worldPush();

SetVar s1 = model.setVar("s1", new int[]{}, new int[]{1, 2, 3, 4, 5});
RealVar r1 = model.realVar("r1", 0, 10, 0.1);
IntVar i1 = model.intVar("i1", 0, 100, true);  // bounded IntVar
IntVar i2 = model.intVar("i2", 0, 100, false); // enumerated IntVar

model.getEnvironment().worldPush();
model.getSolver().hardReset();

For the example, you can see that the domains of RealVar (r1) and bounded IntVar (i1) are correctly restored to the creation state. However, the domains of SetVar (s1), which is reset to {} (empty set), and Enumerated IntVar ( i2 ), which automatically takes on the first value of the domain, do not reflect the expected behavior.

On investigation, it seems to me that the problem begins during the creation of the Environment type variables for the two cases (s1 and i2), which are either created empty (SetVar) or are initially created assuming the first value of the domain (IntVar) unlike bounded domain variables. Later, with hardReset, this state of inconsistency will be returned, making it impossible to use the mechanism in this context.

It seems to me that some treatments could be used to solve this problem:

1. Modify the creation of variables in the Environment: SetVar and enumerated IntVar could be initialized with the full domain to avoid inconsistencies;

2. Prevent the return to an incorrect state during the hardReset.

I also tried inserting a new state into the search tree with worldPush(), but the behavior remained inconsistent. Are there any suggestions or possible solutions for dealing with this incorrect behavior? It seems to me that only by modifying the shock can this problem be solved.

[cprudhom]

I see two options here:

1. Remove the first call to model.getEnvironment().worldPush(),
2. Call model.getEnvironment().worldCommit() after the model's population.