# 3.4. Automatic Solution Checking¶

MiniZinc provides support for checking the correctness of solutions of a model by using a *checker model*. This approach has two main applications:

- Instructors can provide checker models for student assignments. This provides students with immediate, detailed feedback on their modelling attempts.
- A simplified checker model can be used to verify a complex model used for solving. This can be useful when experimenting with new decompositions of constraints, or for post-processing solutions if some constraints cannot be added to the original model (e.g. in case the solver does not support certain constructs).

Running MiniZinc with solution checking is easy. On the command line, simply pass the name of the checker model in addition to the problem model:

```
minizinc model.mzn model.mzc.mzn data.dzn
```

Checker models can be pre-compiled in order to obfuscate their contents (e.g. if they contain clues to students how to model the problem):

```
minizinc --compile-solution-checker model.mzc.mzn
```

This will create the compiled checker model `model.mzc`, which can be used instead of the clear text version for checking:

```
minizinc model.mzn model.mzc data.dzn
```

The MiniZinc IDE has built-in support for solution checkers. Whenever the current project contains a file with the same file name as the current model but file extension `.mzc` or `.mzc.mzn`, the “Run” button turns into a “Run+Check” button.

The rest of this section describes how to implement checker models.

## 3.4.1. Basic checker models¶

At its core, a checker model takes each solution that a solver produces as input, and outputs whether the solution is correct or not. Let’s use the simple map colouring model from the tutorial as an example. Here is the model again:

```
% Colouring Australia using nc colours
int: nc = 3;
var 1..nc: wa; var 1..nc: nt; var 1..nc: sa; var 1..nc: q;
var 1..nc: nsw; var 1..nc: v; var 1..nc: t;
constraint wa != nt;
constraint wa != sa;
constraint nt != sa;
constraint nt != q;
constraint sa != q;
constraint sa != nsw;
constraint sa != v;
constraint q != nsw;
constraint nsw != v;
solve satisfy;
output ["wa=\(wa)\t nt=\(nt)\t sa=\(sa)\n",
"q=\(q)\t nsw=\(nsw)\t v=\(v)\n",
"t=", show(t), "\n"];
```

A checker model for this model requires the values of the variables `wa`

, `nt`

, and so on, for each solution, and then has to test whether all constraints hold. The output of the checker model should contain a line starting with `CORRECT` if the solution passes the test, or `INCORRECT` if it doesn’t.

Since these values will be fixed in any solution, checker models simply declare parameters with the same name as the model variables:

```
int: wa;
int: nt;
int: sa;
int: q;
int: nsw;
int: v;
int: t;
output [
if wa!=nt /\ wa!=sa /\ nt!=sa /\ nt!=q /\ sa!=q /\
sa!=nsw /\ sa!=v /\ q!=nsw /\ nsw!=v
then "CORRECT\n"
else "INCORRECT\n"
endif
];
```

Running the model and the checker will produce output like this:

```
wa=3 nt=2 sa=1
q=3 nsw=2 v=3
t=3
% Solution checker report:
% CORRECT
----------
```

The solution checker report is embedded as comments in the original output.

## 3.4.2. Detailed feedback¶

The basic checker model above only reports whether the solutions satisfy the constraints, but it doesn’t provide any insights into the nature of the error if a constraint is violated.

We can use standard MiniZinc functionality to provide much more detailed feedback. The following checker model introduces a helper function `check`

which outputs a detailed error message if a constraint doesn’t hold. The results of all the checks are combined, and if any of the constraints was violated, the output is `INCORRECT`, otherwise it is `CORRECT`.

```
int: wa;
int: nt;
int: sa;
int: q;
int: nsw;
int: v;
int: t;
test check(bool: b,string: s) =
if b then true else trace_stdout("ERROR: "++s++"\n",false) endif;
output [
if check(wa!=nt, "wa and nt have the same colour")
/\ check(wa!=sa, "wa and sa have the same colour")
/\ check(nt!=sa, "nt and sa have the same colour")
/\ check(nt!=q, "nt and q have the same colour")
/\ check(sa!=q, "sa and q have the same colour")
/\ check(sa!=nsw, "sa and nsw have the same colour")
/\ check(sa!=v, "sa and v have the same colour")
/\ check(q!=nsw, "q and nsw have the same colour")
/\ check(nsw!=v, "nsw and v have the same colour")
then "CORRECT: all constraints hold.\n"
else "INCORRECT\n"
endif
];
```

However, the checker model will only report the first violated constraint, since the conjunction operator short-circuits the evaluation when one of its arguments is false. For example, if we remove all constraints from the original model, the output would be:

```
wa=3 nt=3 sa=3
q=3 nsw=3 v=3
t=3
% Solution checker report:
% ERROR: wa and nt have the same colour
% INCORRECT
----------
```

In order to get all error messages, we can force the evaluation of all checkers by creating an auxiliary array of check results:

```
int: wa;
int: nt;
int: sa;
int: q;
int: nsw;
int: v;
int: t;
test check(bool: b,string: s) =
if b then true else trace_stdout("ERROR: "++s++"\n",false) endif;
array[int] of bool: checks = [
check(wa!=nt, "wa and nt have the same colour"),
check(wa!=sa, "wa and sa have the same colour"),
check(nt!=sa, "nt and sa have the same colour"),
check(nt!=q, "nt and q have the same colour"),
check(sa!=q, "sa and q have the same colour"),
check(sa!=nsw, "sa and nsw have the same colour"),
check(sa!=v, "sa and v have the same colour"),
check(q!=nsw, "q and nsw have the same colour"),
check(nsw!=v, "nsw and v have the same colour")
];
output [
if forall(checks)
then "CORRECT: all constraints hold.\n"
else "INCORRECT\n"
endif
];
```

Now the output contains all error messages (for the case where the model has no constraints):

```
wa=3 nt=3 sa=3
q=3 nsw=3 v=3
t=3
% Solution checker report:
% ERROR: nsw and v have the same colour
% ERROR: q and nsw have the same colour
% ERROR: sa and v have the same colour
% ERROR: sa and nsw have the same colour
% ERROR: sa and q have the same colour
% ERROR: nt and q have the same colour
% ERROR: nt and sa have the same colour
% ERROR: wa and sa have the same colour
% ERROR: wa and nt have the same colour
% INCORRECT
----------
```

## 3.4.3. Instance data in checker models¶

The map colouring example was quite simple because the model did not contain any parameter declarations. For a model that is parameterised, the checker model simply contains the same parameter declarations. MiniZinc will then pass the actual parameters of the instance being solved to the checker model.

For example, the following checker model could be used for the *n*-Queens problem.

```
int: n; % number of queens
set of int: ROW = 1..n;
set of int: COL = 1..n;
array[int] of int: q; % col of queen in each row
test check(bool: b,string: s) =
if b then true else trace_stdout("ERROR: "++s++"\n",false) endif;
output [
if check(index_set(q)=1..n, "ERROR: array q should have index set 1..\(n)")
/\ forall(i in 1..n)(check(q[i] in 1..n, "ERROR: q[\(i)] should have a value in 1..\(n)"))
/\ forall(r1, r2 in 1..n where r1 < r2)
(check(q[r1] != q[r2],
"queens in rows \(r1) and \(r2) are on same column\n")
/\
check(q[r1]+r1 != q[r2]+r2,
"queens in rows \(r1) and \(r2) are on same up diagonal\n")
/\
check(q[r1]-r1 != q[r2]-r2,
"queens in rows \(r1) and \(r2) are on same down diagonal\n")
)
then "CORRECT: All constraints hold"
else "INCORRECT" endif ];
```

The checker model first makes sure that the solution has the right dimensions (correct array index set, and each variable is assigned a value in the correct domain), and then uses standard MiniZinc constructs to check each constraint.

## 3.4.4. Checking optimisation problems¶

Optimisation problems pose a difficulty for automatic checking. When a solver claims to prove optimality, we cannot easily verify this claim without solving the problem again (using a different model that is known to be correct). At the moment, solution checking for optimisation problems is restricted to checking that the objective has been computed correctly.

To check that the objective value is calculated correctly, define a parameter called `_objective`

of the appropriate type (`int`

or `float`

) in the checker. The example in the next section illustrates this.