Added output fails on enum type.

[Moisan]

When a model contains variables whose value are of type enum, pymzn includes the enum in the output with a show function which fails on enumerated enums (that are not ranges like 1..10).
Here is a small example coming from the minizinc coursera course (temperature.mzn):

var int: cost;
array[int] of int: readings = [35,35,20,20,20];
int: start = 25;
int: n = length(readings);
set of int: HOUR = 1..n;
set of int: HOUR0 = 0..n;

array[HOUR] of var ACTION: choice;

enum ACTION = {heat, strong_heat, cool, strongly_cool, do_nothing};

array[HOUR0] of var int: temp;

array[ACTION] of int: adjust = [+1, 4, -2, -5, 0];
array[ACTION] of int: adjust_cost = [1, 5, 3, 9, 0];

constraint temp[0] = start;
constraint forall(h in HOUR) (temp[h] = (temp[h-1] + readings[h]) div 2  + adjust[choice[h]]);
constraint cost = sum(t in HOUR)(adjust_cost[choice[t]]);
constraint forall(h in HOUR) (temp[h] >= 25);
constraint forall(h in HOUR) (temp[h] <= 30);

solve minimize cost;

minizinc temperature.mzn leads to a solution while the following script fails with INCOMPLETE:

import pymzn
pymzn.minizinc("temperature.mzn", keep=True)

You can note the show(ACTION) in the output of the generated mzn file. When removing that part of the line, the model works fine.

[paolodragone]

Hi @Moisan, thank you for reporting this. This looks more likely to be a bug in MiniZinc. I can reproduce it with a much simpler model:

enum ACTION = {A1, A2};
array[1 .. 10] of var ACTION: choice;
solve satisfy;
output [show(ACTION), " ", show(choice)];

Running minizinc on this model gives me the same error I get with your model using PyMzn:

MiniZinc: evaluation error: not an array expression

It would be probably better if you could open an issue on libminizinc?

In the meantime I found a workaround: swapping the order of show(ACTION) (the enum) and show(choice) (the array) in the output statement makes minizinc (and pymzn) succeed. I'll implement a fix for the bug with this workaround in the next few days.

[Moisan]

I think that might be this issue. It seems to have been solved in the develop branch.

[paolodragone]

Ok, good to know! Thanks. I'll check whether it makes sense to make a fix in PyMzn in the near future then. I'll keep this issue open until either I make a fix using the workaround or the bug is fixed in MiniZinc.