Different FZN files

[vdi...@gmail.com]

I noticed that the Windows and Linux version of minizinc 2.6.0 sometimes produce different fzn files for the same model.Does that harm and can the outcome also different?

[Michael Marte]

Hi,

it depends on the differences.

Some changes (like ordering of constraints or ordering of variables in a clause) are semantically irrelevant but other changes should raise your eye-brows.

If you conclude that there is something wrong, then raise an issue on GitHub.

Yesterday, I found the following issue: https://github.com/MiniZinc/libminizinc/issues/563
In this case, different runs on the same system yield different FlatZinc outputs. The differences are semantically irrelevant but might affect the runtime behavior of certain solvers and hence affect results.

It would certainly be helpful if MiniZinc was deterministic, also cross-platform.

Cheers,

Michael