Bugs related to division by zero

[Martin Sjölund]

class A
  constant Real x = 0;
  Real r = x/x;
end A;

Gives
r = 0.0
Both wrong result, and no assertion added.

class A
  Real x = 0;
  Real r = x/x;
end A;

Gives
r = 1.0
No assertion is added.

[Martin Sjölund]

Note: 1.0 / 0.0 should become inf, 0.0 / 0.0 nan (if anything)

[Per Östlund]

The first example now gives an error and fails in r25153, instead of giving an error and continuing. The second example still behaves the same. It could be fixed by removing the x/x = 1 rule in ExpressionSimplify, but I'm not sure what that will affect. Interesting to note is that Dymola says r = 1.0 in both cases though.

[anonymous]

I don't think equating 1.0/0.0 to inf (i.e. +inf) good.

In fact, the limit of 1/x for x->0 is +inf or -inf depending on whether we consider the right or left limit.
For simulation purposes, I think that Dymola's way to is ok: lim (x/x) for x->0 is actually 1.0.

Replying to sjoelund.se:

Note: 1.0 / 0.0 should become inf, 0.0 / 0.0 nan (if anything)

[Francesco Casella]

Replying to perost:

The first example now gives an error and fails in r25153, instead of giving an error and continuing. The second example still behaves the same. It could be fixed by removing the x/x = 1 rule in ExpressionSimplify, but I'm not sure what that will affect. Interesting to note is that Dymola says r = 1.0 in both cases though.

I guess this behaviour makes sense. If x is a constant and is equal to zero the model doesn't make sense. If x is a Real variable (thus in general time-varying) it makes sense to consider x/x = 1 also when x approaches zero.

If there are objections please re-open the ticket and make a concrete proposal