Thread: Weird arithmetic failure on perfectly legit operation.

Here is my code. Its purpose is to test my method for generating pseudorandom numbers for my diamond-square algorithm terrain generator. It was supposed to be a proof-of-concept, to see whether the numbers it gave were random enough for my purposes.

Code:

#include <stdio.h>

int main () {
	unsigned int seed, r, h, i; // Declare all our delicious unsigned ints.
	printf ("Please input the seed, the range, the decrement (to the range), and the initial.\n"); // Just so the user knows what to input where.
	scanf ("%d, %d, %d, %d", &seed, &r, &h, &i); // Get the parameters from the user.
	int v = i + (((seed * (seed % 10) + 5) % r) - (r/2)); // Create our printending number, and give it a believable value.
	int n;
	for (n = 0; n <= 50; n++) {
		// To get the nth digit of integer i, use ((i % (10^n)) / (10^(n - 1))).
		printf ("%d ", v);
		h++;
		v = i + (((seed * ((seed % (10^n)) / (10^(n - 1))) + n) % (r - h)) - ((r - h)/2)); // Generate a new v, with a smaller range than last time.
	}
}

Here is the error Xcode gave me. Notice that I have my ints declared as ints, rather than unsigned ints. I did this to alleviate problems that might have occurred from comparing incompatible data types (although I didn't think that this was the problem). The code builds, and gets as far as the scans, which I complete with glee... and then it gives me some bull about EXC_ARITHMETIC. What's that supposed to be?

Look up pow. And yes, exc_i386_div = division by zero.

But why do they need to be doubles? D:

@mszegedy the problem is when you raise values to a certain power it would go out of the int range creating undefined behaviour. so preferably use the pow(x,y) provided by the stdlib header. hope it helps.

Thanks for the explanation! It doesn't exactly work yet: modulus is being highlighted as only taking doubles, which makes no sense, but thanks a lot! I think I know what I'll try, then...

Probably because the possible range of values is so large. Maybe you could cast back to int or long?
*Edit : Nyah Check beat me to the explanation

@patricio2626 i think using doubles or long values would be more precise since the modular aritmetic would be performed on the variables.

Urr, I got it to compile again but it still says I'm dividing by zero. I swear I'm not! I specifically raised initial n to 2 to combat this nonsense. I'm passing ints into modulus and everything... :/ urrgh...

Interesting. Would you mind posting the new code?

Okay, looking at your screenshot, maybe this is some kind of numeric overflow. Look at the value of tmpxp1, and look at the likely min value of int: –2147483647 – 1.

But it crashes before even a single iteration. It's not like it's 10000000000000 or anything. It's 100. I get the same error if I use 123 as the seed.

(If there's a simpler way to get the nth digit of something, feel free to share.)

Just curious, what is the value of tmpxp2 at that point in your debug?

> But, now that I think about it, must've meant bitwise AND.
Wrong again, it's Exclusive-Or (xor).

You could split your algorithm into many different lines, and/or try using inline asm. Both methods would help you debug where the actual problem is.