Riddle me this one batman

SynapticBytes

Hey Joe, here's a brain teaser for you.

So in a function, I have these lines calling my collision function

x=0
y=1
test = polygon + [polygon[0]] //close the polygon for line segment testing
collide = util.collision(polygon,[50, 100], 10)

Polygon is a list of points similar to this

My collision function, edited for just the essentials, looks like this

collision = function(polygon, centre, radius)
	for point in range(0,polygon.len-2)
		x1 = polygon[point][x]
		x2 = polygon[point+1][x]
		y1 = polygon[point][y]
		y2 = polygon[point+1][y]
		denominator = sqrt((y2-y1)^2 + (x2-x1)^2)
		distance = abs((y2-y1)*centre[x] - (x2-x1)*centre[y] + x2*y1-y2*x1)/denominator
	end for
end function

One loop through the collision function takes about 0.3 milliseconds.

If I change the third line of code above to this

test = polygon + polygon[0] //close the polygon for line segment testing

Each loop through the collision function now takes 66 - 70 milliseconds. I have my thoughts why, but I'd love to get an expert opinion.

Russell.

jstrout

Hmm, interesting. In the changed case, you're just adding the numbers 226 and 482 to your list (which otherwise contains xy pairs). And I think you're going to hit the first of your numbers, which would be at polygon.len-2, in your for loop. And then you'd be trying to index that number with [0] and [1], which I would expect to throw a runtime error.

My vague guess at the moment is that this error is somehow messing up your time measurements. Though why it's not just aborting the program in a very obvious way, I have no idea.

SynapticBytes

Yeah I’m not seeing any error. I’m guessing it’s just too clever at avoiding problems. My thoughts are that a null reference is being returned when trying to access an index of a list that does’t exist and somehow that null is being handled without erroring, in all the implicit type conversion happening, and the internal management of what to do with null is just causing it to have a brain fade on every calculation.

Those single digit points, what happens internally when you reference say the number 3, with 3[0] and 3[1]? It’ll return null and then try to further process it, but what does it do to get null in the first place, and then subsequently process it through all the math?

jstrout

Yeah, when you index into a number, it actually throws a runtime exception, which ought to appear in Mini Micro as "can't index into this type." If that's not happening, then it means I'm catching that exception higher up and allowing the code to continue... which I really shouldn't do, but that could certainly explain the drastic difference in measured time.