Ray to Circle collision broken about cute_headers HOT 10 CLOSED

Thanks taking a look now !

from cute_headers.

Sure I'll add in a big fat comment right above the definition of c2Ray. Good idea!

from cute_headers.

Here's the rendering of your code:

All is working as intended! Except I bet you meant to normalize your ray direction :)

Try:

ray.d = c2Norm(c2V(100, 100));

Here's the code I used to do the rendering in the tinyc2 example demo:

void sro5hRayBug( )
{
	c2Ray ray;
	ray.p = c2V( 100.0f, 100.0f );
	ray.d = c2V( 100.0f, 100.0f );
	ray.t = 1.0f;

	c2Circle circle;
	circle.r = 30.0f;
	circle.p = c2V( 200.0f, 200.0f );

	DrawCircle( circle.p, circle.r );

	c2Raycast cast;
	if ( c2RaytoCircle( ray, circle, &cast ) )
	{
		c2v impact = c2Impact( ray, ray.t );
		c2v end = c2Add( impact, c2Mulvs( cast.n, 1.0f ) );
		tgLineColor( ctx, 1.0f, 0.2f, 0.4f );
		tgLine( ctx, impact.x, impact.y, 0, end.x, end.y, 0 );
	}

	c2v end = c2Add( ray.p, c2Mulvs( ray.d, ray.t ) );
	tgLineColor( ctx, 1.0f, 1.0f, 1.0f );
	tgLine( ctx, ray.p.x, ray.p.y, 0, end.x, end.y, 0 );
}

from cute_headers.

In your function sro5hRayBug you don't seem to normalize ray.d, is that a typo?

from cute_headers.

Nope not a typo. The ray does not have to be normalized, but it seemed like you wanted it to be judging by your post. Here's a pretty good link on vectors and normalization, take a peek if any of the math here is confusing. Rays are just vectors for the most part :)

https://www.khanacademy.org/computing/computer-programming/programming-natural-simulations/programming-vectors/a/vector-magnitude-normalization

from cute_headers.

Well I feel like I understand the math (normalization creates a Vector of length 1 with the same direction) :)
But I really don't understand why my code gives me the wrong impact point.

from cute_headers.

Why are you using ray.t in c2Impact, wouldn't that just give you the endpoint of the ray rather than the actual impact?

from cute_headers.

What really confuses me is that if I take the code from this issues and replace the circle with an AABB the raycast gives me the correct impact. So there has to be some difference between c2RaytoCircle and c2RaytoAABB if one time cast.t is correct even if ray.d is not normalized and the other time it isn't

from cute_headers.

The impact point is off because the function c2RaytoCircle is yet again written with speed in mind. It's not really your fault, it's just confusing. Most other libraries will take extra care and effort to prevent this kind of thing from happening, but tinyc2 is much more "bare metal".

If we look here we can see a huge floating point value. It doesn't even have decimal points. That's because the function is taking big vectors and dotting them together, making a giant value for the disc variable. This giant value has such poor precision that the t value, calculated later, is visibly inaccurate. This is because large floating point numbers have very low precision.

To prevent this kind of thing from happening normalizing the Ray's direction vector results in a much smaller disc value, of only 900.003906. That is plenty small enough to give an accurate result.

c2RaytoAABB worked out OK because the algorithm inside is not very sensitive to the large ray.d inputs. The c2RaytoCircle function uses an algorithm that is very sensitive to the length of the ray's direction vector.

The solution is to normalize the ray's d vector and scale t by the length. In this way the magnitude of the old longer d vector is transferred to the t variable. In general the raycast functions in tinyc2 are not very sensitive to fairly large t values.

Something like this:

c2Ray ray;
ray.p = c2V( 100.0f, 100.0f );
ray.d = c2Norm( c2V( 100.0f, 100.0f ) );
ray.t = c2Length( c2V( 100.0f, 100.0f ) );

from cute_headers.

Ok now I understand it, thanks for the help.

Maybe a notice along the lines of "Keep in mind that direction vectors with a huge length can introduce major inaccuracy in the result. As a fix it is best to normalize the direction of the ray" in a comment above the function would be helpful for others?

from cute_headers.