Clarification for makePerspective docs

[parasyte]

https://github.com/greggman/webgl-fundamentals/blob/master/docs/module-webgl-3d-math.html#L1768-L1770

Note that near and far are not z coordinates, but rather they are distances along the negative z-axis.

IMHO, this could be more specific; If near and far are not coordinates, what are they?

I'm trying to understand the relationship of the frustum to clip space. From experimental testing, I can see that near fragments are always clipped somewhere between Z -1 and -2, regardless of how I set near in makePerspective, and I have been unable to find the far clipping plane, as even incredibly distant objects (Z -1e12) are still drawn (can be seen better when scaled appropriately).

I'm really just trying to get an understanding for how much play room I have within my 3D space. It seems like pretty much every number less than -2 will render.

In more concrete terms, when I use:

makePerspective(Math.PI / 2, width / height, 10, 50);

What is the expected valid range for the Z axis, before it exits clips space?

[greggman]

They are z-coordinates in the -z axis from the camera AFAIK. Sorry that comment in the code was written by another person ~6 years ago and I've just copied and pasted it ever since.

Here's an example

The sphere has a radius of 1 unit. It's center is at z = -5 which means it's closest point is at z = -4. Slide the zNear to just past 4 and you'll see it's clipped exactly as you'd expect. slide the zFar to something less than 5 and you'll see it's also clipped exactly where you'd expect. Of course it will actually start being clipped at 6 but you can't see the back of the sphere so you've got to take it < 5 to see zFar clip. Although, if you set zNear to ~4.5 then there will be a hole through the sphere and you can see it start clipping the back when zFar is under 6.

[greggman]

You might also find these answers useful

http://stackoverflow.com/a/28301213/128511

http://stackoverflow.com/a/21106656/128511

[parasyte]

The second answer is great! I also saw this article, which provides a graph of the expected coordinate mapping, near the end: https://unspecified.wordpress.com/2012/06/21/calculating-the-gluperspective-matrix-and-other-opengl-matrix-maths/

And this video touches on the subject at about 1:15: https://www.youtube.com/watch?v=Sfy6pB0susM

But empirically, I haven't been able to replicate the clipping shown in your example. (I need to identify a bug in my code somewhere.) Anyway, without going too off-topic for this ticket, I guess the docs here are incorrect that near and far are "not z coordinates".

[parasyte]

FYI, my bug was an obscure oversight (on my part). In my vertex shader, I was multiplying my vertex position by the matrix (this operation is not commutative). Whoops! No wonder my input coordinates were generating nonsensical renderings.

Hard to catch, and even harder to debug. Thanks for being my rubber duck on this one!