If something falls into the depths, the near world will be clipped eventually

[microaaron]

https://video.twimg.com/ext_tw_video/1519159159647719425/pu/vid/640x360/lMB3hiUIzGA8jQAs.mp4?tag=12

[andreasplesch]

:) Try a fixed zFar/zNear with the Viewpoint ?

[microaaron]

:) Try a fixed zFar/zNear with the Viewpoint ?

Yes, it's effective.

[microaaron]

Does it need to be fixed?

By the way, according to X3D standard, zFar/zNear should be named farDistance/nearDistance.

[andreasplesch]

That is new in the standard. But changing the names will break older scenes, and making both names work is a bit harder than it sounds.

X3D defines fixed distances. But for most situations dynamic, and automatic determinations are more robust.

[microaaron]

I found the key codes:
var zNearLimit = zfar / this._zRatio;
Increasing zfar leads to the increase of zNearLimit.

[andreasplesch]

I believe this is intentional for numerical precision reasons required for perspective projections. The ratio needs to stay reasonable, within a certain range.

[microaaron]

Maybe it's not a bug, but this phenomenon is often not as expected.

[microaaron]

What are the negative effects of large zFar and small zNear?

[andreasplesch]

mrdoob/three.js#11755

is related. A solution would be to add support for the infinite projection matrix.

[microaaron]

Thanks.

[microaaron]

is related. A solution would be to add support for the infinite projection matrix.

This seems to mean _zRatio is no longer meaningful, and inevitably leads to low precision.

[microaaron]

As zfar increases, the precision decrease gradually converges. When zfar approaches infinity, the infinite projection matrix is obtained.

new x3dom.fields.SFMatrix4f(
        f / aspect, 0, 0, 0,
        0, f, 0, 0,
        0, 0, -1, -2 * near,
        0, 0, -1, 0
    );

Obviously, the precision is only affected by znear at this time.

[microaaron]

[microaaron]

I got it, the original intention of the design should be to improve distant details by increasing znear.
The infinite projection matrix does not improve the precision problem.The value of znear still requires trade-offs.
There is an improved way for webgpu: reversedZ.
But this is not suitable for webgl.