three.js
three.js copied to clipboard
mrdoob
Reverse-Z state problem
Description
Reverse Z PR not fully correctly implement reverse-Z.
Problem in state: https://github.com/mrdoob/three.js/blob/0c7cf78a64a724c24fcca51500b8f3d1b9931af5/src/renderers/webgl/WebGLCapabilities.js#L100
This flag applied in capabilities that persistent after initialisation, that is wrong due to context sharing.
ThreeJS state can be flushed for some specific cases like a 1 contex for N-renderers or libs. For ex: Three + PIXI, Three + WebXR emulator, Three+potree.
WebGLState has reset for that: https://github.com/mrdoob/three.js/blob/0c7cf78a64a724c24fcca51500b8f3d1b9931af5/src/renderers/webgl/WebGLState.js#L1152
Because clipControl is context-related three also should reset this state and change it when needed back
@CodyJasonBennett can you apply some changes and move call of clipControl into WebGLLState instead of WebGLCapabilities?
@Mugen87 what you think about that?
Reproduction steps
Code
Live example
Screenshots
No response
Version
r169
Device
No response
Browser
No response
OS
No response
Also we have state problem with clearDepth:
Reverse Z upgrade depth inside renderer: https://github.com/mrdoob/three.js/blob/0c7cf78a64a724c24fcca51500b8f3d1b9931af5/src/renderers/WebGLRenderer.js#L597
BUT Three has state for that: https://github.com/mrdoob/three.js/blob/0c7cf78a64a724c24fcca51500b8f3d1b9931af5/src/renderers/webgl/WebGLState.js#L190 ReversedZ will broke state upgrade, because clear value will always 0 or 1.
This flag applied in
capabilitiesthat persistent after initialisation, that is wrong due to context sharing.ThreeJS state can be flushed for some specific cases like a 1 contex for N-renderers or libs. For ex: Three + PIXI, Three + WebXR emulator, Three+potree.
How could this happen? Do they have a different WebGL implementation than the browser or are still using WebGL 1?
Because clipControl is context-related three also should reset this state and change it when needed back
I can agree with this and your changes in #29579. I would much prefer if renderer.capabilities had some hint as to what the context supports. Not sure anyone is using that though since this feature is so new.
How could this happen? Do they have a different WebGL implementation than the browser or are still using WebGL 1?
Simple case: Potree is outdated library that use parts of threejs with manual state switch, it not know about reverse-Z at all. For potree to important use valid depth info because it use postprocess for resolve dom lighting.
https://github.com/potree/potree/blob/develop/src/materials/shaders/edl.fs#L38
But in general a reverseZ must have for BIMs ( cad system ), because BIMs have billions of near placed polygons on non-normalized space( about 10000x10000 metres with step in 10sm ). Without reverseZ we to tired use hacks that reducing z-fights.
