I’ve encountered a very bizare & unexpected issue! :))
I have a scene where I’m using a CascadedShadowGenerator, and everything works flawlessly on localhost. However, when I build the code and upload it to the server, I get a WebGL warning, and the shadows appear cut off (as if the shadowZ is very small). Interestingly, if I use a ShadowGenerator, everything works well both locally and in the development environment.
I have attached a few images to illustrate the issue.
I’ve tried to reproduce this on the pg, but without success so far. I’m still working on it.
Any insights are welcome! @Evgeni_Popov maybe
Thank you!
Understood, let me try to reproduce this on an online editor that supports ES6. It is quite hard for me to provide details about the packager and optimizations at the package level because the entire 3D stuff is built in a different repository. It’s built as a package and used in another repository, which is also built and uploaded, making it not too easy to track.
Regarding filters, I’m using PCF (you can check the pg, it’s a simplified version of my scene). I’ll try without and let you know. I’ll also give WebGPU a try. I’ll be back with updates
It looks like webGPU doesn’t like more from my code
I don’t know why there is a connection between RenderTargetTexture and ShadowGenerator, we’ll find out
You where right!
So I’m using webGPU, and no filters and I got this on localhost and the scene freeze. I think it is an infinite loop of warnings, because of this the scene freeze
There’s something really strange here, the first error is a uniformity error, but we’d also get it in a normal PG if there was such a problem in the shader code (and you would also have it locally)… Also, I can see the use of the round function in the error message, but I looked at our shadow code and couldn’t find any occurrence of round (but maybe Tint adds it itself)…
Your second message shows that you’re using the viewer (SceneManager comes from the viewer): perhaps there’s a problem with this package?
Would you have a direct link we could check? I think it would be difficult to find the problem otherwise…