Outline rendering glitches in wasm

[luca-della-vedova]

Hi and thanks for the great crate!

We are using the crate for our application, we insert a Flat SetOutlineDepth component and show the outline when the mouse is hovering.
While this looks great on a desktop (Linux, AMD64, integrated GPU with Vulkan) build:

desktop_view.webm

It looks very odd on wasm:

wasm_view.webm

The code is exactly the same, do you have any idea what might be happening?

[komadori]

The effect you're seeing is Z-fighting between the outline and the stencil. Since the fix for gfx-rs/naga#14, bevy_mod_outline uses a depth offset to separate the outline and stencil. For some reason, the required separation is much larger in WebGL than the native back-ends. I'm still investigating.

[komadori]

The issue is, due to gfx-rs/wgpu#4553, the floating point error in the depth values is much bigger with the WebGL backend than non-OpenGL backends, requiring a much bigger depth bias. I've added a GL-specific workaround to bevy_mod_outline which avoids the increased floating point error at the cost of some performance.

Released in 0.4.2.

[luca-della-vedova]

Wow I had a feeling the fix was tricky but didn't think it would require this depth of digging.
Thanks a lot for looking into it and releasing the fix so quickly!

[dannymcgee]

Hi there! Unfortunately it seems like this issue has cropped up again (or maybe my particular hardware setup is just immune to the previous fix). I forked this locally and was able to fix the issue by increasing the offset to a cartoonishly large amount, which doesn't seem to have introduced any negative side effects. (I can email you before & after screenshots if that would be helpful.)

This is the change I made:

bevy_mod_outline/src/outline.wgsl

Line 111 in 3740766

out.normalised_depth = 0.5 + 0.5 * (out_z / clip_pos.w);

    out.normalized_depth = 10.0 + 0.5 * (out_z / clip_pos.w);

Would be happy to open a PR if you'd like to incorporate it, but no hard feelings if you'd prefer to try a more robust solution or wait for an upstream fix from wgpu.

[komadori]

@dannymcgee I've opened a new issue, #26, for what I believe is the same issue you're experiencing. Please let me know if the screenshot on that issue resembles what you're seeing.

[dannymcgee]

Sorry I missed your update last week @komadori. Yes, that is exactly what I was seeing — thanks for the quick response!