Rendering issue at the boundary of the clear color

[allsey87]

Bevy version

0.5.0

Operating system & version

Arch Linux (5.12.9-arch1-1) running on a Toshiba laptop with Intel UHD Graphics 620 graphics.

What you did

Created a simple scene with a couple of meshes representing a floor and two boxes, although the problem was also recreated by just running the examples from bevy-inspector-egui.

What you expected to happen

The shadows from the egui widgets should blend nicely with the clear color set by AppBuilder's insert_resource(ClearColor(...)). Similarly, I suspect the meshes in my scene should not have this fine white outline where they transition from the mesh to the clear color.

What actually happened

The shadows from the egui widgets and the outline of the meshes interact strangely with the clear color. When changing the clear color dynamically using the widgets in the bevy-inspector-egui world example, it appears that the pixels in question are picking up the clear color but at a brightness several times higher than it should be.

Additional information

This problem is present either the X11 or Wayland features enabled on the Bevy crate (see below) and is present with/without egui/egui_bevy included in the project.

Screenshot from Wayland

Screenshot from X11

[CptPotato]

I have the hunch that this is because of transparency ending up in the swap chain image.
Even if the background is opaque, drawing something transparent on top of it will result in the render target becoming transparent (with default alpha blending ops).

Using a separate OP for blending the alpha channel might fix it if this is really the case.

[allsey87]

@CptPotato it's a bit out of my league to come up with a solution to this problem, but I can definitely run some tests if you have an idea as to where the problem lies.

[mockersf]

@allsey87 would you happen to have the code to reproduce this somewhere available?

[allsey87]

@mockersf for the white outline between the meshes and clear color, I can see it when I pull the main branch and run the 3d_scene example with cargo run --example 3d_scene:

Tomorrow, I will add an egui window on top of this and post the code and screenshots for it.

[allsey87]

By the way, I get this warning when I run Bevy examples and apps:

MESA-INTEL: warning: Performance support disabled, consider sysctl dev.i915.perf_stream_paranoid=0

Probably nothing, but I am making a note of it since it is related to graphics.

[lukors]

@allsey87 I believe that latest note is more of a configuration thing for your computer. At least for me, I got that same message for many applications.

Luckily I recorded what I did to do the configuration the message suggests:

1. sudo -i
2. echo dev.i915.perf_stream_paranoid=0 > /etc/sysctl.d/60-mdapi.conf
3. Rebooted computer
4. Verified the value with sysctl -n dev.i915.perf_stream_paranoid

Source: https://ubuntuforums.org/showthread.php?t=2457426

[bjorn3]

You don't have to do this if you don't actually want to measure the performance of rendering beyond fps or don't want to measure it at all.

[mockersf]

@mockersf for the white outline between the meshes and clear color, I can see it when I pull the main branch and run the 3d_scene example with cargo run --example 3d_scene:

Oh, definitely not what I'm seeing, even your clear color seems wrong, it should be #666666:

[mockersf]

It was mentioned on discord that this may be a case of #419. Could you check by playing with the msaa parameter?

[allsey87]

@mockersf you are correct. Setting Msaa { samples: 1 } solves the problem. When switching to Msaa { samples: 2 } or back to Msaa { samples: 4 }, the problem shows up again. So it seems to me that the clear color is not correct when MSAA samples is greater than one and that this edge (or in the case of the egui widgets, that incorrect shadow) that I am seeing is actually what the clear color should be.

Issue gfx-rs/wgpu#725 reports this problem, specifically calling out Linux and Intel graphics chips.

[DJMcNab]

Is this therefore a duplicate of #419?

Maybe we should add a comment on Msaa mentioning that it causes issues on certain configurations, pointing to #419.