Alpha chanel causes transparent window with glium backend under Wayland

[alvinhochun]

I believe this is related to glium/glium#1678. I'm absolutely not familiar with how OpenGL surfaces should interact with Wayland, but what I gathered is that the resulting alpha channel will affect the compositing of the window.

The glium backend uses the glium::Blend::alpha_blending() alpha blending function (

conrod/backends/conrod_glium/src/lib.rs

Line 294 in 351367e

let blend = glium::Blend::alpha_blending();

) in which the alpha channel is calculated by srcAlpha * srcAlpha + dstAlpha * (1 - srcAlpha) (https://github.com/glium/glium/blob/776413b4eb8da16a4e9801feb27754e400e7d25a/src/draw_parameters/blend.rs#L28). To me this seems really wrong. I believe the usual way of combining the alpha channels is srcAlpha + dstAlpha * (1 - srcAlpha), which means the code should be changed into this:

diff --git a/backends/conrod_glium/src/lib.rs b/backends/conrod_glium/src/lib.rs
index d58bd56..8e2d7eb 100644
--- a/backends/conrod_glium/src/lib.rs
+++ b/backends/conrod_glium/src/lib.rs
@@ -291,7 +291,18 @@ pub fn program<F>(facade: &F) -> Result<glium::Program, glium::program::ProgramC
 
 /// Default glium `DrawParameters` with alpha blending enabled.
 pub fn draw_parameters() -> glium::DrawParameters<'static> {
-    let blend = glium::Blend::alpha_blending();
+    use glium::{Blend, BlendingFunction, LinearBlendingFactor};
+    let blend = Blend {
+        color: BlendingFunction::Addition {
+            source: LinearBlendingFactor::SourceAlpha,
+            destination: LinearBlendingFactor::OneMinusSourceAlpha,
+        },
+        alpha: BlendingFunction::Addition {
+            source: LinearBlendingFactor::One,
+            destination: LinearBlendingFactor::OneMinusSourceAlpha,
+        },
+        constant_value: (0.0, 0.0, 0.0, 0.0),
+    };
     glium::DrawParameters { multisampling: true, blend: blend, ..Default::default() }
 }
 

However I do wonder if this should be fixed in glium by changing glium::Blend::alpha_blending().

The new blending function matches that of of the wgpu backend, which does not exhibit this issue.

The gfx backend, while it to does not appear to exhibit this issue, uses gfx::preset::blend::ALPHA which is srcAlpha + dstAlpha, so it is also incorrect.