Fix half-pixel offsets in scaling renderer (#231)

* Fix half-pixel offsets in scaling renderer

- This bug was very subtle. It can be hard to notice!
- Context: When the scaling renderer transformation matrix is created,
  it needs to center the image within the border.
- The previous code always evaluated the translation to (0, 0)
- This PR makes half-pixel adjustments to the translation when needed,
  making it impossible to rasterize the texture on a half-pixel
  boundary.
- I spotted this while working on pixel-aspect-ratio support, but it can
  most easily be witnessed in the `conway` example by grabbing the top
  or bottom resize handle on the window and slow dragging it up and down
  by 1-pixel-at-a-time. When you hit a half-pixel bug the entire texture
  will change slightly; some pixels will become hidden or duplicated.

* Bump MSRV

.github/workflows/ci.yml

@@ -13,7 +13,7 @@ jobs:
         rust:
           - stable
           - beta
-          - 1.52.0
+          - 1.53.0
     steps:
       - name: Checkout sources
         uses: actions/checkout@v2
@@ -73,7 +73,7 @@ jobs:
         rust:
           - stable
           - beta
-          - 1.52.0
+          - 1.53.0
     steps:
       - name: Checkout sources
         uses: actions/checkout@v2

MSRV.md

@@ -2,6 +2,7 @@
 
 | `pixels` version | `rustc` version |
 |------------------|-----------------|
+| `0.9.0`          | `1.53.0`        |
 | `0.8.0`          | `1.52.0`        |
 | `0.7.0`          | `1.52.0`        |
 | `0.6.0`          | `1.52.0`        |

src/renderers.rs

@@ -243,8 +243,8 @@ impl ScalingMatrix {
         // Create a transformation matrix
         let sw = scaled_width / screen_width;
         let sh = scaled_height / screen_height;
-        let tx = (texture_width / screen_width - 1.0).max(0.0);
+        let tx = (screen_width / 2.0).fract() / screen_width;
-        let ty = (1.0 - texture_height / screen_height).min(0.0);
+        let ty = (screen_height / 2.0).fract() / screen_height;
         #[rustfmt::skip]
         let transform: [f32; 16] = [
             sw,  0.0, 0.0, 0.0,