fix: shutter sound before the window shadow; faster vertical blur

src/main.rs

@@ -221,7 +221,9 @@ fn capture(mode: Mode, common: Common) -> Result<()> {
     // Thumbnails from earlier captures must not end up in this one.
     let mut hide = ipc::HideGuard::new(&ipc::socket_path());
 
-    let (shots, clip, source): Shots = match mode {
+    // Window shots get their (slow) shadow after the shutter sound.
+    let mut windowed = false;
+    let (mut shots, clip, source): Shots = match mode {
         Mode::Screen { all } => {
             let focused = focused_output(&outputs);
             let source = outputs[focused].geom.name.clone();
@@ -246,11 +248,15 @@ fn capture(mode: Mode, common: Common) -> Result<()> {
                 }
                 region::Choice::Window => {
                     drop(frames);
-                    window_shot(&outputs, cursor, config.capture.window_shadow)?
+                    windowed = true;
+                    window_shot(&outputs, cursor)?
                 }
             }
         }
-        Mode::Window => window_shot(&outputs, cursor, config.capture.window_shadow)?,
+        Mode::Window => {
+            windowed = true;
+            window_shot(&outputs, cursor)?
+        }
         Mode::Zoom => {
             let focused = focused_output(&outputs);
             let frame = wl.capture(&outputs[focused..=focused], cursor)?.remove(0);
@@ -267,6 +273,11 @@ fn capture(mode: Mode, common: Common) -> Result<()> {
         }
     };
     sound::play(&config.sound, sound);
+    if windowed && config.capture.window_shadow {
+        for (image, _) in &mut shots {
+            *image = shadow::add(image);
+        }
+    }
     drop(wl);
 
     let target = Target::from_args(common.output, common.clipboard_only);
@@ -293,9 +304,8 @@ fn capture(mode: Mode, common: Common) -> Result<()> {
 
 /// A window shot, shaped like the other modes' results. The preview goes to
 /// the window's output, or niri's focused one if that is unknown.
-fn window_shot(outputs: &[wayland::Output], cursor: bool, shadow: bool) -> Result<Shots> {
+fn window_shot(outputs: &[wayland::Output], cursor: bool) -> Result<Shots> {
     let (image, output) = window::capture(cursor)?;
-    let image = if shadow { shadow::add(&image) } else { image };
     let source = output.unwrap_or_else(|| outputs[focused_output(outputs)].geom.name.clone());
     Ok((vec![(image, None)], 0, source))
 }

src/shadow.rs

@@ -19,7 +19,7 @@ pub fn add(image: &RgbaImage) -> RgbaImage {
             alpha[(sy * cw + sx) as usize] = p.0[3] as f32 / 255.0;
         }
     }
-    // Three box blurs approximate a Gaussian of about BLUR / 2 sigma.
+    // Three box blurs approximate a Gaussian (visible extent about BLUR px).
     let r = (BLUR / 3).max(1) as usize;
     for _ in 0..3 {
         alpha = box_blur(&alpha, cw as usize, ch as usize, r);
@@ -69,17 +69,27 @@ fn box_blur(src: &[f32], w: usize, h: usize, r: usize) -> Vec<f32> {
             }
         }
     }
+    // Vertically with a running sum per column, walking rows so memory is
+    // read in order.
     let mut out = vec![0f32; src.len()];
-    for x in 0..w {
-        let at = |y: usize| tmp[y * w + x];
-        let mut sum: f32 = (0..=r.min(h - 1)).map(at).sum();
+    let mut sums = vec![0f32; w];
+    for y in 0..=r.min(h - 1) {
+        for (sum, v) in sums.iter_mut().zip(&tmp[y * w..(y + 1) * w]) {
+            *sum += v;
+        }
+    }
     for y in 0..h {
-            out[y * w + x] = sum / n;
+        for (o, sum) in out[y * w..(y + 1) * w].iter_mut().zip(&sums) {
+            *o = sum / n;
+        }
         if y + r + 1 < h {
-                sum += at(y + r + 1);
+            for (sum, v) in sums.iter_mut().zip(&tmp[(y + r + 1) * w..(y + r + 2) * w]) {
+                *sum += v;
+            }
         }
         if y >= r {
-                sum -= at(y - r);
+            for (sum, v) in sums.iter_mut().zip(&tmp[(y - r) * w..(y - r + 1) * w]) {
+                *sum -= v;
             }
         }
     }