Answer "have I asked this child?" in one read

A container's draw asked it once per child by searching the list of children
it had added so far, and four other per-child steps searched a list too, so
one draw cost the square of its children: 70% of a 1,600-child redraw was
those searches. A draw takes a DrawId and leaves it on every widget it asks
about; one note per widget is enough because the handle a container holds a
child by cannot be cloned. tests/children_cost.rs is the rig that shows it,
and it is the only one here that varies width: 3.680 ms to 0.811 ms at 1,600
children, and flat per child.

Beside it, the rest of the fourteenth sweep of #19: a mask's rectangle
resolved once per fragment instead of once per instance, which takes the
storage buffers out of the fragment stage and is 8.8x on a screenful of
deeply nested clips; TextBuffer::shape copying its attrs before the check
that would not need them, which allocated once per named-family text per
frame; a should_panic test on a debug assertion that made cargo test
--release fail; Fixed::div, reached only by its own test; Moves::remove
re-uploading an array it cannot have changed; and two comments the branch
itself falsified.

docs/LAYOUT_LOG.md has all eight with their measurements, the five things
looked at and left, and what was verified.

tests/chain_cost.rs

@@ -8,13 +8,15 @@
 //! submitted and waited on, so this is the GPU's cost and not the recording
 //! loop's -- which is what `draw_cost.rs` measures instead.
 //!
-//! The instances are two pixels wide so that vertex work dominates; a chain
-//! walk that does not show up against small quads will not show up against
-//! anything.
+//! Two fixtures, because the walk happens in both stages. `chain_cost_by_depth`
+//! draws instances two pixels wide so that vertex work dominates; a walk that
+//! does not show up against small quads will not show up against anything.
+//! `mask_cost_by_depth` draws one screenful through a mask instead, which is
+//! where a walk in the fragment stage would show and nowhere else.
 
 use iris::prelude::*;
 use iris_core::{
-    Len, MaskIdx, MoveIdx, PrimitiveInst, RectPrimitive, UiData, UiRegion, UiRenderNode,
+    Len, Mask, MaskIdx, MoveIdx, PrimitiveInst, RectPrimitive, UiData, UiRegion, UiRenderNode,
     UiRenderState, UiSpan,
 };
 use wgpu::{Color as GpuColor, *};
@@ -45,8 +47,17 @@ fn gpu() -> Option<(Device, Queue, f32)> {
     Some((device, queue, period))
 }
 
+/// Which stage the fill puts the work in: many small quads, where a walk per
+/// vertex is what shows, or one screenful of masked rows, where a walk per
+/// fragment would.
+#[derive(Clone, Copy)]
+enum Fixture {
+    Quads,
+    Masked,
+}
+
 /// A chain `depth` slots long, and instances that all resolve through its end.
-fn fill(ui: &mut UiData, render: &mut UiRenderState, depth: usize) {
+fn fill(ui: &mut UiData, render: &mut UiRenderState, depth: usize, fixture: Fixture) {
     let kind = ui.primitives.kind::<RectPrimitive>();
     let id = ui.widgets.add_strong(Rect::new(UiColor::WHITE)).id();
 
@@ -56,20 +67,51 @@ fn fill(ui: &mut UiData, render: &mut UiRenderState, depth: usize) {
     }
 
     let px = |v: f32| Len::px(v);
-    for i in 0..INSTANCES {
-        let x = (i % (SIZE as usize / 2)) as f32 * 2.0;
-        let y = (i / (SIZE as usize / 2)) as f32;
+    let rows = SIZE as usize;
+    let mask_idx = match fixture {
+        Fixture::Quads => MaskIdx::NONE,
+        // Its own chain as long as the instances', since a viewport sits as
+        // deep in the tree as the content it clips.
+        Fixture::Masked => {
+            let idx = ui.masks.push(Mask {
+                region: UiRegion::FULL,
+                move_idx: slot,
+            });
+            // Nothing frees it here, but the owner's reference is what a real
+            // one is kept alive by.
+            ui.masks.push_ref(idx);
+            idx
+        }
+    };
+    let instances = match fixture {
+        Fixture::Quads => INSTANCES,
+        Fixture::Masked => rows,
+    };
+    for i in 0..instances {
+        let region = match fixture {
+            Fixture::Quads => {
+                let x = (i % (rows / 2)) as f32 * 2.0;
+                let y = (i / (rows / 2)) as f32;
+                UiRegion::new(
+                    UiSpan::new(px(x), px(x + 2.0)),
+                    UiSpan::new(px(y), px(y + 1.0)),
+                )
+            }
+            // A full row each, so one screenful of fragments goes through the
+            // mask and the vertex stage is four corners per row.
+            Fixture::Masked => UiRegion::new(
+                UiSpan::new(px(0.0), px(SIZE as f32)),
+                UiSpan::new(px(i as f32), px(i as f32 + 1.0)),
+            ),
+        };
         render.layers.write(
             0,
             PrimitiveInst {
                 kind,
                 id,
                 primitive: RectPrimitive::color(UiColor::WHITE),
-                region: UiRegion::new(
-                    UiSpan::new(px(x), px(x + 2.0)),
-                    UiSpan::new(px(y), px(y + 1.0)),
-                ),
-                mask_idx: MaskIdx::NONE,
+                region,
+                mask_idx,
                 move_idx: slot,
             },
         );
@@ -77,28 +119,15 @@ fn fill(ui: &mut UiData, render: &mut UiRenderState, depth: usize) {
 }
 
 /// Nanoseconds the pass took on the GPU, best of `BATCHES`.
-fn pass_cost(device: &Device, queue: &Queue, period: f32, depth: usize) -> f64 {
+fn pass_cost(device: &Device, queue: &Queue, period: f32, depth: usize, fixture: Fixture) -> f64 {
     let format = TextureFormat::Bgra8Unorm;
     let mut node = UiRenderNode::new(device, &gpu::config(format, SIZE));
     let mut ui = UiData::default();
     let mut render = UiRenderState::new();
-    fill(&mut ui, &mut render, depth);
+    fill(&mut ui, &mut render, depth, fixture);
     node.update(device, queue, &mut ui, &mut render);
 
-    let target = device.create_texture(&TextureDescriptor {
-        label: Some("chain cost"),
-        size: Extent3d {
-            width: SIZE,
-            height: SIZE,
-            depth_or_array_layers: 1,
-        },
-        mip_level_count: 1,
-        sample_count: 1,
-        dimension: TextureDimension::D2,
-        format,
-        usage: TextureUsages::RENDER_ATTACHMENT,
-        view_formats: &[],
-    });
+    let target = gpu::target(device, format, SIZE, false);
     let view = target.create_view(&TextureViewDescriptor::default());
 
     let queries = device.create_query_set(&QuerySetDescriptor {
@@ -178,17 +207,15 @@ fn pass_cost(device: &Device, queue: &Queue, period: f32, depth: usize) -> f64 {
     best
 }
 
-#[test]
-#[ignore = "measurement, not a check"]
-fn chain_cost_by_depth() {
+fn by_depth(fixture: Fixture, instances: usize) {
     let Some((device, queue, period)) = gpu() else {
         println!("no gpu with timestamps; nothing measured");
         return;
     };
-    println!("{INSTANCES} instances, {SIZE}x{SIZE}, best of {BATCHES} batches");
+    println!("{instances} instances, {SIZE}x{SIZE}, best of {BATCHES} batches");
     let mut base = None;
     for depth in [1, 2, 4, 8, 16, 32, 64] {
-        let ns = pass_cost(&device, &queue, period, depth);
+        let ns = pass_cost(&device, &queue, period, depth, fixture);
         let base = *base.get_or_insert(ns);
         println!(
             "depth {depth:>3}: {:>9.1} us   {:+6.1}% against depth 1",
@@ -197,3 +224,19 @@ fn chain_cost_by_depth() {
         );
     }
 }
+
+#[test]
+#[ignore = "measurement, not a check"]
+fn chain_cost_by_depth() {
+    by_depth(Fixture::Quads, INSTANCES);
+}
+
+/// One screenful of rows, every one clipped by a mask whose own chain is that
+/// deep. What this says that the quads cannot is whether a mask costs the walk
+/// once per instance or once per fragment: at a screenful of fragments per
+/// chain, the second is the difference between these two tables.
+#[test]
+#[ignore = "measurement, not a check"]
+fn mask_cost_by_depth() {
+    by_depth(Fixture::Masked, SIZE as usize);
+}