Profiling

Built-in Profiler

The workspace includes a ring-buffer frame profiler that tracks cascade, layout, and paint timing.

Enabling

let mut devtools = Devtools::attach(&mut tree, true); // true = enable profiler

Or create a profiler directly:

tree.profiler = Some(Profiler::new());

Viewing Results

Press F9 to dump a summary to stderr:

Frame #123 — FullPipeline (60.0 fps)
  cascade    1.23ms ████
  layout     2.45ms ████████
  paint      0.89ms ███
  nodes: 456  quads: 234  glyphs: 1234  boxes: 89

Profiler Architecture

• Ring buffer: 240 frames (~4 seconds at 60fps)
• Scope-based: RAII guards with prof_scope! macro
• Zero-cost when disabled: expands to Option::map eliminated by compiler
• Alert threshold: only reports frames exceeding a configurable threshold

PipelineTimings (always available)

pub struct PipelineTimings {
    pub cascade_ms: f64,
    pub layout_ms: f64,
    pub paint_ms: f64,
}

let (_, _, timings) = paint_tree_returning_layout_profiled(...);
println!("Frame: {timings:?}");

PipelineCache

The PipelineCache classifies frames for incremental rendering:

pub enum PipelineAction {
    FullPipeline,        // Cascade + layout + paint
    PartialCascade,      // Re-cascade dirty nodes only
    LayoutOnly,          // Re-layout, then paint
    PatchFormControls,   // Patch form control state, repaint only
    RepaintOnly,         // Re-paint with existing layout
}

Frame classification considers tree.generation, cascade_generation, form_control_generation, and dirty_paths.

Performance Characteristics

Stage	Cold Frame	Incremental Frame
Cascade	O(n × rules)	O(dirty_nodes × rules)
Layout (block)	O(n)	O(dirty_subtree)
Layout (flex)	O(items²)	O(items) if container dirty
Layout (grid)	O(items × tracks)	O(items × tracks) if container dirty
Paint	O(draw_commands)	O(draw_commands)

Performance Tips

• Register fonts before the first cascade/layout
• Use PipelineCache to avoid redundant cascade on hover/focus changes
• Avoid frequent tree mutations that dirty large subtrees
• Preload images with tree.preload_asset() to avoid blocking layout
• Set reasonable asset_cache_ttl to avoid redundant image re-decodes