:has() subtree benchmark

:has() subtree benchmark

Compare 0, 1 or 2 :has() checks against a fresh tree with a hidden subtree.

Chain depth Descendants below the matching child. Extra leaves per level Add width without increasing chain depth. Preceding sibling elements 511 preserves the original 512-element group. Selectors Custom :has() count Used only for “Custom count”; all match the same child. Measured repeats per mode Warmup repeats per mode Recorded separately; excluded from summaries. Forced evaluation Pause between trials (ms) Not included in measured times. Stop after a slow trial (ms) Construction + attachment + evaluation + removal. 0 disables. Build / profile label Copy exact version and build ID from about:support.

Ready.

Results

Times in milliseconds. Cells show median [minimum–maximum]. Warmups are excluded.
Batch	:has count	Depth / leaves	Elements	Samples	Build DOM	Attach	Forced evaluation	Remove	Evaluation mode

Every trial

Batch	Phase	Repeat	:has count	Elements	Build DOM	Attach	Evaluation	Remove	Color check

Notes

Two checks use #subject:has(.first):has(.second). Both match the first child; the second lookup triggers subtree-filter construction.
Evaluation includes the browser work needed for the style read and optional layout flush. DOM creation, attachment and removal are timed separately.
Keep this tab in the foreground and close DevTools for timing. Browser timer precision and GC affect results.
The fix removes recursion. It still visits the subtree. Crash thresholds depend on stack size and style-thread scheduling; this page does not change Firefox preferences.
URL settings: depth, leaves, siblings, mode (compare, 0, 1, 2, custom), count, repeats, warmups, flush (layout-color or color), pause, slow, build, stress (1 or 0). Settings load without running a trial.

Test document: