perf: reduce framework hydration allocations

[Mohamed Mansour (mohamedmansour)]

Summary

• allocate the condition resolver only for components that evaluate a runtime condition
• share the existing empty-array sentinel for absent text, attribute, condition, and repeat metadata in both client wiring and SSR hydration
• clear populated binding arrays in place while guarding the shared empty sentinel

Performance

Measured in Chromium with the direct streaming-browser-bench lazy-hydration matrix at 10, 100, and 1,000 elements. Each arm ran 60 samples in counterbalanced baseline → branch → branch → baseline order with forced-GC CDP heap samples. Both source trees included the same PR #475 runtime fix and differed only in template-element.ts.

Mode	Elements	Baseline retained heap	Branch retained heap	Delta
eager	10	214.3 KiB	214.5 KiB	noise (+0.1%)
eager	100	335.5 KiB	326.8 KiB	-2.6%
eager	1,000	1,378.0 KiB	1,244.6 KiB	-9.7% (-133.4 KiB)
lazy-hydrate	1,000	635.1 KiB	588.1 KiB	-7.4% (-47.0 KiB)
lazy-render	1,000	652.0 KiB	605.3 KiB	-7.2% (-46.7 KiB)

A separate object-identity probe confirmed exactly four avoided allocations per element: one resolver closure and three empty binding arrays, or 40 / 400 / 4,000 allocations at 10 / 100 / 1,000 elements. Baseline roots retained N resolver functions and 3N unique empty arrays; the branch retained no resolver on condition-free roots and reused one pre-existing empty sentinel.

End-to-end custom-element definition medians remained effectively flat:

• eager 1,000: 26.95 ms → 27.25 ms (+0.30 ms)
• lazy-hydrate 1,000: 21.70 ms → 21.75 ms (+0.05 ms)
• lazy-render 1,000: 21.45 ms → 21.55 ms (+0.10 ms)

Bundle size changes by +83 B gzip for eager and +100 B gzip for lazy modes. Exact body.innerHTML SHA-256 values matched between baseline and branch at all three scales.

Validation

• pnpm --dir packages/webui-framework test:unit
• focused Playwright fixtures: 82 passed, 2 skipped
• cargo xtask check

[Copilot]

Pull request overview

This PR reduces per-element allocations during WebUI framework hydration/wiring by avoiding unnecessary resolver closure creation and by reusing a shared empty-array sentinel for binding arrays when a template block has no bindings of a given kind.

Changes:

• Added a bindingArray() helper to reuse EMPTY_ARR for zero-length binding collections while still allocating mutable arrays when bindings exist.
• Deferred allocating the condition resolver closure until the first runtime condition evaluation via $conditionResolver().
• Avoided unnecessary .length = 0 writes during instance teardown by clearing binding arrays only when they’re non-empty.

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.