perf(handler): eliminate repeated render scans and lookups

[Mohamed Mansour (mohamedmansour)]

Why

Large component-list renders repeated two protocol-invariant or already-completed operations on the request hot path: identical inline CSS was rescanned for </style for every instance, and state-backed dynamic component attributes resolved the same path once for HTML output and again for child-scope capture.

What changed

• Classify component and bundled inline style resources once when the immutable runtime Protocol loads.
• Directly write known-safe CSS while preserving the existing escape writer for unsafe CSS.
• Reuse a successful immutable-state attribute lookup for both HTML output and child-scope capture, while retaining the existing resolver fallback for owned, synthetic, missing, and streaming values.
• Share protocol-owned CSS metadata across buffered and streaming render contexts.
• Cover detector/writer parity, component and bundled-chunk cache classification, resolver provenance parity, and escaped borrowed attribute output.

Architecture

Both optimizations remain inside webui-handler and preserve existing ownership boundaries. Runtime-only metadata stays on immutable Protocol, not the protobuf wire model. Attribute lookup reuse preserves the established precedence across loop bindings, borrowed locals, owned locals, and global state. No public API, protocol, rendering semantics, or dependencies change.

Performance

Measured with:

cargo bench -p microsoft-webui --bench contact_book_bench -- 'contact_book_contacts_render/contacts/1000'

Metric	Before	After	Change
Criterion mean render time	2.7039 ms	1.4830 ms	-44.5%
Render throughput	857.66 MiB/s	1.5271 GiB/s	+80.1%
Render/1000 P50	2.62 ms	1.48 ms	-43.5%
FAST Render/1000 P50	3.66 ms	2.49 ms	-32.0%
Output bytes	2,431,665	2,431,665	unchanged

The attribute lookup reuse alone improved the post-CSS-cache contact benchmark by 16.0% and the focused nested-component benchmark by 7.5%.

Validation

• cargo xtask check
• Independent review of cache lifecycle, Style/Module/Link behavior, buffered/streaming parity, resolver precedence, provenance, lifetimes, synthetic values, and escaping

[Copilot]

Pull request overview

This PR improves handler render performance by caching which inline style resources require </style-escaping at Protocol load time, so safe CSS can be written directly without rescanning identical bytes on every render.

Changes:

• Added a protocol-owned HashSet cache of style resources whose CSS contains a case-insensitive </style sequence.
• Updated the <style> emission path to use the cached classification: write raw CSS for known-safe resources, and fall back to the existing escape writer for unsafe resources.
• Introduced a shared detector (style_text_needs_escape) and added tests for detector/writer parity and protocol cache classification.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
crates/webui-handler/src/streaming/session.rs	Plumbs the protocol-owned escape cache into streaming render context initialization.
crates/webui-handler/src/streaming/inventory.rs	Updates streaming inventory test context initialization to include the new escape cache reference.
crates/webui-handler/src/route_handler.rs	Stores and builds the protocol-owned escape classification cache; adds a targeted unit test.
crates/webui-handler/src/lib.rs	Uses the cache to select raw vs escaped CSS writing; adds the cache field to the render context.
crates/webui-handler/src/html_encode.rs	Adds a fast </style detector shared by both the writer and the protocol classification; adds parity tests.

---

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