Implements a grapheme cluster iterator over UTF-8 byte slices with SIMD-accelerated whitespace skipping. Uses wasm32 SIMD intrinsics (v128_load, i8x16_eq) to process 16 bytes at a time for fast whitespace detection. Provides take_until, take_until_one_found, and skip_whitespace methods used by the parser. Demonstrates non-obvious skill in leveraging SIMD for text processing within a Rust-WASM context.

The main JS/TS entry point defining the SAXParser class and data decoders (Tag, Text, Attribute, ProcInst). Handles WebAssembly instantiation, manages a shared memory buffer, and provides both a callback-based eventHandler and an async generator for streaming. Includes a neat approach to memory management: it caches a Uint8Array view of WASM memory and re-creates it when memory grows. The event constructor lookup table avoids per-event branching. The class is well-documented with JSDoc and is the primary reusable component for integrating the parser into JS projects.

Implements the core SAX parser state machine (~1400 lines) handling all XML constructs: tags, attributes (various quoting styles), JSX expressions, CDATA, comments, doctype declarations, processing instructions, and BOM. Uses a 32-state enum and processes grapheme clusters one at a time via the GraphemeClusters iterator. Supports streaming via fragment buffering when chunks break in the middle of syntax constructs. Demonstrates careful handling of edge cases like orphaned close tags, multi-byte characters across chunk boundaries, and empty CDATA. The parser is tightly coupled to the internal memory structure but shows high craftsmanship in state management.