[fallintoplace]

What

Scan long textproto string literals with memchr3.

Why

lex_string_run checks each byte in long unescaped strings.

Implementation

• Search for quote, backslash, and newline in one call.
• Keep the byte loop for short strings and after a special byte.
• Keep memchr optional and preserve no_std support.

Benchmark

Tokenizer microbenchmark on Apple M1 Pro, Rust 1.95 release. Exact upstream base versus this branch, seven alternating runs; median ns/op. Lower is better.

Input	Before	After	Speedup
Plain ASCII, 16 B	34.121 ns	35.270 ns	0.97x
Plain ASCII, 64 B	71.436 ns	28.868 ns	2.47x
Plain ASCII, 2 KiB	1,584.087 ns	118.391 ns	13.38x
Plain ASCII, 16 KiB	12,387.722 ns	771.388 ns	16.06x
UTF-8, about 2 KiB	1,593.851 ns	117.812 ns	13.53x
One escape, 2 KiB	1,587.509 ns	987.433 ns	1.61x
Dense escapes, 2 KiB	697.039 ns	806.209 ns	0.86x
Two adjacent literals, 2 KiB	1,592.575 ns	122.178 ns	13.03x

The 16 B case is about 3% slower and dense escapes about 14% slower in this run. Large plain literals are much faster. Benchmark harness.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}