strpack

A tiny, fast, dependency-free binary encoding for strings and counts.

• A string is a u8 length followed by its bytes, or 0xFF followed by a little-endian u32 length for strings of 255 bytes or more.
• A count is a little-endian u32.

Writer copies strings of 4 to 64 bytes with a few inline loads and stores instead of calling memcpy, and Reader skips full UTF-8 validation for ASCII strings.

Reader never panics: it returns None on truncated or corrupted input, and a count can't claim more items than the rest of the input could hold, so a corrupted count can't make you allocate a huge collection.

Example

let mut writer = strpack::Writer::default();
writer.write_count(2);
writer.write_str("hello");
writer.write_str("world");
let bytes = writer.into_bytes();

let mut reader = strpack::Reader::new(&bytes);
let count = reader.read_count(1).unwrap();
let strings: Vec<&str> = (0..count).map(|_| reader.read_str().unwrap()).collect();
assert_eq!(strings, ["hello", "world"]);

Testing

strpack's unsafe code is tested with Miri, including on 32-bit, big-endian, and 32-bit ARM targets:

cargo test
MIRIFLAGS="-Zmiri-strict-provenance -Zmiri-symbolic-alignment-check" cargo +nightly miri test
cargo +nightly miri test --target i686-unknown-linux-gnu
cargo +nightly miri test --target s390x-unknown-linux-gnu
cargo +nightly miri test --target armv7-unknown-linux-gnueabihf

The tests round-trip every string length from 0 to 300 at different alignments, plus long strings, and check that truncated, corrupted, and invalid UTF-8 input is never misread and never panics.