Tagcode

Four Rust implementations of the "3y3" / "secondsight" Unicode Tags steganography technique (hiding ASCII text as invisible characters in the U+E0000..=U+E007F block), from a direct 1:1 port of the original algorithm up to a byte-level "ultra opti" version.

• Original JS algorithm: https://synthetic.garden/3y3.htm, popularized at https://github.com/ArjixWasTaken/3y3
• Rust port this crate started from: https://github.com/bitten2up/secondsight/blob/master/src/lib.rs

All documentation is inline as Rust doc comments (/// / //!). Run cargo doc --open --no-deps for a browsable, cross-linked version.

Modules

Module	File	Summary
naive	src/naive.rs	Direct 1:1 port of the original. Correct, but allocates a String per character.
optimized	src/optimized.rs	One allocation total, no redundant validity checks (char::from_u32_unchecked with a documented safety invariant).
cow_optimized	src/cow_optimized.rs	Same as optimized, but returns Cow<str> — zero allocation when there's nothing to encode/decode.
fast_path	src/fast_path.rs	Byte-level implementation that skips char decoding entirely, exploiting the fixed UTF-8 byte shape of the Tags block.

Running things

cargo test                          # correctness: all 4 impls cross-checked + doctests
cargo doc --open --no-deps          # full docs, browsable
cargo run --release --example bench # small micro-benchmark comparing all 4
just heavy-bench                    # heavy benchmark: 64B..2MiB x 3 content profiles, real stats
just demo                           # end-to-end demo including the `meta` dispatcher

A justfile is included (test, doctest, doc, bench, heavy-bench, fmt, lint, check, build, clean, ci, demo, you can run just with no arguments to list them all).

Safety

optimized, cow_optimized, and fast_path use unsafe (mainly char::from_u32_unchecked and, in fast_path, String::from_utf8_unchecked). Every unsafe block has a // SAFETY: comment explaining exactly why it's sound, and the module-level docs in optimized.rs spell out the invariant all three modules rely on. All four modules are cross-checked against each other in src/lib.rs's test suite, including edge cases (NUL, DEL, the full ASCII range, multi-byte/astral UTF-8, and text that already contains real Tag-block characters before encoding).