v0.1.2 — Full TOON spec v1.5 support (delimiters + key folding)

Highlights

Complete TOON spec v1.5 coverage: 146/146 fixtures passing (up from 111 in v0.1.1). Two major new options for LLM token efficiency: custom delimiters and key folding.

New features

Custom delimiters (saves tokens when values contain commas)

etoon.dumps([{"id": 1, "note": "Hello, world"}, {"id": 2, "note": "Foo, bar"}], delimiter="|")
# [2|]{id|note}:
#   1|Hello, world
#   2|Foo, bar

Supports , (default), \t, |. When the value delimiter is not comma, commas inside strings don't need quoting → fewer tokens.

Key folding (collapse nested chains)

etoon.dumps({"api": {"v2": {"users": {"count": 42}}}}, fold_keys=True)
# api.v2.users.count: 42

Single-key object chains collapse into dot-notation. Saves 30-50% on deeply nested config/schema data. Options:

• fold_keys=True enables safe-mode folding (skips segments needing quotes, avoids collisions with sibling literal keys)
• flatten_depth=N limits max chain length

Stricter key quoting

Keys now enforce the TOON identifier pattern [a-zA-Z_][a-zA-Z0-9_.]*. Keys with hyphens, symbols, or leading digits are now properly quoted ("full-name", "order:id").

Performance

Encoder	Time (neptune 50-doc)	vs etoon
etoon v0.1.2 (Python)	~15 μs	baseline
@toon-format/toon (TS SDK)	35.6 μs	2.4× slower
py-rtoon	85.9 μs	5.7× slower
toons	106.4 μs	7.1× slower

~2μs regression vs v0.1.1 due to new param threading cost (Config struct, allow_fold). See v0.1.3 for the const-generic optimization that partially recovers it.

Install

pip install --upgrade etoon                # Python
cargo add etoon --no-default-features      # Rust library
cargo install etoon --no-default-features  # Rust CLI from source

Full Changelog: v0.1.1...v0.1.2