December 18, 2025
No parse, all spice
Lite^3, a JSON-Compatible Zero-Copy Serialization Format
Dev crowd split: edit-on-the-wire magic or a leaky suitcase
TLDR: Lite³ lets developers edit data directly while it’s ready to send, skipping the usual unpack–repack dance. The crowd is split: fans love the speed; critics warn about unreclaimed space and potential data leaks, and argue it’s not true “serialization,” pressing for comparisons and safety proof.
Lite³ promises a wild trick: you can tweak data while it’s already “packed” to send, no unpacking required. Fans on Hacker News are hyped, with one user cheering that it’s “mutable in place” and linking to the design docs. The author says this smashes the old rule where you must parse and re-serialize—just edit and ship. Cue the drumroll, and the drama.
Then the skeptics enter. A sharp-eyed commenter calls out a spicy line in the docs: overwritten space isn’t reclaimed. Translation for non-devs: imagine a suitcase where you keep stuffing new clothes on top, but never take the old ones out. People worry about “leaked” old data hitching a ride if you send the whole buffer. Privacy alarm bells, anyone?
Another camp argues Lite³ is not really “serialization” at all—more like a portable tree of keys and values. One wag basically said: cool trick, questionable label. Meanwhile, comparison nerds are out in force: think Apache Arrow (a columnar data format) or even GLTF, with folks asking for head‑to‑head benchmarks and wondering why other “binary JSONs” don’t do this.
The vibe? Equal parts “this could be huge for fast systems” and “name it better, fix the trash, and prove it’s safe.” Classic internet: bold claims, big hopes, and meme energy about a magic suitcase that might spill secrets.
Key Points
- •Lite³ enables direct, in-place mutations of serialized data (arbitrary key/value insertion/overwrite) without reserialization or additional allocations.
- •Operations on Lite³ structures have O(log n) amortized time with predictable latency and zero-copy reads/writes in a single contiguous buffer.
- •The format is schemaless and self-describing, fully compatible with JSON, with optional yyjson for conversion to/from JSON.
- •Lite³ targets embedded/no-malloc environments: caller-provided buffers, built-in pointer validation, low memory profile; core is 9.3 kB, C11, dependency-free.
- •A feature matrix contrasts Lite³ with formats like JSON, Protobuf, Flatbuffers, Cap'n Proto, and Apache Arrow, highlighting unique zero-copy write capability; benchmarks reference a simdjson dataset.