Serialization and codecs

const stringify = JIT.json(User).stringify().compile();
const parse = JIT.json(User).parse().compile();

Known JSON keys and punctuation are baked into source. Strings use a fast clean-string path and fall back to native escaping only when necessary.

For large arrays, stream bounded chunks:

const stringifyChunks = JIT.json(Users)
  .stringifyChunks({ chunkBytes: 16 * 1024 })
  .compile();

for (const chunk of stringifyChunks(users)) writable.write(chunk);

The binary codec is a transport format with an explicit version byte. Binary rowsets are different: they are evolvable process-local analytical memory and must not be persisted as a wire contract.

JSON behavior

Known object keys and punctuation are static source fragments. Optional undefined properties are omitted like native JSON. Strings use a short-string character scan or long-string regex probe before falling back to native escaping.

Use JIT.json(S).parse() at an untrusted text boundary. Use the serializer directly only when the input is already typed/validated.

Chunk strategy

chunkBytes is an approximate UTF-16 code-unit budget, not exact UTF-8 bytes. Tune it to downstream backpressure and write overhead. Larger chunks reduce write calls; smaller chunks reduce peak retained text.

Binary wire compatibility

Codec v2 stores a version byte, little-endian numeric payloads, 2-bit optional states and length-prefixed UTF-8 strings. Changing this layout is a protocol change. Keep golden vectors between producers/consumers and reject version drift loudly.

Value codecs are different: they transform between two validated schemas in memory, for example ISO text and native Date.