Backends

An MCodec[T] never names a format. It reads from an abstract Input and writes to an abstract Output, a small streaming API. A backend implements that pair for one wire format, and every derived codec runs against it unchanged.

The Input / Output model

Output offers three things to write:

• writeNull()
• writeSimple(): SimpleOutput, then one of writeString / writeInt / writeLong / writeDouble / writeBoolean / writeBigInt / writeBigDecimal (writeByte, writeChar, writeBinary, … sit on top of those)
• writeList(): ListOutput and writeObject(): ObjectOutput, sequential writers you feed elements or named fields

Input mirrors it: readNull(), readSimple(), readList(), readObject(), skip(), and on backends that can peek, peekKind() — used by @flatten and @outOfOrder.

A product codec calls output.writeObject(), writes each field by name, and closes it. A collection codec calls writeList(). Reading walks ObjectInput / ListInput with hasNext, nextField(), and nextElement().

The JSON backend

Json is the only wired-up entry point. Json.write[T](value) and Json.read[T](string) each need a given MCodec[T]:

import halotukozak.mcodec.*

case class Point(x: Int, y: Int) derives MCodec

val s: String = Json.write(Point(1, 2)) // {"x":1,"y":2}
val p: Point = Json.read[Point](s)

Json.write builds the String through a JsonOutput. Json.read parses it with a JsonReader behind a JsonInput and checks that nothing is left over.

BSON and CBOR

Input / Output implementations for BSON and CBOR sit in the source tree and run in the test suite. There is no public Bson / Cbor read/write object yet, so they are not part of the supported API.

Writing your own backend

Implement Input and Output. The InputAndSimpleInput / OutputAndSimpleOutput helper traits fold the simple readers and writers in when your format doesn't split them out. Every existing MCodec[T] then works against your backend as is.