Code generated from a schema

Here you own the format. You define messages and fields, and the compiler decides the bytes. Protocol Buffers writes each field as a tag holding its field number and wire type, followed by the value: a varint, a fixed-size number, or a length and that many bytes. That is tag-length-value, so an old decoder skips fields added after it was built.

nanopb is a schema compiler for microcontrollers. It reads a standard .proto file and emits C structs plus field descriptors for a runtime of 5 to 20 kB of code that uses about 1 kB of stack. Options such as max_count and max_size turn repeated fields and strings into fixed arrays inside the struct, so decoding needs no malloc, and input that would overflow an array is rejected.

A schema can’t describe a format that already exists, because the compiler chooses the bytes; the Kaitai Struct FAQ says the two kinds of tool solve completely different problems. Schemaless binary formats such as CBOR sit in between: any decoder can read the data without a schema, and the CDDL notation can describe the expected shape when you want one.