• avro is the runtime library, should be reasonably small (depends only on camlzip)
• avro-compiler can read a json schema and generate code from it

The online documentation can be found here

• code generation from json schema (using avro-compiler)
  • NOTE: default values are not supported yet.
• object container file format
• binary encoding
• schema evolution (in particular the schema is emitted as-is, nothing is stripped)
• single object file format
• sorting
• json encoding
• compression codecs:
  • null
  • deflate
  • snappy (optional codec)

Schemas are json files (described here). Some examples can be found in tests/.

For example, the schema:

{
  "name": "employee",
  "type": "record",
  "fields": [
    {"name": "firstname", "type": "string"},
    {"name": "lastname", "type": "string"},
    {"name": "age", "type": "int"},
    {"name": "friendliness", "type": "float"},
    {"name": "job", "type": [
      {"name": "programmer", "type": "record",
        "fields": [{"name": "language", "type": {"type":"enum", "name":"language","symbols": ["cpp", "ocaml", "java", "python"]}}]},
      {"name": "RH", "type": "record",
        "fields": [{"name": "hair_spikiness", "type": {"type":"fixed", "size": 4, "name":"spikiness"}}]}
    ] }
  ]
}

will be compiled to the types:

type nonrec language =   | Cpp | Ocaml | Java | Python

type nonrec programmer = { language: language; }

type nonrec rH = { hair_spikiness: string; }
type nonrec union_0 =  
  | C_programmer of programmer 
  | C_RH of rH 
                       
type nonrec t = {
  firstname: string; 
  lastname: string; 
  age: int; 
  friendliness: float; 
  job: union_0; 
}

New codecs can be registered in Avro.Obj_container_format.Codec. Currently it's quite naive as it goes from a string to a string, instead of being fully streaming. This might change before 1.0.

module Codec : sig
  type t

  (* … *)

  val register :
    name:string ->
    compress:(string -> string) ->
    decompress:(string -> string) ->
    unit -> t
  (** Register decompression codecs. Defaults are "null" and "deflate". *)
end

See some examples from the tests/ directory:

• tests/records.json and tests/records_test.ml that use it. you can see what the schema compiler will produce:

  $ ./compiler.sh ./tests/records.json
  (* generated by avro-compiler *)
  
  open Avro
  
  module Str_map = Map.Make(String)
  
  let schema = "{\"type\":\"record\",\"name\":\"test\",\"fields\":[{\"type\":\"long\",\"name\":\"a\"},{
  \"type\":\"string\",\"name\":\"b\"}]}"
  
  type nonrec t = { a: int64;  b: string; }
  
  let read (input:Input.t) : t =
    (let a = Input.read_int64 input in
     let b = Input.read_string input in
     { a;b })
  
  let write (out:Output.t) (self:t) : unit =
    ((let self = self.a in Output.write_int64 out self);
      (let self = self.b in Output.write_string out self);
      )
  

• similarly for tests/employee.json and tests/employee_test.ml that use it

MIT license.