Type-safe serialization and deserialization of data structures.
This module provides type-safe serialization and deserialization of data structures. Backends are provided to both /ad hoc/ binary, JSON and BSON.
This works by writing type descriptors by hand, using the provided combinators. These combinators can fine-tune the binary representation to be compact and efficient, but also provide proper field names and meta information. As a result, an API that uses those descriptors can be automatically introspected and documented.
Here is an example encoding for type
(int * string).
let enc = obj2 (req "code" uint16) (req "message" string)
In JSON, this encoding maps values of type
int * string to JSON objects with a field
code whose value is a number and a field
message whose value is a string.
In binary, this encoding maps to two raw bytes for the
int followed by the size of the string in bytes, and finally the raw contents of the string. This binary format is mostly tagless, meaning that serialized data cannot be interpreted without the encoding that was used for serialization.
Regarding binary serialization, encodings are classified as either:
- fixed size (booleans, integers, numbers) data is always the same size for that type ;
- dynamically sized (arbitrary strings and bytes) data is of unknown size and requires an explicit length field ;
- variable size (special case of strings, bytes, and arrays) data makes up the remainder of an object of known size, thus its size is given by the context, and does not have to be serialized.
JSON operations are delegated to
Data_encoding module provides multiple submodules:
Encodingcontains the necessary types and constructors for making the type descriptors.
Binarycontain functions to serialize and deserialize values.
module Encoding : sig ... end
include module type of Encoding with type 'a t = 'a Encoding.t
type 'a t = 'a Encoding.t
The type descriptors for values of type
type 'a encoding = 'a t
val null : unit encoding
null in JSON, nothing in binary.
val empty : unit encoding
Empty object (not included in binary, encoded as empty object in JSON).
val unit : unit encoding
Unit value, omitted in binary. Serialized as an empty object in JSON, accepts any object when deserializing.
val constant : string -> unit encoding
Constant string (data is not included in the binary data).
ground numerical types
All encodings are big-endians.
- 8-bit integers (signed or unsigned) are encoded over 1 single byte.
- 16-bit integers (signed or unsigned) are encoded over 2 bytes.
- 31-bit integers are always signed and always encoded over 4 bytes.
- 32-bit integers are always signed and always encoded over 4 bytes.
- 64-bit integers are always signed and always encoded over 8 bytes.
A note on 31-bit integers. The internal representation of integers in OCaml reserves one bit for GC tagging. The remaining bits encode a signed integer. For compatibility with 32-bit machine, we restrict these native integers to the 31-bit range.
val int8 : int encoding
Signed 8 bit integer (data is encoded as a byte in binary and an integer in JSON).
val uint8 : int encoding
Unsigned 8 bit integer (data is encoded as a byte in binary and an integer in JSON).
val int16 : int encoding
Signed 16 bit integer (data is encoded as a short in binary and an integer in JSON).
val uint16 : int encoding
Unsigned 16 bit integer (data is encoded as a short in binary and an integer in JSON).
val int31 : int encoding
Signed 31 bit integer, which corresponds to type int on 32-bit OCaml systems (data is encoded as a 32 bit int in binary and an integer in JSON).
val int32 : int32 encoding
Signed 32 bit integer (data is encoded as a 32-bit int in binary and an integer in JSON).
val int64 : int64 encoding
Signed 64 bit integer (data is encoded as a 64-bit int in binary and a decimal string in JSON).
val ranged_int : int -> int -> int encoding
Integer with bounds in a given range. Both bounds are inclusive.
Big number In JSON, data is encoded as a decimal string. In binary, data is encoded as a variable length sequence of bytes, with a running unary size bit: the most significant bit of each byte tells is this is the last byte in the sequence (0) or if there is more to read (1). The second most significant bit of the first byte is reserved for the sign (positive if zero). Binary_size and sign bits ignored, data is then the binary representation of the absolute value of the number in little-endian order.
val float : float encoding
Encoding of floating point number (encoded as a floating point number in JSON and a double in binary).
val ranged_float : float -> float -> float encoding
Float with bounds in a given range. Both bounds are inclusive
Other ground type encodings
val bool : bool encoding
Encoding of a boolean (data is encoded as a byte in binary and a boolean in JSON).
val string : string encoding
Encoding of a string
- encoded as a byte sequence in binary prefixed by the length of the string
- encoded as a string in JSON.
Encoding of arbitrary bytes (encoded via hex in JSON and directly as a sequence byte in binary).
Combinator to make an optional value (represented as a 1-byte tag followed by the data (or nothing) in binary and either the raw value or a null in JSON).
Note that the JSON representation is only weakly discriminating. Specifically, the value
Some None is represented as the raw value
None and so the two are indistinguishable. For this reason, this combinator does not support nesting, nor does it support use within a recursive (
Combinator to make a
result value (represented as a 1-byte tag followed by the data of either type in binary, and either unwrapped value in JSON (the caller must ensure that both encodings do not collide)).
- encoded as an array in JSON
- encoded as the concatenation of all the element in binary prefixed its length in bytes
- encoded as an array in JSON
- encoded as the concatenation of all the element in binary prefixed its length in bytes
val conv : ('a -> 'b) -> ('b -> 'a) -> ?schema:Json_schema.schema -> 'b encoding -> 'a encoding
val conv_with_guard : ('a -> 'b) -> ('b -> ('a, string) result) -> ?schema:Json_schema.schema -> 'b encoding -> 'a encoding
conv_with_guard is similar to
conv but the function that takes in the value from the outside (untrusted) world has a chance to fail.
Specifically, if the function returns
Error msg then the decoding is interrupted with an error carrying the message
msg. If the function returns
Ok _ then the decoding proceeds normally.
with_decoding_guard g e is similar to
e but decoding fails if
Error _ on the decoded value.
Association list. An object in JSON, a list of pairs in binary.
An enriched encoding to represent a component in a structured type, augmenting the encoding with a name and whether it is a required or optional. Fields are used to encode OCaml tuples as objects in JSON, and as sequences in binary, using combinator
obj1 and the like.
Optional field. Omitted entirely in JSON encoding if None. Omitted in binary if the only optional field in a
`Variable encoding, otherwise a 1-byte prefix (`0` or `255`) tells if the field is present or not.
Optional field of variable length. Only one can be present in a given object.
Required field with a default value. If the default value is passed, the field is omitted in JSON. The value is always serialized in binary.
Constructors for objects with N fields
These are serialized to binary by converting each internal object to binary and placing them in the order of the original object. These are serialized to JSON as a JSON object with the field names. An object might only contains one 'variable' field, typically the last one. If the encoding of more than one field are 'variable', the first ones should be wrapped with
Create a larger object from the encodings of two smaller ones.
Constructors for tuples with N fields
These are serialized to binary by converting each internal object to binary and placing them in the order of the original object. These are serialized to JSON as JSON arrays/lists. Like objects, a tuple might only contains one 'variable' field, typically the last one. If the encoding of more than one field are 'variable', the first ones should be wrapped with
Create a large tuple encoding from two smaller ones.
A partial encoding to represent a case in a variant type. Hides the (existentially bound) type of the parameter to the specific case, providing its encoder, and converter functions to and from the union type.
type 'a matching_function = 'a -> match_result
A sum descriptor can be optimized by providing a specific
matching_function which efficiently determines in which case some value of type
Note that in general you should use a total function (i.e., one defined over the whole of the
'a type) for the
matching_function. However, in the case where you have a good reason to use a partial function, you should raise
No_case_matched in the dead branches. Reasons why you may want to do so include:
'ais an open variant and you will complete the matching function later, and
- there is a code invariant that guarantees that
'ais not fully inhabited.
val matched : ?tag_size:[ `Uint8 | `Uint16 ] -> int -> 'a encoding -> 'a -> match_result
val case : title:string -> ?description:string -> case_tag -> 'a encoding -> ('t -> 'a option) -> ('a -> 't) -> 't case
Encodes a variant constructor. Takes the encoding for the specific parameters, a recognizer function that will extract the parameters in case the expected case of the variant is being serialized, and a constructor function for deserialization.
The tag must be less than the tag size of the union in which you use the case. An optional tag gives a name to a case and should be used to maintain compatibility.
An optional name for the case can be provided, which is used in the binary documentation.
val matching : ?tag_size:[ `Uint8 | `Uint16 ] -> 't matching_function -> 't case list -> 't encoding
Create a single encoding from a series of cases.
In JSON, all cases are tried one after the other using the
case list. The caller is responsible for avoiding collisions. If there are collisions (i.e., if multiple cases produce the same JSON output) then the encoding and decoding processes might not be inverse of each other. In other words,
destruct e (construct e v) may not be equal to
In binary, a prefix tag is added to discriminate quickly between cases. The default is
`Uint8 and you must use a
`Uint16 if you are going to have more than 256 cases.
The matching function is used during binary encoding of a value
v to efficiently determine which of the cases corresponds to
v. The case list is used during decoding to reconstruct a value based on the encoded tag. (Decoding is optimised internally: tag look-up has a constant cost.)
The caller is responsible for ensuring that the
matching_function and the
case list describe the same encoding. If they describe different encodings, then the decoding and encoding processes will not be inverses of each others. In other words,
of_bytes e (to_bytes e v) will not be equal to
If you do not wish to be responsible for this, you can use the unoptimised
union that uses a
case list only (see below). Beware that in
union the complexity of the encoding is linear in the number of cases.
Following: a basic example use. Note that the
matching_function uses the same tags, payload conversions, and payload encoding as the
type t = A of string | B of int * int | C let encoding_t = (* Tags and payload encodings for each constructors *) let a_tag = 0 and a_encoding = string in let b_tag = 1 and b_encoding = obj2 (req "x" int) (req "y" int) in let c_tag = 2 and c_encoding = unit in matching (* optimised encoding function *) (function | A s -> matched a_tag a_encoding s | B (x, y) -> matched b_tag b_encoding (x, y) | C -> matched c_tag c_encoding ()) (* decoding case list *) [ case ~title:"A" (Tag a_tag) a_encoding (function A s -> Some s | _ -> None) (fun s -> A s); case ~title:"B" (Tag b_tag) b_encoding (function B (x, y) -> Some (x, y) | _ -> None) (fun (x, y) -> B (x, y)); case ~title:"C" (Tag c_tag) c_encoding (function C -> Some () | _ -> None) (fun () -> C); ]
Same as matching except that the matching function is a linear traversal of the cases.
Predicates over descriptors
val is_obj : 'a encoding -> bool
Is the given encoding serialized as a JSON object?
val is_tup : 'a encoding -> bool
Does the given encoding encode a tuple?
val classify : 'a encoding -> [ `Fixed of int | `Dynamic | `Variable ]
Classify the binary serialization of an encoding as explained in the preamble.
val string_enum : (string * 'a) list -> 'a encoding
Encode enumeration via association list
- represented as a string in JSON and
- represented as an integer representing the element's position in the list in binary. The integer size depends on the list size.
module Fixed : sig ... end
Create encodings that produce data of a fixed length when binary encoded. See the preamble for an explanation.
module Variable : sig ... end
Create encodings that produce data of a variable length when binary encoded. See the preamble for an explanation.
module Bounded : sig ... end
Mark an encoding as being of dynamic size. Forces the size to be stored alongside content when needed. Typically used to combine two variable encodings in a same objects or tuple, or to use a variable encoding in an array or a list.
check_size size encoding ensures that the binary encoding of a value will not be allowed to exceed
size bytes. The reader and the writer fails otherwise. This function do not modify the JSON encoding.
Recompute the encoding definition each time it is used. Useful for dynamically updating the encoding of values of an extensible type via a global reference (e.g., exceptions).
Define different encodings for JSON and binary serialization.
val mu : string -> ?title:string -> ?description:string -> ('a encoding -> 'a encoding) -> 'a encoding
Combinator for recursive encodings.
Notice that the function passed to
mu must be pure. Otherwise, the behavior is unspecified.
A stateful recursive encoding can still be put under a
delayed combinator to make sure that a new encoding is generated each time it is used. Caching the encoding generation when the state has not changed is then the responsability of the client.
Give a name to an encoding and optionally add documentation to an encoding.
Combinator to have a part of the binary encoding lazily deserialized. This is transparent on the JSON side.
val force_decode : 'a lazy_t -> 'a option
Force the decoding (memoized for later calls), and return the value if successful.
Obtain the bytes without actually deserializing. Will serialize and memoize the result if the value is not the result of a lazy deserialization.
val apply_lazy : fun_value:('a -> 'b) -> fun_bytes:(Bytes.t -> 'b) -> fun_combine:('b -> 'b -> 'b) -> 'a lazy_t -> 'b
Apply on structure of lazy value, and combine results
module Compact : sig ... end
This module provides specialized encoding combinators that are implemented to reduce the size of the serialization result.
type 'a compact = 'a Compact.t
module With_version : sig ... end
Data_encoding.t value which records knowledge of older versions of a given encoding as long as one can "upgrade" from an older version to the next (if upgrade is impossible one should consider that the encoding is completely different).
module Json : sig ... end
module Bson : sig ... end
module Binary_schema : sig ... end
module Binary_stream : sig ... end
module Binary : sig ... end
type json = Json.t
val json : json Encoding.t
json is an encoding for JSON values. It is mostly intended for internal use or for defining your own low-level combinators.
WARNING! Due to a limitation of BSON, this encoding does not safely roundtrip. Specifically,
Json.destruct json (Json.construct json v) is not guaranteed to be equal to
v. More specifically, in BSON, top-level Arrays are represented as number-indexed Objects and this library has no way to distinguish between the two, doubly so for empty collections.
?bson_relaxation optional parameter.
type json_schema = Json.schema
val json_schema : json_schema Encoding.t
type bson = Bson.t
module Registration : sig ... end