Page
Library
Module
Module type
Parameter
Class
Class type
Source
Repr
SourceYet-another type combinator library
Repr
provides type combinators to define runtime representation for OCaml types and generic operations to manipulate values with a runtime type representation.
The type combinators supports all the usual type primitives but also compact definitions of records and variants. It also allows the definition of run-time representations of recursive types.
The type for runtime representation of values of type 'a
.
The type of integer used to store buffers, list or array lengths.
Int
use a (compressed) variable encoding to encode integers in a binary format, while IntX
always use X
bytes. Overflows are not detected.
int
is a representation of integers. Binary serialization uses a varying-width representation.
int63
is a representation of the 63-bit integer type supplied by the Optint
library.
boxed t
is the same as t
but with a binary representation which is always boxed (e.g. top-level values won't be unboxed). This forces Unboxed
functions to be exactly the same as boxed ones.
list t
is a representation of lists of values of type t
.
array t
is a representation of arrays of values of type t
.
triple x y z
is a representation of values of type x * y * z
.
quad w x y z
is a representation of values of type w * x * y * z
.
result a b
is a representation of values of type (a, b) result
.
either a b
is a representation of values of type (a, b) Either.t
.
ref t
is a representation of references to values of type t
.
Note: derived deserialisation functions will not preserve reference sharing.
lazy_t t
is a representation of lazy values of type t
.
Note: derived deserialisation functions on the resulting type will not be lazy.
hashtbl k v
is a representation of hashtables with keys of type k
and values of type v
.
set (module Set) elt
is a representation of sets with elements of type elt
. See Of_set
for a functorised equivalent of this function.
Functor for building representatives of sets from the standard library.
Functor for building representatives of maps from the standard library.
An uninhabited type, defined as a variant with no constructors.
The type for fields holding values of type 'b
and belonging to a record of type 'a
.
field n t g
is the representation of the field called n
of type t
with getter g
. Raises. Invalid_argument
if n
is not valid UTF-8.
The name n
is used for non-binary encoding/decoding and for pretty printing. It must not be used by any other field
in the record.
For instance:
type manuscript = { title : string option }
let manuscript = field "title" (option string) (fun t -> t.title)
r |+ f
is the open record r
augmented with the field f
.
sealr r
seals the open record r
. Raises. Invalid_argument
if two or more fields share the same name.
Putting all together:
type menu = { restaurant : string; items : (string * int32) list }
let t =
record "t" (fun restaurant items -> { restaurant; items })
|+ field "restaurant" string (fun t -> t.restaurant)
|+ field "items" (list (pair string int32)) (fun t -> t.items)
|> sealr
The type for representing variant cases of type 'a
with patterns of type 'b
.
The type for representing patterns for a variant of type 'a
.
case0 n v
is a representation of a variant constructor v
with no arguments and name n
. Raises. Invalid_argument
if n
is not valid UTF-8.
The name n
is used for non-binary encoding/decoding and for pretty printing. It must not by used by any other case0
in the record.
For instance:
type t = Foo
let foo = case0 "Foo" Foo
case1 n t c
is a representation of a variant constructor c
with an argument of type t
and name n
. Raises. Invalid_argument
if n
is not valid UTF-8.
The name n
is used for non-binary encoding/decoding and for pretty printing. It must not by used by any other case1
in the record.
For instance:
type t = Foo of string
let foo = case1 "Foo" string (fun s -> Foo s)
v |~ c
is the open variant v
augmented with the case c
.
sealv v
seals the open variant v
. Raises. Invalid_argument
if two or more cases of same arity share the same name.
Putting all together:
type t = Foo | Bar of string
let t =
variant "t" (fun foo bar -> function Foo -> foo | Bar s -> bar s)
|~ case0 "Foo" Foo
|~ case1 "Bar" string (fun x -> Bar x)
|> sealv
enum n cs
is a representation of the variant type called n
with singleton cases cs
. e.g.
type t = Foo | Bar | Toto
let t = enum "t" [ ("Foo", Foo); ("Bar", Bar); ("Toto", Toto) ]
The name n
and the case names are used for non-binary encoding/decoding and for pretty printing. Raises. Invalid_argument
if two or more cases share the same name.
Repr
allows a limited description of recursive records and variants.
TODO: describe the limitations, e.g. only regular recursion and no use of the generics inside the mu*
functions and the usual caveats with recursive values (such as infinite loops on most of the generics which don't check sharing).
mu f
is the representation r
such that r = mu r
.
For instance:
type x = { x : x option }
let x =
mu (fun x ->
record "x" (fun x -> { x })
|+ field "x" (option x) (fun x -> x.x)
|> sealr)
mu2 f
is the representations r
and s
such that r, s = mu2 r s
.
For instance:
type r = { foo : int; bar : string list; z : z option }
and z = { x : int; r : r list }
(* Build the representation of [r] knowing [z]'s. *)
let mkr z =
record "r" (fun foo bar z -> { foo; bar; z })
|+ field "foo" int (fun t -> t.foo)
|+ field "bar" (list string) (fun t -> t.bar)
|+ field "z" (option z) (fun t -> t.z)
|> sealr
(* And the representation of [z] knowing [r]'s. *)
let mkz r =
record "z" (fun x r -> { x; r })
|+ field "x" int (fun t -> t.x)
|+ field "r" (list r) (fun t -> t.r)
|> sealr
(* Tie the loop. *)
let r, z = mu2 (fun r z -> (mkr z, mkz y))
The type for staged operations.
stage x
stages x
, where x
would typically be a function that is expensive to construct.
unstage x
unstages x
.
Both stage
and unstage
are implemented with the identity function.
As the generic operations tend to be used repeatedly with the same left-most parameters, this type trick encourages the user to specialise them only once for performance reasons.
For instance:
let t = Repr.(pair int bool)
let compare = Repr.(unstage (compare t))
let sorted_list =
List.init 42_000 (fun _ -> (Random.int 100_000, Random.bool ()))
|> List.sort compare
Given a value 'a t
, it is possible to define generic operations on value of type 'a
such as pretty-printing, parsing and unparsing.
The type for pretty-printers.
pp_dump t
is the dump pretty-printer for values of type t
.
This pretty-printer outputs an encoding which is as close as possible to native OCaml syntax, so that the result can easily be copy-pasted into an OCaml REPL to inspect the value further.
random t
is a random value generator for values of type t
. For bounded types, values are sampled uniformly; for unbounded ones (lists, strings etc.), the length is first chosen according to a geometric distribution.
Derived generators use the global PRNG state provided by Stdlib.Random.get_state
.
NOTE: this generator may fail to terminate when sampling a recursive type.
random_state
is a variant of random
that takes an explicit PRNG state to use for random generation.
Attributes provide a mechanism for attaching metadata to type representations.
The type for JSON encoders.
The type for JSON decoders.
Similar to pp_dump
but pretty-prints the JSON representation instead of the OCaml one. See encode_json
for details about the encoding.
For instance:
type t = { foo : int option; bar : string list }
let t =
record "r" (fun foo bar -> { foo; bar })
|+ field "foo" (option int) (fun t -> t.foo)
|+ field "bar" (list string) (fun t -> t.bar)
|> sealr
let s = Fmt.str "%a\n" (pp t) { foo = None; bar = [ "foo" ] }
(* s is "{ foo = None; bar = [\"foo\"]; }" *)
let j = Fmt.str "%a\n" (pp_json t) { foo = None; bar = [ "foo" ] }
(* j is "{ \"bar\":[\"foo\"] }" *)
NOTE: this will automatically convert JSON fragments to valid JSON objects by adding an enclosing array if necessary.
encode_json t e
encodes t
into the jsonm encoder e
. The encoding is a relatively straightforward translation of the OCaml structure into JSON. The main highlights are:
()
is translated into the empty object {}
.None
are removed from the JSON object; record fields with a value of Some x
are automatically unboxed into x; and outside of records, None
is translated into null
and Some x
into {"some": x'}
with x'
the JSON encoding of x
.case0
are represented as strings.case1
are represented as a record with one field; the field name is the name of the variant.NOTE: this can be used to encode JSON fragments. It's the responsibility of the caller to ensure that the encoded JSON fragment fits properly into a well-formed JSON object.
decode_json t e
decodes values of type t
from the jsonm decoder e
.
decode_json_lexemes
is similar to decode_json
but uses an already decoded list of JSON lexemes instead of a decoder.
to_json_string
is encode_json
with a string encoder.
of_json_string
is decode_json
with a string decoder .
The type for binary encoders.
The type for size function related to binary encoder/decoders.
hash t x
is a short hash of x
of type t
.
encode_bin t
is the binary encoder for values of type t
.
decode_bin t
is the binary decoder for values of type t
.
to_bin_string t x
use encode_bin
to convert x
, of type t
, to a string.
NOTE: When t
is string
or bytes
, the original buffer x
is not prefixed by its size as encode_bin
would do. If t
is string
, the result is x
(without copy).
of_bin_string t s
is v
such that s = to_bin_string t v
.
NOTE: When t
is string
, the result is s
(without copy).
size_of t x
is either the size of encode_bin t x
or the binary encoding of x
, if the backend is not able to pre-compute serialisation lengths.
Unboxed operations assumes that value being serialized is fully filling the underlying buffer. When that's the case, it is not necessary to prefix the value's binary representation by its size, as it is exactly the buffer's size.
val abstract :
pp:'a pp ->
of_string:'a of_string ->
json:('a encode_json * 'a decode_json) ->
bin:('a encode_bin * 'a decode_bin * 'a size_of) ->
?unboxed_bin:('a encode_bin * 'a decode_bin * 'a size_of) ->
equal:'a equal ->
compare:'a compare ->
short_hash:'a short_hash ->
pre_hash:'a encode_bin ->
unit ->
'a t
The representation of an abstract type, with an internal structure that is opaque to Repr, that supports the generic operations above.
For a given type representation, each generic operation can be implemented in one of the following ways:
val partially_abstract :
pp:'a pp impl ->
of_string:'a of_string impl ->
json:('a encode_json * 'a decode_json) impl ->
bin:('a encode_bin * 'a decode_bin * 'a size_of) impl ->
unboxed_bin:('a encode_bin * 'a decode_bin * 'a size_of) impl ->
equal:'a equal impl ->
compare:'a compare impl ->
short_hash:'a short_hash impl ->
pre_hash:'a encode_bin impl ->
'a t ->
'a t
partially_abstract t
is a partially-abstract type with internal representation t
. The named arguments specify the implementation of each of the generic operations on this type.
val like :
?pp:'a pp ->
?of_string:'a of_string ->
?json:('a encode_json * 'a decode_json) ->
?bin:('a encode_bin * 'a decode_bin * 'a size_of) ->
?unboxed_bin:('a encode_bin * 'a decode_bin * 'a size_of) ->
?equal:'a equal ->
?compare:'a compare ->
?short_hash:'a short_hash ->
?pre_hash:'a encode_bin ->
'a t ->
'a t
A wrapper around partially_abstract
with each operation defaulting to `Structural
and admitting a `Custom
override.
Note: if ~compare
is passed and ~equal
is not then the default equality function (fun x y -> compare x y = 0)
will be used.
val map :
?pp:'a pp ->
?of_string:'a of_string ->
?json:('a encode_json * 'a decode_json) ->
?bin:('a encode_bin * 'a decode_bin * 'a size_of) ->
?unboxed_bin:('a encode_bin * 'a decode_bin * 'a size_of) ->
?equal:'a equal ->
?compare:'a compare ->
?short_hash:'a short_hash ->
?pre_hash:'a encode_bin ->
'b t ->
('b -> 'a) ->
('a -> 'b) ->
'a t
This combinator allows defining a representative of one type in terms of another by supplying coercions between them. For a representative of Stdlib.Map
, see Of_map
.
This module provides functions for interacting with Repr's binary serialisation format directly (without first constructing a representation of the type being encoded). These can be useful for performance-critical applications, where the runtime overhead of the dynamic specialisation is too large, or when the actual codec being used is too complex to be expressed via a type representation.