Rec2json is a parse transform that takes a module which defines a record of the same name and adds to_json, from_json, and introspection functions. The to_json and from_json convert to and from the map based format used by jsx and other json encoding and decoding libraries.

• Uses a parse transform.
• Type checking on json -> record conversion using primitive built-in types.
• Type checking can be extended to include user defined types.
• Atom 'undefined' fields in records optionally skipped or set to null.
• Atom 'null' in json optionally converted to 'undefined'.
• Post processing options on record -> json convertion.
• Seed json -> record conversion with a record.
• Nested json -> record and record -> json conversions for other records that have been compiled using rec2json.
• Generated module has accessor functions for fields and field_names for list of fields in the record.
• Above feature can be surpressed, and is careful by default.
• Generated module exports functions to examine structure and types for a record.

make

To run tests:

make tests

make check

The rec2json parse_transform looks for a record with the same name as the module. It doesn't matter whether the record was defined directly in the module or in an include file.

-compile([{parse_transform, rec2json}]).

The transformed modules depend on the rec2json application's modules.

The records are unchanged and can be used normally.

Options are passed to the rec2json parse transform through compile options. The parse transform checks for the key 'rec2json' in the compile options. The value is expected to be a proplist.

Options can also be passed in on a per-module basis by adding one or more rec2json module attributes. A rec2json module attribute can either be a tuple for one option, or a list of option tuples. They all get mashed together. Using the same option more than once has undefined behavior.

Options are:

The given examples use the following record and record defintion:

-record(person {
    name :: binary(),
    age = 0 :: pos_integer(),
    spouse :: #person{}
}).
Record = #person{ name = <<"John">>, age = 32, spouse = undefined }.

To convert a record to a json structure:

Json = person:to_json(Record).
#{name := <<"John">>, age := 32} = Json.

The to_json function can take a list of mutators. Mutators are applied in the order listed except {null_is_undefined}. Supported mutators are:

• Turn undefined into null instead of skipping the field

person:to_json(Record, [{null_is_undefined}]).

• Add a property

person:to_json(Record, [{single, true}]).
person:to_json(Record, [#{single => true}).
person:to_json(Record, [#{single => true, employed => true}).

• Remove a property

person:to_json(Record, [age]).

• Modify based only on the json

ModFunc = fun(Json) ->
    case maps:find(spouse, Json) of
        error ->
            Json#{single => true}
        _ ->
            Json#{single => false}
    end
end.
person:to_json(Record, [ModFunc]).

• Modify based on both json and record

ModFunc = fun(Json, Record) ->
    case Record#person.spouse of
        undefined ->
            Json#{single => true};
        _ ->
            Json#{single => false}
    end
end.
person:to_json(Record, [ModFunc]).

Converting from a json structure to a record is just as simple:

{ok, Record} = person:from_json(#{
    <<"name">> => <<"John">>,
    <<"age">> => 32,
    <<"spouse">> => null
}).

It may be desirable to change 'null' into 'undefined' in the record:

{ok, Record} = person:from_json(Json, [null_is_undefined]).

It may be desirable to start with an existing record instead of creating a new one:

{ok, Record2} = person:from_json(Json, Record).
{ok, Record2} = person:from_json(Record, Json).
{ok, Record2} = person:from_json(Record, Json, [null_is_undefined]).

If the json structure has a type that cannot be reconciled with a type specified by the record definition, a list of fields with possible errors is returned. The record will have the data that was in the json structure. An untyped record field is the same as having the type 'any()'. There are no warnings about missing properties in the json, they simply retain the default value of the record.

{ok, Record, [age]} = person:from_json(#{<<"age">> => <<"32">>}).

If all you are using is the parse_transform, simply add rec2json as a required application.

Type conversion attempts to be as transparent and intuitive as possible. There are some types that json does not represent directly, and some types that have additional checking implemented.

Record fields that have atoms as types will have binary values in the json checked. If the atom converted to the binary is equal to the json value, the atom value is put into the record. When converting a record to json, atom values will be converted to binaries.

Lists have their types checked. If there is an invalid type, the invalid type is placed in the list, but the warning message has the index of the invalid type placed in the warning path list. For example:

-record(list_holder, {
    ids :: [integer()]
}).

type_mismatch() ->
    Json = #{ids => [<<"invalid">>, 3]},
    {ok, Record, Warnings} = list_holder:from_json(Json),
    #list_holder{ids = [<<"invalid">>, 3]} = Record,
    [[ids, 1]] = Warnings.

Proplists will match the first record type. A warning is emitted if that record was not compiled using rec2json (eg: RecordName:from_json/2 is not an exported function). If the compilation emits warnings, the resulting warning list has the field name prepended to each. For example:

-record(outer, {
    in_field :: #inner{}
}).
-record(inner, {
    count :: integer()
}).

type_mismatch() ->
    Json = #{in_field => #{count => <<"0">>}},
    {ok, Record, Warnings} = outer:from_json(Json),
    #outer{in_field = #inner{ count = <<"0">> } } = Record,
    [[in_filed, count]] = Warnings.

If a record field is not typed, or has the type "any()", no warning is ever emitted for that field.

Type checking comes in two flavors: built-in, and user defined.

Currently defined types checked:

• integer()
• pos_integer()
• non_neg_integer()
• neg_integer()
• float()
• number()
• boolean()
• binary()
• [supported_type()]
• #record{} when record has record:from_json/2 exported
• atom (note it is not atom())*
• null when converting to undefined or back

The type atom() is not supported because a primary use case for rec2json is to convert untrusted data, such as an http post request. Converting untrusted data into atoms can exhaust the erlang vm's atom table, or worse exhaust the machine's memory. Rec2json is, by default, safe.

It is still possible to apply a type to a record field so that json strings will be converted to the equivalent atom using either a user defined type, or the provided r2j_type:unsafe_atom() type.

A user defined type is the same as an external type. When going to or from json, rec2json will check to see if there is a translation function matching the module and type of the type given. A translation function should have an arity 1 greater than the type has parameters. The function should either return {ok, NewVal} or error.

For example, given the module:

-module(type_example).
-compile([{parse_transform, rec2json}]).

% These two lines exist to satisfy dialyzer.
-type point() :: {number(), number()}.
-export_type([point/0]).

-record(type_example, {
	some_field :: module:function(arg1, arg2),
	xy = {0, 0} :: type_example:point()
}).

-export([point/1]).

point({X,Y}) when is_number(X), is_number(Y) ->
	{ok, [X,Y]};

point([X, Y]) when is_number(X), is_number(Y) ->
	{ok, {X, Y}};

point(_) ->
	error.

When using to_json or from_json, rec2json will check to see if module exports a function named function with arity 3. If it exists, rec2json will call the function with the args listed in the type, and the current value of the field (either from the record or from the json) prepended to the arguments.

When used in from_json, returns error, and all other types have been tested, {ok, Rec, Warnings} is returned. When used in to_json, an error is thrown.

Fork and submit a pull request with relevant tests.