json

-- Encoding

json.encode({'foo', {bar = {'baz', json.null, 1, 2}}})
--< '["foo", {"bar": ["baz", null, 1, 2]}]'
print(json.encode('\\'))
--< "\\"
print(json.encode({c = 0, b = 0, a = 0}))
--< {"a": 0, "b": 0, "c": 0}

-- Decoding

local obj = json.decode('["foo", {"bar":["baz", null, 1.0, 2]}]')
print(json.decode('"\\"foo\\bar"'))

Types

Constants

null: unspecified

The single object that represents the JSON null value.

It’s safe to compare against this object to test for JSON’s null.

If you call tostring() on this object, the string "null" will be returned.

lexer_ecat

It comes straight from the imported library and we don’t really control the error codes.

dom_ecat

Errors from this category don’t mean the textual JSON representation is invalid. Rather, conversion to/from lua value failed (e.g. number overflow would occur, nesting level too deep, cyclic references, …​).

This error category represents the very membrane between textual and Lua data representation.

Functions

decode(raw_json: string) → value

Deserialize raw_json to a lua value.

local json_str = '{"items":[],"properties":{}}'
print(json_str)
print(json.encode(json.decode(json_str)))

will output (do note that order is unspecified and might change from emilua version to version):

{"items":[],"properties":{}}
{"properties":{},"items":[]}

encode(value[, opts: table]) → string

Serialize value to a JSON formatted string.

print(json.encode(json.null))
print(json.encode({hello = 'world', what = json.null,
                   animals = {'cow', 'coelho'}}))
print(json.encode(json.into_array()))
print(json.encode('hey "pretty"'))

will output:

null
{"what":null,"hello":"world","animals":["cow","coelho"]}
[]
"hey \"pretty\""

If value (or any nested element) has a __tojson() metamethod, it’ll be used to serialize that nesting level. Check __tojson() below to see parameters documentation.

opts is an options table that might contain the following fields:

  • indent: the indentation string (or nil if a compact ugly JSON is desired).

  • state: the state object passed in the __tojson() call. Useful to serialize further subobjects from the metamethod site. This option overrides other options in the opts table.

    If called with state, encode() will NOT return the generated string as it expects to write a partial value using state.writer only.

is_array(json: value) → boolean

Test if json is a lua table and it has been tagged using the json.into_array() function to indicate that it represents a JSON array.

Example
local raw_json = [[ ["test", 4, false] ]]

function poor_print(value)
   if json.is_array(value) then
      print(unpack(value))
   elseif type(value) == 'table' then
      print('{')
      for k, v in pairs(value) do
         print('', '"' .. k .. '"', v)
      end
      print('}')
   elseif type(value) == 'string' then
      print('"' .. value .. '"')
   else
      print(value)
   end
end

poor_print(json.decode(raw_json))

into_array([json: table]) → table

Change json's metatable to a certain tag that indicates either:

  • The associated table was created from the result of parsing a JSON array.

  • If this table is used to generate JSON textual representation, it should be encoded as a JSON array.

json is returned from this function to favour certain useful syntactic idioms.

If called with no arguments, a new array is created and returned.

Use json.is_array() to check if some value has been marked using this function.

Customization point metamethods

__tojson(self, state)

Called to write current node in the JSON tree.

state is a table with the following fields:

  • writer: the generator.

  • visited: a table to detect reference cycles. Before serializing a suboject, check whether visited already contains the to-be-serialized table. If a cycle is detected, raise cycle_exists error. If all is good, set visited[t] = true before calling getmetatable(t).__tojson(t, state) on the subobject t.

  • indent: the indentation string (or nil if a compact ugly JSON is desired). Current level of nested containers can be queried through writer, so you should write this string as many times as this reported level.

A trick to avoid the error-prone interactions involving state (e.g. updating visited, etc) to serialize subobjects is to call json.encode(t, { state = state }) on the subobject t. This way, you move the responsibility away to the json module itself. Example:

-- NOTE: this example ignores `indent`
mt = {
    __tojson = function(o, state)
        local writer = state.writer

        writer:begin_object()
        writer:value('foo')
        writer:value(o.foo) --< a number
        writer:value('bar')

        -- a subobject
        -- might contain its own `__tojson()`
        json.encode(o.bar, { state = state })

        writer:end_object()
    end
}

Conversion table

Lua type JSON type Notes

json.null

null

boolean

boolean

number

number

string

string

table

array

On decode(raw_json):

  • The lua table is marked with the json.into_array() function.

On encode(lua_obj):

  • lua_obj is encoded as a JSON array if it has been marked as so using json.into_array() or #lua_obj evaluates to a value larger than 0.

  • Non-integer keys are ignored.

table

object

On encode(lua_obj):

  • Non-string keys are ignored.

Rationale

These choices are also used by other lua libraries in the wild.

David Heiko Kolf's work on collecting and comparing JSON libraries for Lua, and generally documenting common pitfalls as well, was specially helpful. Thanks to his work it was much easier for me to design my own solution.

null

Encoding the JSON null value is a problem. Lua treats nil as indistinguishable from an absent value so we can’t really map null to nil. This problem only gets worse when interactions with sparse tables begin. However, JavaScript uses a different value for absent, undefined. And the same solution is chosen here with the introduction of a json.null value.

JSON arrays

JSON arrays and JSON objects will map to the same type — lua tables. How do we differentiate them? This problem isn’t exclusive to Lua. JavaScript itself suffers from this problem:

> typeof({})
'object'
> typeof([])
'object'

The solution chosen by JavaScript is an Array.isArray() function:

> Array.isArray({})
false
> Array.isArray([])
true

Therefore the same solution is chosen here:

local value = json.decode(raw_json)
if json.is_array(value) then
  -- ...
end

And json.into_aray() is introduced to make certain patterns easier to work with (especially for the encode() function).

I acknowledge that dkjson’s __jsontype metafield is more general, but JSON doesn’t really need this kind of generality. JSON is a closed world.

encode()

The following libraries and pages inspired this function:

The decode() function avoids a recursive implementation. However, the encode() function does not share the same property. The reason why no effort was made to offer a recursion-free encode() implementation is the __tojson() metamethod. This metamethod would force an unbounded call-stack anyway, so there is no point. However, the recursion was implemented in lua bytecode, so at least your process shouldn’t crash on stack overflow. If you wish for a recursion-free implementation, you can use the generator interface directly and avoid __tojson() yourself.