Fast JSON parsing and encoding support for Lua
C Lua Perl
Switch branches/tags
Nothing to show
Latest commit 4f27182 Mar 18, 2014 @efelix Merge pull request #2 from yueyoum/README
add Note and Fix of error: undefined symbol: luaL_register


Lua CJSON v1.0.3

Lua CJSON is covered by the MIT license. See the file "LICENSE" for

Lua CJSON provides fast JSON parsing and encoding support for Lua.

- 10x to 20x quicker (or more) than the fastest pure Lua JSON modules.
- Full support for JSON with UTF-8, including decoding surrogate
- Optional run-time support for common exceptions to the JSON
  specification (NaN, Inf,..).

- UTF-16 and UTF-32 are not supported.
- Multiple OS threads within a single Lua state are not currently
  supported. However, this is an extremely uncommon configuration due
  to performance limitations.

To obtain the latest version of Lua CJSON visit:

Feel free to email me if you have any patches, suggestions, or

- Mark Pulford <>


Build requirements:
- Lua (
- LuaJIT (

There are 3 build methods available:
- Gmake: POSIX, OSX
- RPM: Various Linux distributions
- LuaRocks ( POSIX, OSX, Windows


Review and update the included Makefile to suit your platform. Next,
build and install the module:

  # gmake
  # gmake install
  # cp [your_module_directory]

Note: Some Solaris platforms are missing isinf(). You can work around
      this bug by adding -DMISSING_ISINF to CFLAGS in the Makefile.

Note: For Lua 5.2. You may got a 'undefined symbol: luaL_register' error.
      You can adding -DLUA_COMPAT_MODULE to CFLAGS in the Makefile to fix it.


RPM-based Linux distributions should be able to create a package using
the included RPM spec file. Install the "rpm-build" package, or
similar, then:

  # rpmbuild -tb lua-cjson-1.0.3.tar.gz


LuaRocks ( can be used to install and manage Lua
modules on a wide range of platforms (including Windows).

Extract the Lua CJSON source package into a directory and run:

  # cd lua-cjson-1.0.3; luarocks make

See the LuaRocks documentation for further details.



  require "cjson"
  -- Or:
  local cjson = require "cjson"

  -- Translate Lua value to/from JSON
  text = cjson.encode(value)
  value = cjson.decode(text)

  -- Get and/or set CJSON configuration
  setting = cjson.refuse_invalid_numbers([setting])
  depth = cjson.encode_max_depth([depth])
  convert, ratio, safe = cjson.encode_sparse_array([convert[, ratio[, safe]]])
  keep = cjson.encode_keep_buffer([keep])


  json_text = cjson.encode(value)

cjson.encode() will serialise the following types:
  * number, string, table, boolean, lightuserdata (NULL) or nil

The remaining Lua types cannot be serialised:
  * thread, userdata, lightuserdata (non-NULL), function

Numbers are encoded using the standard Lua number format.

ASCII 0 - 31, double-quote, forward-slash, black-slash and ASCII 127
are escaped when encoding strings. Other octets are passed
transparently. It is expected the application will perform UTF-8 error
checking if required.

A Lua table will only be recognised as an array if all keys are type
"number" and are positive integers (>0). Otherwise CJSON will encode
the table as a JSON object.

CJSON will also recognise and handle sparse arrays. Missing entries
will be encoded as "null". Eg:
  { [3] = "data" }

Note: standards compliant JSON must be encapsulated in either an
object ({}) or an array ([]). You must pass a table to cjson.encode()
if you want to generate standards compliant JSON output.

By default, errors will be raised for:
- Excessively sparse arrays (see below)
- More than 20 nested tables
- Invalid numbers (NaN, Infinity)

These defaults can be changed with:
- cjson.encode_sparse_array()
- cjson.encode_max_depth()
- cjson.refuse_invalid_numbers()

  data_obj = { true, { foo = "bar" } }
  data_json = cjson.encode(data_obj)


  value = cjson.decode(json_text)

cjson.decode() will deserialise any UTF-8 JSON string into a Lua data
structure. It can return any of the types that cjson.encode()

UTF-16 and UTF-32 JSON strings are not supported.

CJSON requires that NULL (\0) and double quote (\") are escaped within
strings. All escape codes will be decoded and other characters will be
passed transparently. UTF-8 characters are not validated during
decoding and should be checked elsewhere if required.

JSON "null" will be converted to a NULL lightuserdata value. This can
be compared with cjson.null for convenience.

By default, invalid numbers (NaN, Infinity, Hexidecimal) will be

  data_json = '[ true, { "foo": "bar" } ]'
  data_obj = cjson.decode(data_json)

Invalid numbers

  setting = cjson.refuse_invalid_numbers([setting])
  -- "setting" must be on of:
  --       false, "encode", "decode", "both", true

CJSON considers numbers which are outside the JSON specification to be
"invalid". Eg:
- Infinity
- NaN
- Hexadecimal numbers

By default CJSON will decode "invalid" numbers, but will refuse to
encode them.

This setting can be configured separately for encoding and/or
- Enabled: an error will be generated if an invalid number is found.
- Disabled (encoding): NaN and Infinity can be encoded.
- Disabled (decoding): All numbers supported by strtod(3) will be

Sparse arrays

  convert, ratio, safe = cjson.encode_sparse_array([convert[, ratio[, safe]]])
  -- "convert" must be a boolean. Default: false.
  -- "ratio" must be a positive integer (>0). Default: 2
  -- "safe" must be a positive integer (>0). Default: 10

A Lua array is sparse if it is missing a value for at least 1 index.
Lua CJSON encodes missing values as "null". Eg:
  Lua array:  { [3] = "sparse" }
  JSON array: [null,null,"sparse"]

CJSON detects excessively sparse arrays by comparing the number of
items in a Lua array with the maximum index. In particular:

  maximum index > safe AND maximum index > array_items * ratio

By default, attempting to encode excessively sparse arrays will
generate an error.

If "convert" is set to "true", excessively sparse arrays will be
encoded as a JSON object:
  Lua array:  { [1000] = "excessively sparse" }
  JSON array: {"1000":"excessively sparse"}

Setting "ratio" to 0 disables checking for excessively sparse arrays.

Nested tables

  depth = cjson.encode_max_depth([depth])
  -- "depth" must be a positive integer (>0).

By default, CJSON will reject data structure with more than 20 nested

This check is used to prevent a nested data structure from crashing
the application. Eg:
  a = {}; b = { a }; a[1] = b

Number precision

  precision = cjson.encode_number_precision([precision])
  -- "precision" must be between 1 and 14 (inclusive)

By default CJSON will output 14 significant digits when converting a
number to text.

Reducing number precision to 3 can improve performance of number
heavy conversions by up to 50%.

Persistent encoding buffer

  keep = cjson.keep_encode_buffer([keep])
  -- "keep" must be a boolean

By default, CJSON will reuse the JSON encoding buffer to improve
performance. The buffer will grow to the largest size required and is
not freed until CJSON is garbage collected. Setting this option to
"false" will cause the buffer to be freed after each call to

Lua / JSON limitations and CJSON

Null handling

Lua CJSON decodes JSON "null" as a Lua lightuserdata NULL pointer.

CJSON provides "cjson.null" as a convenience for comparison.

Table keys

JSON object keys must be strings - other types are not supported. Lua
CJSON will convert numeric keys to a string, and other non-string
types will generate an error.

JSON object keys are always be decoded as Lua strings.

If all Lua table keys are numbers (not strings), Lua CJSON will
encode the table as a JSON array. See "Sparse arrays" above for
more details.


Lua CJSON does not use metamethods when serialising tables.
- next() is used to iterate over tables.
- rawget() is used when iterating over arrays.

Functions, Userdata, Threads

Lua CJSON will generate an error if asked to serialise Lua functions,
userdata, lightuserdata or threads.