Jansson Documentation

This is the documentation for Jansson 2.10, last updated Jul 12, 2017.

Introduction

Jansson is a C library for encoding, decoding and manipulating JSON data. Its main features and design principles are:

  • Simple and intuitive API and data model
  • Comprehensive documentation
  • No dependencies on other libraries
  • Full Unicode support (UTF-8)
  • Extensive test suite

Jansson is licensed under the MIT license; see LICENSE in the source distribution for details.

Jansson is used in production and its API is stable. It works on numerous platforms, including numerous Unix like systems and Windows. It’s suitable for use on any system, including desktop, server, and small embedded systems.

Contents

Getting Started

Compiling and Installing Jansson

The Jansson source is available at http://www.digip.org/jansson/releases/.

Unix-like systems (including MinGW)

Unpack the source tarball and change to the source directory:

bunzip2 -c jansson-2.10.tar.bz2 | tar xf -
cd jansson-2.10

The source uses GNU Autotools (autoconf, automake, libtool), so compiling and installing is extremely simple:

./configure
make
make check
make install

To change the destination directory (/usr/local by default), use the --prefix=DIR argument to ./configure. See ./configure --help for the list of all possible configuration options.

The command make check runs the test suite distributed with Jansson. This step is not strictly necessary, but it may find possible problems that Jansson has on your platform. If any problems are found, please report them.

If you obtained the source from a Git repository (or any other source control system), there’s no ./configure script as it’s not kept in version control. To create the script, the build system needs to be bootstrapped. There are many ways to do this, but the easiest one is to use autoreconf:

autoreconf -fi

This command creates the ./configure script, which can then be used as described above.

CMake (various platforms, including Windows)

Jansson can be built using CMake. Create a build directory for an out-of-tree build, change to that directory, and run cmake (or ccmake, cmake-gui, or similar) to configure the project.

See the examples below for more detailed information.

Note

In the below examples .. is used as an argument for cmake. This is simply the path to the jansson project root directory. In the example it is assumed you’ve created a sub-directory build and are using that. You could use any path you want.

Unix (Make files)

Generating make files on unix:

bunzip2 -c jansson-2.10.tar.bz2 | tar xf -
cd jansson-2.10

mkdir build
cd build
cmake .. # or ccmake .. for a GUI.

Then to build:

make
make check
make install
Windows (Visual Studio)

Creating Visual Studio project files from the command line:

<unpack>
cd jansson-2.10

md build
cd build
cmake -G "Visual Studio 10" ..

You will now have a Visual Studio Solution in your build directory. To run the unit tests build the RUN_TESTS project.

If you prefer a GUI the cmake line in the above example can be replaced with:

cmake-gui ..

For command line help (including a list of available generators) for CMake simply run:

cmake

To list available CMake settings (and what they are currently set to) for the project, run:

cmake -LH ..
Mac OSX (Xcode)

If you prefer using Xcode instead of make files on OSX, do the following. (Use the same steps as for Unix):

...
cmake -G "Xcode" ..
Additional CMake settings
Shared library

By default the CMake project will generate build files for building the static library. To build the shared version use:

...
cmake -DJANSSON_BUILD_SHARED_LIBS=1 ..
Changing install directory (same as autoconf –prefix)

Just as with the autoconf project you can change the destination directory for make install. The equivalent for autoconfs ./configure --prefix in CMake is:

...
cmake -DCMAKE_INSTALL_PREFIX:PATH=/some/other/path ..
make install
Android

Jansson can be built for Android platforms. Android.mk is in the source root directory. The configuration header file is located in the android directory in the source distribution.

Other Systems

On non Unix-like systems, you may be unable to run the ./configure script. In this case, follow these steps. All the files mentioned can be found in the src/ directory.

  1. Create jansson_config.h (which has some platform-specific parameters that are normally filled in by the ./configure script). Edit jansson_config.h.in, replacing all @variable@ placeholders, and rename the file to jansson_config.h.
  2. Make jansson.h and jansson_config.h available to the compiler, so that they can be found when compiling programs that use Jansson.
  3. Compile all the .c files (in the src/ directory) into a library file. Make the library available to the compiler, as in step 2.
Building the Documentation

(This subsection describes how to build the HTML documentation you are currently reading, so it can be safely skipped.)

Documentation is in the doc/ subdirectory. It’s written in reStructuredText with Sphinx annotations. To generate the HTML documentation, invoke:

make html

and point your browser to doc/_build/html/index.html. Sphinx 1.0 or newer is required to generate the documentation.

Compiling Programs that Use Jansson

Jansson involves one C header file, jansson.h, so it’s enough to put the line

#include <jansson.h>

in the beginning of every source file that uses Jansson.

There’s also just one library to link with, libjansson. Compile and link the program as follows:

cc -o prog prog.c -ljansson

Starting from version 1.2, there’s also support for pkg-config:

cc -o prog prog.c `pkg-config --cflags --libs jansson`

Upgrading from 1.x

This chapter lists the backwards incompatible changes introduced in Jansson 2.0, and the steps that are needed for upgrading your code.

The incompatibilities are not dramatic. The biggest change is that all decoding functions now require and extra parameter. Most programs can be modified to work with 2.0 by adding a 0 as the second parameter to all calls of json_loads(), json_loadf() and json_load_file().

Compatibility

Jansson 2.0 is backwards incompatible with the Jansson 1.x releases. It is ABI incompatible, i.e. all programs dynamically linking to the Jansson library need to be recompiled. It’s also API incompatible, i.e. the source code of programs using Jansson 1.x may need modifications to make them compile against Jansson 2.0.

All the 2.x releases are guaranteed to be backwards compatible for both ABI and API, so no recompilation or source changes are needed when upgrading from 2.x to 2.y.

List of Incompatible Changes

Decoding flags

For future needs, a flags parameter was added as the second parameter to all decoding functions, i.e. json_loads(), json_loadf() and json_load_file(). All calls to these functions need to be changed by adding a 0 as the second argument. For example:

/* old code */
json_loads(input, &error);

/* new code */
json_loads(input, 0, &error);
Underlying type of JSON integers

The underlying C type of JSON integers has been changed from int to the widest available signed integer type, i.e. long long or long, depending on whether long long is supported on your system or not. This makes the whole 64-bit integer range available on most modern systems.

jansson.h has a typedef json_int_t to the underlying integer type. int should still be used in most cases when dealing with smallish JSON integers, as the compiler handles implicit type coercion. Only when the full 64-bit range is needed, json_int_t should be explicitly used.

Maximum encoder indentation depth
The maximum argument of the JSON_INDENT() macro has been changed from 255 to 31, to free up bits from the flags parameter of json_dumps(), json_dumpf() and json_dump_file(). If your code uses a bigger indentation than 31, it needs to be changed.
Unsigned integers in API functions
Version 2.0 unifies unsigned integer usage in the API. All uses of unsigned int and unsigned long have been replaced with size_t. This includes flags, container sizes, etc. This should not require source code changes, as both unsigned int and unsigned long are usually compatible with size_t.

Tutorial

In this tutorial, we create a program that fetches the latest commits of a repository in GitHub over the web. GitHub API uses JSON, so the result can be parsed using Jansson.

To stick to the the scope of this tutorial, we will only cover the the parts of the program related to handling JSON data. For the best user experience, the full source code is available: github_commits.c. To compile it (on Unix-like systems with gcc), use the following command:

gcc -o github_commits github_commits.c -ljansson -lcurl

libcurl is used to communicate over the web, so it is required to compile the program.

The command line syntax is:

github_commits USER REPOSITORY

USER is a GitHub user ID and REPOSITORY is the repository name. Please note that the GitHub API is rate limited, so if you run the program too many times within a short period of time, the sever starts to respond with an error.

The GitHub Repo Commits API

The GitHub Repo Commits API is used by sending HTTP requests to URLs like https://api.github.com/repos/USER/REPOSITORY/commits, where USER and REPOSITORY are the GitHub user ID and the name of the repository whose commits are to be listed, respectively.

GitHub responds with a JSON array of the following form:

[
    {
        "sha": "<the commit ID>",
        "commit": {
            "message": "<the commit message>",
            <more fields, not important to this tutorial...>
        },
        <more fields...>
    },
    {
        "sha": "<the commit ID>",
        "commit": {
            "message": "<the commit message>",
            <more fields...>
        },
        <more fields...>
    },
    <more commits...>
]

In our program, the HTTP request is sent using the following function:

static char *request(const char *url);

It takes the URL as a parameter, preforms a HTTP GET request, and returns a newly allocated string that contains the response body. If the request fails, an error message is printed to stderr and the return value is NULL. For full details, refer to the code, as the actual implementation is not important here.

The Program

First the includes:

#include <string.h>
#include <jansson.h>

Like all the programs using Jansson, we need to include jansson.h.

The following definitions are used to build the GitHub API request URL:

#define URL_FORMAT   "https://api.github.com/repos/%s/%s/commits"
#define URL_SIZE     256

The following function is used when formatting the result to find the first newline in the commit message:

/* Return the offset of the first newline in text or the length of
   text if there's no newline */
static int newline_offset(const char *text)
{
    const char *newline = strchr(text, '\n');
    if(!newline)
        return strlen(text);
    else
        return (int)(newline - text);
}

The main function follows. In the beginning, we first declare a bunch of variables and check the command line parameters:

int main(int argc, char *argv[])
{
    size_t i;
    char *text;
    char url[URL_SIZE];

    json_t *root;
    json_error_t error;

    if(argc != 3)
    {
        fprintf(stderr, "usage: %s USER REPOSITORY\n\n", argv[0]);
        fprintf(stderr, "List commits at USER's REPOSITORY.\n\n");
        return 2;
    }

Then we build the request URL using the user and repository names given as command line parameters:

snprintf(url, URL_SIZE, URL_FORMAT, argv[1], argv[2]);

This uses the URL_SIZE and URL_FORMAT constants defined above. Now we’re ready to actually request the JSON data over the web:

text = request(url);
if(!text)
    return 1;

If an error occurs, our function request prints the error and returns NULL, so it’s enough to just return 1 from the main function.

Next we’ll call json_loads() to decode the JSON text we got as a response:

root = json_loads(text, 0, &error);
free(text);

if(!root)
{
    fprintf(stderr, "error: on line %d: %s\n", error.line, error.text);
    return 1;
}

We don’t need the JSON text anymore, so we can free the text variable right after decoding it. If json_loads() fails, it returns NULL and sets error information to the json_error_t structure given as the second parameter. In this case, our program prints the error information out and returns 1 from the main function.

Now we’re ready to extract the data out of the decoded JSON response. The structure of the response JSON was explained in section The GitHub Repo Commits API.

We check that the returned value really is an array:

if(!json_is_array(root))
{
    fprintf(stderr, "error: root is not an array\n");
    json_decref(root);
    return 1;
}

Then we proceed to loop over all the commits in the array:

for(i = 0; i < json_array_size(root); i++)
{
    json_t *data, *sha, *commit, *message;
    const char *message_text;

    data = json_array_get(root, i);
    if(!json_is_object(data))
    {
        fprintf(stderr, "error: commit data %d is not an object\n", i + 1);
        json_decref(root);
        return 1;
    }
...

The function json_array_size() returns the size of a JSON array. First, we again declare some variables and then extract the i’th element of the root array using json_array_get(). We also check that the resulting value is a JSON object.

Next we’ll extract the commit ID (a hexadecimal SHA-1 sum), intermediate commit info object, and the commit message from that object. We also do proper type checks:

    sha = json_object_get(data, "sha");
    if(!json_is_string(sha))
    {
        fprintf(stderr, "error: commit %d: sha is not a string\n", i + 1);
        json_decref(root);
        return 1;
    }

    commit = json_object_get(data, "commit");
    if(!json_is_object(commit))
    {
        fprintf(stderr, "error: commit %d: commit is not an object\n", i + 1);
        json_decref(root);
        return 1;
    }

    message = json_object_get(commit, "message");
    if(!json_is_string(message))
    {
        fprintf(stderr, "error: commit %d: message is not a string\n", i + 1);
        json_decref(root);
        return 1;
    }
...

And finally, we’ll print the first 8 characters of the commit ID and the first line of the commit message. A C-style string is extracted from a JSON string using json_string_value():

    message_text = json_string_value(message);
    printf("%.8s %.*s\n",
           json_string_value(id),
           newline_offset(message_text),
           message_text);
}

After sending the HTTP request, we decoded the JSON text using json_loads(), remember? It returns a new reference to the JSON value it decodes. When we’re finished with the value, we’ll need to decrease the reference count using json_decref(). This way Jansson can release the resources:

json_decref(root);
return 0;

For a detailed explanation of reference counting in Jansson, see Reference Count in API Reference.

The program’s ready, let’s test it and view the latest commits in Jansson’s repository:

$ ./github_commits akheron jansson
1581f26a Merge branch '2.3'
aabfd493 load: Change buffer_pos to be a size_t
bd72efbd load: Avoid unexpected behaviour in macro expansion
e8fd3e30 Document and tweak json_load_callback()
873eddaf Merge pull request #60 from rogerz/contrib
bd2c0c73 Ignore the binary test_load_callback
17a51a4b Merge branch '2.3'
09c39adc Add json_load_callback to the list of exported symbols
cbb80baf Merge pull request #57 from rogerz/contrib
040bd7b0 Add json_load_callback()
2637faa4 Make test stripping locale independent
<...>

Conclusion

In this tutorial, we implemented a program that fetches the latest commits of a GitHub repository using the GitHub Repo Commits API. Jansson was used to decode the JSON response and to extract the commit data.

This tutorial only covered a small part of Jansson. For example, we did not create or manipulate JSON values at all. Proceed to API Reference to explore all features of Jansson.

RFC Conformance

JSON is specified in RFC 4627, “The application/json Media Type for JavaScript Object Notation (JSON)”.

Character Encoding

Jansson only supports UTF-8 encoded JSON texts. It does not support or auto-detect any of the other encodings mentioned in the RFC, namely UTF-16LE, UTF-16BE, UTF-32LE or UTF-32BE. Pure ASCII is supported, as it’s a subset of UTF-8.

Strings

JSON strings are mapped to C-style null-terminated character arrays, and UTF-8 encoding is used internally.

All Unicode codepoints U+0000 through U+10FFFF are allowed in string values. However, U+0000 is not allowed in object keys because of API restrictions.

Unicode normalization or any other transformation is never performed on any strings (string values or object keys). When checking for equivalence of strings or object keys, the comparison is performed byte by byte between the original UTF-8 representations of the strings.

Numbers

Real vs. Integer

JSON makes no distinction between real and integer numbers; Jansson does. Real numbers are mapped to the double type and integers to the json_int_t type, which is a typedef of long long or long, depending on whether long long is supported by your compiler or not.

A JSON number is considered to be a real number if its lexical representation includes one of e, E, or .; regardless if its actual numeric value is a true integer (e.g., all of 1E6, 3.0, 400E-2, and 3.14E3 are mathematical integers, but will be treated as real values). With the JSON_DECODE_INT_AS_REAL decoder flag set all numbers are interpreted as real.

All other JSON numbers are considered integers.

When encoding to JSON, real values are always represented with a fractional part; e.g., the double value 3.0 will be represented in JSON as 3.0, not 3.

Overflow, Underflow & Precision

Real numbers whose absolute values are too small to be represented in a C double will be silently estimated with 0.0. Thus, depending on platform, JSON numbers very close to zero such as 1E-999 may result in 0.0.

Real numbers whose absolute values are too large to be represented in a C double will result in an overflow error (a JSON decoding error). Thus, depending on platform, JSON numbers like 1E+999 or -1E+999 may result in a parsing error.

Likewise, integer numbers whose absolute values are too large to be represented in the json_int_t type (see above) will result in an overflow error (a JSON decoding error). Thus, depending on platform, JSON numbers like 1000000000000000 may result in parsing error.

Parsing JSON real numbers may result in a loss of precision. As long as overflow does not occur (i.e. a total loss of precision), the rounded approximate value is silently used. Thus the JSON number 1.000000000000000005 may, depending on platform, result in the double value 1.0.

Signed zeros

JSON makes no statement about what a number means; however Javascript (ECMAscript) does state that +0.0 and -0.0 must be treated as being distinct values, i.e. -0.0 ≠ 0.0. Jansson relies on the underlying floating point library in the C environment in which it is compiled. Therefore it is platform-dependent whether 0.0 and -0.0 will be distinct values. Most platforms that use the IEEE 754 floating-point standard will support signed zeros.

Note that this only applies to floating-point; neither JSON, C, or IEEE support the concept of signed integer zeros.

Types

No support is provided in Jansson for any C numeric types other than json_int_t and double. This excludes things such as unsigned types, long double, etc. Obviously, shorter types like short, int, long (if json_int_t is long long) and float are implicitly handled via the ordinary C type coercion rules (subject to overflow semantics). Also, no support or hooks are provided for any supplemental “bignum” type add-on packages.

Depth of nested values

To avoid stack exhaustion, Jansson currently limits the nesting depth for arrays and objects to a certain value (default: 2048), defined as a macro JSON_PARSER_MAX_DEPTH within jansson_config.h.

The limit is allowed to be set by the RFC; there is no recommended value or required minimum depth to be supported.

Portability

Thread safety

Jansson is thread safe and has no mutable global state. The only exceptions are the hash function seed and memory allocation functions, see below.

There’s no locking performed inside Jansson’s code, so a multithreaded program must perform its own locking if JSON values are shared by multiple threads. Jansson’s reference counting semantics may make this a bit harder than it seems, as it’s possible to have a reference to a value that’s also stored inside a list or object. Modifying the container (adding or removing values) may trigger concurrent access to such values, as containers manage the reference count of their contained values. Bugs involving concurrent incrementing or decrementing of deference counts may be hard to track.

The encoding functions (json_dumps() and friends) track reference loops by modifying the internal state of objects and arrays. For this reason, encoding functions must not be run on the same JSON values in two separate threads at the same time. As already noted above, be especially careful if two arrays or objects share their contained values with another array or object.

If you want to make sure that two JSON value hierarchies do not contain shared values, use json_deep_copy() to make copies.

Hash function seed

To prevent an attacker from intentionally causing large JSON objects with specially crafted keys to perform very slow, the hash function used by Jansson is randomized using a seed value. The seed is automatically generated on the first explicit or implicit call to json_object(), if json_object_seed() has not been called beforehand.

The seed is generated by using operating system’s entropy sources if they are available (/dev/urandom, CryptGenRandom()). The initialization is done in as thread safe manner as possible, by using architecture specific lockless operations if provided by the platform or the compiler.

If you’re using threads, it’s recommended to autoseed the hashtable explicitly before spawning any threads by calling json_object_seed(0) , especially if you’re unsure whether the initialization is thread safe on your platform.

Memory allocation functions

Memory allocation functions should be set at most once, and only on program startup. See Custom Memory Allocation.

Locale

Jansson works fine under any locale.

However, if the host program is multithreaded and uses setlocale() to switch the locale in one thread while Jansson is currently encoding or decoding JSON data in another thread, the result may be wrong or the program may even crash.

Jansson uses locale specific functions for certain string conversions in the encoder and decoder, and then converts the locale specific values to/from the JSON representation. This fails if the locale changes between the string conversion and the locale-to-JSON conversion. This can only happen in multithreaded programs that use setlocale(), because setlocale() switches the locale for all running threads, not only the thread that calls setlocale().

If your program uses setlocale() as described above, consider using the thread-safe uselocale() instead.

API Reference

Preliminaries

All declarations are in jansson.h, so it’s enough to

#include <jansson.h>

in each source file.

All constants are prefixed with JSON_ (except for those describing the library version, prefixed with JANSSON_). Other identifiers are prefixed with json_. Type names are suffixed with _t and typedef‘d so that the struct keyword need not be used.

Library Version

The Jansson version is of the form A.B.C, where A is the major version, B is the minor version and C is the micro version. If the micro version is zero, it’s omitted from the version string, i.e. the version string is just A.B.

When a new release only fixes bugs and doesn’t add new features or functionality, the micro version is incremented. When new features are added in a backwards compatible way, the minor version is incremented and the micro version is set to zero. When there are backwards incompatible changes, the major version is incremented and others are set to zero.

The following preprocessor constants specify the current version of the library:

JANSSON_MAJOR_VERSION, JANSSON_MINOR_VERSION, JANSSON_MICRO_VERSION
Integers specifying the major, minor and micro versions, respectively.
JANSSON_VERSION
A string representation of the current version, e.g. "1.2.1" or "1.3".
JANSSON_VERSION_HEX

A 3-byte hexadecimal representation of the version, e.g. 0x010201 for version 1.2.1 and 0x010300 for version 1.3. This is useful in numeric comparisons, e.g.:

#if JANSSON_VERSION_HEX >= 0x010300
/* Code specific to version 1.3 and above */
#endif

Value Representation

The JSON specification (RFC 4627) defines the following data types: object, array, string, number, boolean, and null. JSON types are used dynamically; arrays and objects can hold any other data type, including themselves. For this reason, Jansson’s type system is also dynamic in nature. There’s one C type to represent all JSON values, and this structure knows the type of the JSON value it holds.

json_t

This data structure is used throughout the library to represent all JSON values. It always contains the type of the JSON value it holds and the value’s reference count. The rest depends on the type of the value.

Objects of json_t are always used through a pointer. There are APIs for querying the type, manipulating the reference count, and for constructing and manipulating values of different types.

Unless noted otherwise, all API functions return an error value if an error occurs. Depending on the function’s signature, the error value is either NULL or -1. Invalid arguments or invalid input are apparent sources for errors. Memory allocation and I/O operations may also cause errors.

Type
enum json_type

The type of a JSON value. The following members are defined:

JSON_OBJECT
JSON_ARRAY
JSON_STRING
JSON_INTEGER
JSON_REAL
JSON_TRUE
JSON_FALSE
JSON_NULL

These correspond to JSON object, array, string, number, boolean and null. A number is represented by either a value of the type JSON_INTEGER or of the type JSON_REAL. A true boolean value is represented by a value of the type JSON_TRUE and false by a value of the type JSON_FALSE.

int json_typeof(const json_t *json)

Return the type of the JSON value (a json_type cast to int). json MUST NOT be NULL. This function is actually implemented as a macro for speed.

json_is_object(const json_t *json)
json_is_array(const json_t *json)
json_is_string(const json_t *json)
json_is_integer(const json_t *json)
json_is_real(const json_t *json)
json_is_true(const json_t *json)
json_is_false(const json_t *json)
json_is_null(const json_t *json)

These functions (actually macros) return true (non-zero) for values of the given type, and false (zero) for values of other types and for NULL.

json_is_number(const json_t *json)

Returns true for values of types JSON_INTEGER and JSON_REAL, and false for other types and for NULL.

json_is_boolean(const json_t *json)

Returns true for types JSON_TRUE and JSON_FALSE, and false for values of other types and for NULL.

json_boolean_value(const json_t *json)

Alias of json_is_true(), i.e. returns 1 for JSON_TRUE and 0 otherwise.

New in version 2.7.

Reference Count

The reference count is used to track whether a value is still in use or not. When a value is created, it’s reference count is set to 1. If a reference to a value is kept (e.g. a value is stored somewhere for later use), its reference count is incremented, and when the value is no longer needed, the reference count is decremented. When the reference count drops to zero, there are no references left, and the value can be destroyed.

json_t *json_incref(json_t *json)

Increment the reference count of json if it’s not NULL. Returns json.

void json_decref(json_t *json)

Decrement the reference count of json. As soon as a call to json_decref() drops the reference count to zero, the value is destroyed and it can no longer be used.

Functions creating new JSON values set the reference count to 1. These functions are said to return a new reference. Other functions returning (existing) JSON values do not normally increase the reference count. These functions are said to return a borrowed reference. So, if the user will hold a reference to a value returned as a borrowed reference, he must call json_incref(). As soon as the value is no longer needed, json_decref() should be called to release the reference.

Normally, all functions accepting a JSON value as an argument will manage the reference, i.e. increase and decrease the reference count as needed. However, some functions steal the reference, i.e. they have the same result as if the user called json_decref() on the argument right after calling the function. These functions are suffixed with _new or have _new_ somewhere in their name.

For example, the following code creates a new JSON array and appends an integer to it:

json_t *array, *integer;

array = json_array();
integer = json_integer(42);

json_array_append(array, integer);
json_decref(integer);

Note how the caller has to release the reference to the integer value by calling json_decref(). By using a reference stealing function json_array_append_new() instead of json_array_append(), the code becomes much simpler:

json_t *array = json_array();
json_array_append_new(array, json_integer(42));

In this case, the user doesn’t have to explicitly release the reference to the integer value, as json_array_append_new() steals the reference when appending the value to the array.

In the following sections it is clearly documented whether a function will return a new or borrowed reference or steal a reference to its argument.

Circular References

A circular reference is created when an object or an array is, directly or indirectly, inserted inside itself. The direct case is simple:

json_t *obj = json_object();
json_object_set(obj, "foo", obj);

Jansson will refuse to do this, and json_object_set() (and all the other such functions for objects and arrays) will return with an error status. The indirect case is the dangerous one:

json_t *arr1 = json_array(), *arr2 = json_array();
json_array_append(arr1, arr2);
json_array_append(arr2, arr1);

In this example, the array arr2 is contained in the array arr1, and vice versa. Jansson cannot check for this kind of indirect circular references without a performance hit, so it’s up to the user to avoid them.

If a circular reference is created, the memory consumed by the values cannot be freed by json_decref(). The reference counts never drops to zero because the values are keeping the references to each other. Moreover, trying to encode the values with any of the encoding functions will fail. The encoder detects circular references and returns an error status.

Scope Dereferencing

New in version 2.9.

It is possible to use the json_auto_t type to automatically dereference a value at the end of a scope. For example:

void function(void) {
  json_auto_t *value = NULL;
  value = json_string("foo");
  /* json_decref(value) is automatically called. */
}

This feature is only available on GCC and Clang. So if your project has a portability requirement for other compilers, you should avoid this feature.

Additionally, as always, care should be taken when passing values to functions that steal references.

True, False and Null

These three values are implemented as singletons, so the returned pointers won’t change between invocations of these functions.

json_t *json_true(void)
Return value: New reference.

Returns the JSON true value.

json_t *json_false(void)
Return value: New reference.

Returns the JSON false value.

json_t *json_boolean(val)
Return value: New reference.

Returns JSON false if val is zero, and JSON true otherwise. This is a macro, and equivalent to val ? json_true() : json_false().

New in version 2.4.

json_t *json_null(void)
Return value: New reference.

Returns the JSON null value.

String

Jansson uses UTF-8 as the character encoding. All JSON strings must be valid UTF-8 (or ASCII, as it’s a subset of UTF-8). All Unicode codepoints U+0000 through U+10FFFF are allowed, but you must use length-aware functions if you wish to embed null bytes in strings.

json_t *json_string(const char *value)
Return value: New reference.

Returns a new JSON string, or NULL on error. value must be a valid null terminated UTF-8 encoded Unicode string.

json_t *json_stringn(const char *value, size_t len)
Return value: New reference.

Like json_string(), but with explicit length, so value may contain null characters or not be null terminated.

json_t *json_string_nocheck(const char *value)
Return value: New reference.

Like json_string(), but doesn’t check that value is valid UTF-8. Use this function only if you are certain that this really is the case (e.g. you have already checked it by other means).

json_t *json_stringn_nocheck(const char *value, size_t len)
Return value: New reference.

Like json_string_nocheck(), but with explicit length, so value may contain null characters or not be null terminated.

const char *json_string_value(const json_t *string)

Returns the associated value of string as a null terminated UTF-8 encoded string, or NULL if string is not a JSON string.

The returned value is read-only and must not be modified or freed by the user. It is valid as long as string exists, i.e. as long as its reference count has not dropped to zero.

size_t json_string_length(const json_t *string)

Returns the length of string in its UTF-8 presentation, or zero if string is not a JSON string.

int json_string_set(json_t *string, const char *value)

Sets the associated value of string to value. value must be a valid UTF-8 encoded Unicode string. Returns 0 on success and -1 on error.

int json_string_setn(json_t *string, const char *value, size_t len)

Like json_string_set(), but with explicit length, so value may contain null characters or not be null terminated.

int json_string_set_nocheck(json_t *string, const char *value)

Like json_string_set(), but doesn’t check that value is valid UTF-8. Use this function only if you are certain that this really is the case (e.g. you have already checked it by other means).

int json_string_setn_nocheck(json_t *string, const char *value, size_t len)

Like json_string_set_nocheck(), but with explicit length, so value may contain null characters or not be null terminated.

Number

The JSON specification only contains one numeric type, “number”. The C programming language has distinct types for integer and floating-point numbers, so for practical reasons Jansson also has distinct types for the two. They are called “integer” and “real”, respectively. For more information, see RFC Conformance.

json_int_t

This is the C type that is used to store JSON integer values. It represents the widest integer type available on your system. In practice it’s just a typedef of long long if your compiler supports it, otherwise long.

Usually, you can safely use plain int in place of json_int_t, and the implicit C integer conversion handles the rest. Only when you know that you need the full 64-bit range, you should use json_int_t explicitly.

JSON_INTEGER_IS_LONG_LONG

This is a preprocessor variable that holds the value 1 if json_int_t is long long, and 0 if it’s long. It can be used as follows:

#if JSON_INTEGER_IS_LONG_LONG
/* Code specific for long long */
#else
/* Code specific for long */
#endif
JSON_INTEGER_FORMAT

This is a macro that expands to a printf() conversion specifier that corresponds to json_int_t, without the leading % sign, i.e. either "lld" or "ld". This macro is required because the actual type of json_int_t can be either long or long long, and printf() requires different length modifiers for the two.

Example:

json_int_t x = 123123123;
printf("x is %" JSON_INTEGER_FORMAT "\n", x);
json_t *json_integer(json_int_t value)
Return value: New reference.

Returns a new JSON integer, or NULL on error.

json_int_t json_integer_value(const json_t *integer)

Returns the associated value of integer, or 0 if json is not a JSON integer.

int json_integer_set(const json_t *integer, json_int_t value)

Sets the associated value of integer to value. Returns 0 on success and -1 if integer is not a JSON integer.

json_t *json_real(double value)
Return value: New reference.

Returns a new JSON real, or NULL on error.

double json_real_value(const json_t *real)

Returns the associated value of real, or 0.0 if real is not a JSON real.

int json_real_set(const json_t *real, double value)

Sets the associated value of real to value. Returns 0 on success and -1 if real is not a JSON real.

double json_number_value(const json_t *json)

Returns the associated value of the JSON integer or JSON real json, cast to double regardless of the actual type. If json is neither JSON real nor JSON integer, 0.0 is returned.

Array

A JSON array is an ordered collection of other JSON values.

json_t *json_array(void)
Return value: New reference.

Returns a new JSON array, or NULL on error. Initially, the array is empty.

size_t json_array_size(const json_t *array)

Returns the number of elements in array, or 0 if array is NULL or not a JSON array.

json_t *json_array_get(const json_t *array, size_t index)
Return value: Borrowed reference.

Returns the element in array at position index. The valid range for index is from 0 to the return value of json_array_size() minus 1. If array is not a JSON array, if array is NULL, or if index is out of range, NULL is returned.

int json_array_set(json_t *array, size_t index, json_t *value)

Replaces the element in array at position index with value. The valid range for index is from 0 to the return value of json_array_size() minus 1. Returns 0 on success and -1 on error.

int json_array_set_new(json_t *array, size_t index, json_t *value)

Like json_array_set() but steals the reference to value. This is useful when value is newly created and not used after the call.

int json_array_append(json_t *array, json_t *value)

Appends value to the end of array, growing the size of array by 1. Returns 0 on success and -1 on error.

int json_array_append_new(json_t *array, json_t *value)

Like json_array_append() but steals the reference to value. This is useful when value is newly created and not used after the call.

int json_array_insert(json_t *array, size_t index, json_t *value)

Inserts value to array at position index, shifting the elements at index and after it one position towards the end of the array. Returns 0 on success and -1 on error.

int json_array_insert_new(json_t *array, size_t index, json_t *value)

Like json_array_insert() but steals the reference to value. This is useful when value is newly created and not used after the call.

int json_array_remove(json_t *array, size_t index)

Removes the element in array at position index, shifting the elements after index one position towards the start of the array. Returns 0 on success and -1 on error. The reference count of the removed value is decremented.

int json_array_clear(json_t *array)

Removes all elements from array. Returns 0 on success and -1 on error. The reference count of all removed values are decremented.

int json_array_extend(json_t *array, json_t *other_array)

Appends all elements in other_array to the end of array. Returns 0 on success and -1 on error.

json_array_foreach(array, index, value)

Iterate over every element of array, running the block of code that follows each time with the proper values set to variables index and value, of types size_t and json_t * respectively. Example:

/* array is a JSON array */
size_t index;
json_t *value;

json_array_foreach(array, index, value) {
    /* block of code that uses index and value */
}

The items are returned in increasing index order.

This macro expands to an ordinary for statement upon preprocessing, so its performance is equivalent to that of hand-written code using the array access functions. The main advantage of this macro is that it abstracts away the complexity, and makes for more concise and readable code.

New in version 2.5.

Object

A JSON object is a dictionary of key-value pairs, where the key is a Unicode string and the value is any JSON value.

Even though null bytes are allowed in string values, they are not allowed in object keys.

json_t *json_object(void)
Return value: New reference.

Returns a new JSON object, or NULL on error. Initially, the object is empty.

size_t json_object_size(const json_t *object)

Returns the number of elements in object, or 0 if object is not a JSON object.

json_t *json_object_get(const json_t *object, const char *key)
Return value: Borrowed reference.

Get a value corresponding to key from object. Returns NULL if key is not found and on error.

int json_object_set(json_t *object, const char *key, json_t *value)

Set the value of key to value in object. key must be a valid null terminated UTF-8 encoded Unicode string. If there already is a value for key, it is replaced by the new value. Returns 0 on success and -1 on error.

int json_object_set_nocheck(json_t *object, const char *key, json_t *value)

Like json_object_set(), but doesn’t check that key is valid UTF-8. Use this function only if you are certain that this really is the case (e.g. you have already checked it by other means).

int json_object_set_new(json_t *object, const char *key, json_t *value)

Like json_object_set() but steals the reference to value. This is useful when value is newly created and not used after the call.

int json_object_set_new_nocheck(json_t *object, const char *key, json_t *value)

Like json_object_set_new(), but doesn’t check that key is valid UTF-8. Use this function only if you are certain that this really is the case (e.g. you have already checked it by other means).

int json_object_del(json_t *object, const char *key)

Delete key from object if it exists. Returns 0 on success, or -1 if key was not found. The reference count of the removed value is decremented.

int json_object_clear(json_t *object)

Remove all elements from object. Returns 0 on success and -1 if object is not a JSON object. The reference count of all removed values are decremented.

int json_object_update(json_t *object, json_t *other)

Update object with the key-value pairs from other, overwriting existing keys. Returns 0 on success or -1 on error.

int json_object_update_existing(json_t *object, json_t *other)

Like json_object_update(), but only the values of existing keys are updated. No new keys are created. Returns 0 on success or -1 on error.

New in version 2.3.

int json_object_update_missing(json_t *object, json_t *other)

Like json_object_update(), but only new keys are created. The value of any existing key is not changed. Returns 0 on success or -1 on error.

New in version 2.3.

json_object_foreach(object, key, value)

Iterate over every key-value pair of object, running the block of code that follows each time with the proper values set to variables key and value, of types const char * and json_t * respectively. Example:

/* obj is a JSON object */
const char *key;
json_t *value;

json_object_foreach(obj, key, value) {
    /* block of code that uses key and value */
}

The items are returned in the order they were inserted to the object.

Note: It’s not safe to call json_object_del(object, key) during iteration. If you need to, use json_object_foreach_safe() instead.

This macro expands to an ordinary for statement upon preprocessing, so its performance is equivalent to that of hand-written iteration code using the object iteration protocol (see below). The main advantage of this macro is that it abstracts away the complexity behind iteration, and makes for more concise and readable code.

New in version 2.3.

json_object_foreach_safe(object, tmp, key, value)

Like json_object_foreach(), but it’s safe to call json_object_del(object, key) during iteration. You need to pass an extra void * parameter tmp that is used for temporary storage.

New in version 2.8.

The following functions can be used to iterate through all key-value pairs in an object. The items are returned in the order they were inserted to the object.

void *json_object_iter(json_t *object)

Returns an opaque iterator which can be used to iterate over all key-value pairs in object, or NULL if object is empty.

void *json_object_iter_at(json_t *object, const char *key)

Like json_object_iter(), but returns an iterator to the key-value pair in object whose key is equal to key, or NULL if key is not found in object. Iterating forward to the end of object only yields all key-value pairs of the object if key happens to be the first key in the underlying hash table.

void *json_object_iter_next(json_t *object, void *iter)

Returns an iterator pointing to the next key-value pair in object after iter, or NULL if the whole object has been iterated through.

const char *json_object_iter_key(void *iter)

Extract the associated key from iter.

json_t *json_object_iter_value(void *iter)
Return value: Borrowed reference.

Extract the associated value from iter.

int json_object_iter_set(json_t *object, void *iter, json_t *value)

Set the value of the key-value pair in object, that is pointed to by iter, to value.

int json_object_iter_set_new(json_t *object, void *iter, json_t *value)

Like json_object_iter_set(), but steals the reference to value. This is useful when value is newly created and not used after the call.

void *json_object_key_to_iter(const char *key)

Like json_object_iter_at(), but much faster. Only works for values returned by json_object_iter_key(). Using other keys will lead to segfaults. This function is used internally to implement json_object_foreach(). Example:

/* obj is a JSON object */
const char *key;
json_t *value;

void *iter = json_object_iter(obj);
while(iter)
{
    key = json_object_iter_key(iter);
    value = json_object_iter_value(iter);
    /* use key and value ... */
    iter = json_object_iter_next(obj, iter);
}

New in version 2.3.

void json_object_seed(size_t seed)

Seed the hash function used in Jansson’s hashtable implementation. The seed is used to randomize the hash function so that an attacker cannot control its output.

If seed is 0, Jansson generates the seed itself by reading random data from the operating system’s entropy sources. If no entropy sources are available, falls back to using a combination of the current timestamp (with microsecond precision if possible) and the process ID.

If called at all, this function must be called before any calls to json_object(), either explicit or implicit. If this function is not called by the user, the first call to json_object() (either explicit or implicit) seeds the hash function. See Thread safety for notes on thread safety.

If repeatable results are required, for e.g. unit tests, the hash function can be “unrandomized” by calling json_object_seed() with a constant value on program startup, e.g. json_object_seed(1).

New in version 2.6.

Error reporting

Jansson uses a single struct type to pass error information to the user. See sections Decoding, Building Values and Parsing and Validating Values for functions that pass error information using this struct.

json_error_t
char text[]

The error message (in UTF-8), or an empty string if a message is not available.

char source[]

Source of the error. This can be (a part of) the file name or a special identifier in angle brackets (e.g. <string>).

int line

The line number on which the error occurred.

int column

The column on which the error occurred. Note that this is the character column, not the byte column, i.e. a multibyte UTF-8 character counts as one column.

int position

The position in bytes from the start of the input. This is useful for debugging Unicode encoding problems.

The normal use of json_error_t is to allocate it on the stack, and pass a pointer to a function. Example:

int main() {
    json_t *json;
    json_error_t error;

    json = json_load_file("/path/to/file.json", 0, &error);
    if(!json) {
        /* the error variable contains error information */
    }
    ...
}

Also note that if the call succeeded (json != NULL in the above example), the contents of error are generally left unspecified. The decoding functions write to the position member also on success. See Decoding for more info.

All functions also accept NULL as the json_error_t pointer, in which case no error information is returned to the caller.

Encoding

This sections describes the functions that can be used to encode values to JSON. By default, only objects and arrays can be encoded directly, since they are the only valid root values of a JSON text. To encode any JSON value, use the JSON_ENCODE_ANY flag (see below).

By default, the output has no newlines, and spaces are used between array and object elements for a readable output. This behavior can be altered by using the JSON_INDENT and JSON_COMPACT flags described below. A newline is never appended to the end of the encoded JSON data.

Each function takes a flags parameter that controls some aspects of how the data is encoded. Its default value is 0. The following macros can be ORed together to obtain flags.

JSON_INDENT(n)

Pretty-print the result, using newlines between array and object items, and indenting with n spaces. The valid range for n is between 0 and 31 (inclusive), other values result in an undefined output. If JSON_INDENT is not used or n is 0, no newlines are inserted between array and object items.

The JSON_MAX_INDENT constant defines the maximum indentation that can be used, and its value is 31.

Changed in version 2.7: Added JSON_MAX_INDENT.

JSON_COMPACT
This flag enables a compact representation, i.e. sets the separator between array and object items to "," and between object keys and values to ":". Without this flag, the corresponding separators are ", " and ": " for more readable output.
JSON_ENSURE_ASCII
If this flag is used, the output is guaranteed to consist only of ASCII characters. This is achieved by escaping all Unicode characters outside the ASCII range.
JSON_SORT_KEYS
If this flag is used, all the objects in output are sorted by key. This is useful e.g. if two JSON texts are diffed or visually compared.
JSON_PRESERVE_ORDER

Deprecated since version 2.8: Order of object keys is always preserved.

Prior to version 2.8: If this flag is used, object keys in the output are sorted into the same order in which they were first inserted to the object. For example, decoding a JSON text and then encoding with this flag preserves the order of object keys.

JSON_ENCODE_ANY

Specifying this flag makes it possible to encode any JSON value on its own. Without it, only objects and arrays can be passed as the json value to the encoding functions.

Note: Encoding any value may be useful in some scenarios, but it’s generally discouraged as it violates strict compatibility with RFC 4627. If you use this flag, don’t expect interoperability with other JSON systems.

New in version 2.1.

JSON_ESCAPE_SLASH

Escape the / characters in strings with \/.

New in version 2.4.

JSON_REAL_PRECISION(n)

Output all real numbers with at most n digits of precision. The valid range for n is between 0 and 31 (inclusive), and other values result in an undefined behavior.

By default, the precision is 17, to correctly and losslessly encode all IEEE 754 double precision floating point numbers.

New in version 2.7.

JSON_EMBED

If this flag is used, the opening and closing characters of the top-level array (‘[‘, ‘]’) or object (‘{‘, ‘}’) are omitted during encoding. This flag is useful when concatenating multiple arrays or objects into a stream.

New in version 2.10.

These functions output UTF-8:

char *json_dumps(const json_t *json, size_t flags)

Returns the JSON representation of json as a string, or NULL on error. flags is described above. The return value must be freed by the caller using free().

size_t json_dumpb(const json_t *json, char *buffer, size_t size, size_t flags)

Writes the JSON representation of json to the buffer of size bytes. Returns the number of bytes that would be written or 0 on error. flags is described above. buffer is not null-terminated.

This function never writes more than size bytes. If the return value is greater than size, the contents of the buffer are undefined. This behavior enables you to specify a NULL buffer to determine the length of the encoding. For example:

size_t size = json_dumpb(json, NULL, 0, 0);
if (size == 0)
    return -1;

char *buf = alloca(size);

size = json_dumpb(json, buf, size, 0);

New in version 2.10.

int json_dumpf(const json_t *json, FILE *output, size_t flags)

Write the JSON representation of json to the stream output. flags is described above. Returns 0 on success and -1 on error. If an error occurs, something may have already been written to output. In this case, the output is undefined and most likely not valid JSON.

int json_dumpfd(const json_t *json, int output, size_t flags)

Write the JSON representation of json to the stream output. flags is described above. Returns 0 on success and -1 on error. If an error occurs, something may have already been written to output. In this case, the output is undefined and most likely not valid JSON.

It is important to note that this function can only succeed on stream file descriptors (such as SOCK_STREAM). Using this function on a non-stream file descriptor will result in undefined behavior. For non-stream file descriptors, see instead json_dumpb().

This function requires POSIX and fails on all non-POSIX systems.

New in version 2.10.

int json_dump_file(const json_t *json, const char *path, size_t flags)

Write the JSON representation of json to the file path. If path already exists, it is overwritten. flags is described above. Returns 0 on success and -1 on error.

json_dump_callback_t

A typedef for a function that’s called by json_dump_callback():

typedef int (*json_dump_callback_t)(const char *buffer, size_t size, void *data);

buffer points to a buffer containing a chunk of output, size is the length of the buffer, and data is the corresponding json_dump_callback() argument passed through.

On error, the function should return -1 to stop the encoding process. On success, it should return 0.

New in version 2.2.

int json_dump_callback(const json_t *json, json_dump_callback_t callback, void *data, size_t flags)

Call callback repeatedly, passing a chunk of the JSON representation of json each time. flags is described above. Returns 0 on success and -1 on error.

New in version 2.2.

Decoding

This sections describes the functions that can be used to decode JSON text to the Jansson representation of JSON data. The JSON specification requires that a JSON text is either a serialized array or object, and this requirement is also enforced with the following functions. In other words, the top level value in the JSON text being decoded must be either array or object. To decode any JSON value, use the JSON_DECODE_ANY flag (see below).

See RFC Conformance for a discussion on Jansson’s conformance to the JSON specification. It explains many design decisions that affect especially the behavior of the decoder.

Each function takes a flags parameter that can be used to control the behavior of the decoder. Its default value is 0. The following macros can be ORed together to obtain flags.

JSON_REJECT_DUPLICATES

Issue a decoding error if any JSON object in the input text contains duplicate keys. Without this flag, the value of the last occurrence of each key ends up in the result. Key equivalence is checked byte-by-byte, without special Unicode comparison algorithms.

New in version 2.1.

JSON_DECODE_ANY

By default, the decoder expects an array or object as the input. With this flag enabled, the decoder accepts any valid JSON value.

Note: Decoding any value may be useful in some scenarios, but it’s generally discouraged as it violates strict compatibility with RFC 4627. If you use this flag, don’t expect interoperability with other JSON systems.

New in version 2.3.

JSON_DISABLE_EOF_CHECK

By default, the decoder expects that its whole input constitutes a valid JSON text, and issues an error if there’s extra data after the otherwise valid JSON input. With this flag enabled, the decoder stops after decoding a valid JSON array or object, and thus allows extra data after the JSON text.

Normally, reading will stop when the last ] or } in the JSON input is encountered. If both JSON_DISABLE_EOF_CHECK and JSON_DECODE_ANY flags are used, the decoder may read one extra UTF-8 code unit (up to 4 bytes of input). For example, decoding 4true correctly decodes the integer 4, but also reads the t. For this reason, if reading multiple consecutive values that are not arrays or objects, they should be separated by at least one whitespace character.

New in version 2.1.

JSON_DECODE_INT_AS_REAL

JSON defines only one number type. Jansson distinguishes between ints and reals. For more information see Real vs. Integer. With this flag enabled the decoder interprets all numbers as real values. Integers that do not have an exact double representation will silently result in a loss of precision. Integers that cause a double overflow will cause an error.

New in version 2.5.

JSON_ALLOW_NUL

Allow \u0000 escape inside string values. This is a safety measure; If you know your input can contain null bytes, use this flag. If you don’t use this flag, you don’t have to worry about null bytes inside strings unless you explicitly create themselves by using e.g. json_stringn() or s# format specifier for json_pack().

Object keys cannot have embedded null bytes even if this flag is used.

New in version 2.6.

Each function also takes an optional json_error_t parameter that is filled with error information if decoding fails. It’s also updated on success; the number of bytes of input read is written to its position field. This is especially useful when using JSON_DISABLE_EOF_CHECK to read multiple consecutive JSON texts.

New in version 2.3: Number of bytes of input read is written to the position field of the json_error_t structure.

If no error or position information is needed, you can pass NULL.

json_t *json_loads(const char *input, size_t flags, json_error_t *error)
Return value: New reference.

Decodes the JSON string input and returns the array or object it contains, or NULL on error, in which case error is filled with information about the error. flags is described above.

json_t *json_loadb(const char *buffer, size_t buflen, size_t flags, json_error_t *error)
Return value: New reference.

Decodes the JSON string buffer, whose length is buflen, and returns the array or object it contains, or NULL on error, in which case error is filled with information about the error. This is similar to json_loads() except that the string doesn’t need to be null-terminated. flags is described above.

New in version 2.1.

json_t *json_loadf(FILE *input, size_t flags, json_error_t *error)
Return value: New reference.

Decodes the JSON text in stream input and returns the array or object it contains, or NULL on error, in which case error is filled with information about the error. flags is described above.

This function will start reading the input from whatever position the input file was in, without attempting to seek first. If an error occurs, the file position will be left indeterminate. On success, the file position will be at EOF, unless JSON_DISABLE_EOF_CHECK flag was used. In this case, the file position will be at the first character after the last ] or } in the JSON input. This allows calling json_loadf() on the same FILE object multiple times, if the input consists of consecutive JSON texts, possibly separated by whitespace.

json_t *json_loadfd(int input, size_t flags, json_error_t *error)
Return value: New reference.

Decodes the JSON text in stream input and returns the array or object it contains, or NULL on error, in which case error is filled with information about the error. flags is described above.

This function will start reading the input from whatever position the input file descriptor was in, without attempting to seek first. If an error occurs, the file position will be left indeterminate. On success, the file position will be at EOF, unless JSON_DISABLE_EOF_CHECK flag was used. In this case, the file descriptor’s position will be at the first character after the last ] or } in the JSON input. This allows calling json_loadfd() on the same file descriptor multiple times, if the input consists of consecutive JSON texts, possibly separated by whitespace.

It is important to note that this function can only succeed on stream file descriptors (such as SOCK_STREAM). Using this function on a non-stream file descriptor will result in undefined behavior. For non-stream file descriptors, see instead json_loadb().

This function requires POSIX and fails on all non-POSIX systems.

New in version 2.10.

json_t *json_load_file(const char *path, size_t flags, json_error_t *error)
Return value: New reference.

Decodes the JSON text in file path and returns the array or object it contains, or NULL on error, in which case error is filled with information about the error. flags is described above.

json_load_callback_t

A typedef for a function that’s called by json_load_callback() to read a chunk of input data:

typedef size_t (*json_load_callback_t)(void *buffer, size_t buflen, void *data);

buffer points to a buffer of buflen bytes, and data is the corresponding json_load_callback() argument passed through.

On success, the function should return the number of bytes read; a returned value of 0 indicates that no data was read and that the end of file has been reached. On error, the function should return (size_t)-1 to abort the decoding process.

New in version 2.4.

json_t *json_load_callback(json_load_callback_t callback, void *data, size_t flags, json_error_t *error)
Return value: New reference.

Decodes the JSON text produced by repeated calls to callback, and returns the array or object it contains, or NULL on error, in which case error is filled with information about the error. data is passed through to callback on each call. flags is described above.

New in version 2.4.

Building Values

This section describes functions that help to create, or pack, complex JSON values, especially nested objects and arrays. Value building is based on a format string that is used to tell the functions about the expected arguments.

For example, the format string "i" specifies a single integer value, while the format string "[ssb]" or the equivalent "[s, s, b]" specifies an array value with two strings and a boolean as its items:

/* Create the JSON integer 42 */
json_pack("i", 42);

/* Create the JSON array ["foo", "bar", true] */
json_pack("[ssb]", "foo", "bar", 1);

Here’s the full list of format specifiers. The type in parentheses denotes the resulting JSON type, and the type in brackets (if any) denotes the C type that is expected as the corresponding argument or arguments.

s (string) [const char *]
Convert a null terminated UTF-8 string to a JSON string.
s? (string) [const char *]

Like s, but if the argument is NULL, output a JSON null value.

New in version 2.8.

s# (string) [const char *, int]

Convert a UTF-8 buffer of a given length to a JSON string.

New in version 2.5.

s% (string) [const char *, size_t]

Like s# but the length argument is of type size_t.

New in version 2.6.

+ [const char *]

Like s, but concatenate to the previous string. Only valid after s, s#, + or +#.

New in version 2.5.

+# [const char *, int]

Like s#, but concatenate to the previous string. Only valid after s, s#, + or +#.

New in version 2.5.

+% (string) [const char *, size_t]

Like +# but the length argument is of type size_t.

New in version 2.6.

n (null)
Output a JSON null value. No argument is consumed.
b (boolean) [int]
Convert a C int to JSON boolean value. Zero is converted to false and non-zero to true.
i (integer) [int]
Convert a C int to JSON integer.
I (integer) [json_int_t]
Convert a C json_int_t to JSON integer.
f (real) [double]
Convert a C double to JSON real.
o (any value) [json_t *]
Output any given JSON value as-is. If the value is added to an array or object, the reference to the value passed to o is stolen by the container.
O (any value) [json_t *]
Like o, but the argument’s reference count is incremented. This is useful if you pack into an array or object and want to keep the reference for the JSON value consumed by O to yourself.
o?, O? (any value) [json_t *]

Like o and O?, respectively, but if the argument is NULL, output a JSON null value.

New in version 2.8.

[fmt] (array)
Build an array with contents from the inner format string. fmt may contain objects and arrays, i.e. recursive value building is supported.
{fmt} (object)
Build an object with contents from the inner format string fmt. The first, third, etc. format specifier represent a key, and must be a string (see s, s#, + and +# above), as object keys are always strings. The second, fourth, etc. format specifier represent a value. Any value may be an object or array, i.e. recursive value building is supported.

Whitespace, : and , are ignored.

json_t *json_pack(const char *fmt, ...)
Return value: New reference.

Build a new JSON value according to the format string fmt. For each format specifier (except for {}[]n), one or more arguments are consumed and used to build the corresponding value. Returns NULL on error.

json_t *json_pack_ex(json_error_t *error, size_t flags, const char *fmt, ...)
json_t *json_vpack_ex(json_error_t *error, size_t flags, const char *fmt, va_list ap)
Return value: New reference.

Like json_pack(), but an in the case of an error, an error message is written to error, if it’s not NULL. The flags parameter is currently unused and should be set to 0.

As only the errors in format string (and out-of-memory errors) can be caught by the packer, these two functions are most likely only useful for debugging format strings.

More examples:

/* Build an empty JSON object */
json_pack("{}");

/* Build the JSON object {"foo": 42, "bar": 7} */
json_pack("{sisi}", "foo", 42, "bar", 7);

/* Like above, ':', ',' and whitespace are ignored */
json_pack("{s:i, s:i}", "foo", 42, "bar", 7);

/* Build the JSON array [[1, 2], {"cool": true}] */
json_pack("[[i,i],{s:b}]", 1, 2, "cool", 1);

/* Build a string from a non-null terminated buffer */
char buffer[4] = {'t', 'e', 's', 't'};
json_pack("s#", buffer, 4);

/* Concatenate strings together to build the JSON string "foobarbaz" */
json_pack("s++", "foo", "bar", "baz");

Parsing and Validating Values

This section describes functions that help to validate complex values and extract, or unpack, data from them. Like building values, this is also based on format strings.

While a JSON value is unpacked, the type specified in the format string is checked to match that of the JSON value. This is the validation part of the process. In addition to this, the unpacking functions can also check that all items of arrays and objects are unpacked. This check be enabled with the format specifier ! or by using the flag JSON_STRICT. See below for details.

Here’s the full list of format specifiers. The type in parentheses denotes the JSON type, and the type in brackets (if any) denotes the C type whose address should be passed.

s (string) [const char *]
Convert a JSON string to a pointer to a null terminated UTF-8 string. The resulting string is extracted by using json_string_value() internally, so it exists as long as there are still references to the corresponding JSON string.
s% (string) [const char *, size_t *]

Convert a JSON string to a pointer to a null terminated UTF-8 string and its length.

New in version 2.6.

n (null)
Expect a JSON null value. Nothing is extracted.
b (boolean) [int]
Convert a JSON boolean value to a C int, so that true is converted to 1 and false to 0.
i (integer) [int]
Convert a JSON integer to C int.
I (integer) [json_int_t]
Convert a JSON integer to C json_int_t.
f (real) [double]
Convert a JSON real to C double.
F (integer or real) [double]
Convert a JSON number (integer or real) to C double.
o (any value) [json_t *]
Store a JSON value with no conversion to a json_t pointer.
O (any value) [json_t *]
Like O, but the JSON value’s reference count is incremented.
[fmt] (array)
Convert each item in the JSON array according to the inner format string. fmt may contain objects and arrays, i.e. recursive value extraction is supported.
{fmt} (object)

Convert each item in the JSON object according to the inner format string fmt. The first, third, etc. format specifier represent a key, and must be s. The corresponding argument to unpack functions is read as the object key. The second fourth, etc. format specifier represent a value and is written to the address given as the corresponding argument. Note that every other argument is read from and every other is written to.

fmt may contain objects and arrays as values, i.e. recursive value extraction is supported.

New in version 2.3: Any s representing a key may be suffixed with a ? to make the key optional. If the key is not found, nothing is extracted. See below for an example.

!
This special format specifier is used to enable the check that all object and array items are accessed, on a per-value basis. It must appear inside an array or object as the last format specifier before the closing bracket or brace. To enable the check globally, use the JSON_STRICT unpacking flag.
*
This special format specifier is the opposite of !. If the JSON_STRICT flag is used, * can be used to disable the strict check on a per-value basis. It must appear inside an array or object as the last format specifier before the closing bracket or brace.

Whitespace, : and , are ignored.

int json_unpack(json_t *root, const char *fmt, ...)

Validate and unpack the JSON value root according to the format string fmt. Returns 0 on success and -1 on failure.

int json_unpack_ex(json_t *root, json_error_t *error, size_t flags, const char *fmt, ...)
int json_vunpack_ex(json_t *root, json_error_t *error, size_t flags, const char *fmt, va_list ap)

Validate and unpack the JSON value root according to the format string fmt. If an error occurs and error is not NULL, write error information to error. flags can be used to control the behaviour of the unpacker, see below for the flags. Returns 0 on success and -1 on failure.

Note

The first argument of all unpack functions is json_t *root instead of const json_t *root, because the use of O format specifier causes the reference count of root, or some value reachable from root, to be increased. Furthermore, the o format specifier may be used to extract a value as-is, which allows modifying the structure or contents of a value reachable from root.

If the O and o format specifiers are not used, it’s perfectly safe to cast a const json_t * variable to plain json_t * when used with these functions.

The following unpacking flags are available:

JSON_STRICT
Enable the extra validation step checking that all object and array items are unpacked. This is equivalent to appending the format specifier ! to the end of every array and object in the format string.
JSON_VALIDATE_ONLY
Don’t extract any data, just validate the JSON value against the given format string. Note that object keys must still be specified after the format string.

Examples:

/* root is the JSON integer 42 */
int myint;
json_unpack(root, "i", &myint);
assert(myint == 42);

/* root is the JSON object {"foo": "bar", "quux": true} */
const char *str;
int boolean;
json_unpack(root, "{s:s, s:b}", "foo", &str, "quux", &boolean);
assert(strcmp(str, "bar") == 0 && boolean == 1);

/* root is the JSON array [[1, 2], {"baz": null} */
json_error_t error;
json_unpack_ex(root, &error, JSON_VALIDATE_ONLY, "[[i,i], {s:n}]", "baz");
/* returns 0 for validation success, nothing is extracted */

/* root is the JSON array [1, 2, 3, 4, 5] */
int myint1, myint2;
json_unpack(root, "[ii!]", &myint1, &myint2);
/* returns -1 for failed validation */

/* root is an empty JSON object */
int myint = 0, myint2 = 0, myint3 = 0;
json_unpack(root, "{s?i, s?[ii]}",
            "foo", &myint1,
            "bar", &myint2, &myint3);
/* myint1, myint2 or myint3 is no touched as "foo" and "bar" don't exist */

Equality

Testing for equality of two JSON values cannot, in general, be achieved using the == operator. Equality in the terms of the == operator states that the two json_t pointers point to exactly the same JSON value. However, two JSON values can be equal not only if they are exactly the same value, but also if they have equal “contents”:

  • Two integer or real values are equal if their contained numeric values are equal. An integer value is never equal to a real value, though.
  • Two strings are equal if their contained UTF-8 strings are equal, byte by byte. Unicode comparison algorithms are not implemented.
  • Two arrays are equal if they have the same number of elements and each element in the first array is equal to the corresponding element in the second array.
  • Two objects are equal if they have exactly the same keys and the value for each key in the first object is equal to the value of the corresponding key in the second object.
  • Two true, false or null values have no “contents”, so they are equal if their types are equal. (Because these values are singletons, their equality can actually be tested with ==.)
int json_equal(json_t *value1, json_t *value2)

Returns 1 if value1 and value2 are equal, as defined above. Returns 0 if they are unequal or one or both of the pointers are NULL.

Copying

Because of reference counting, passing JSON values around doesn’t require copying them. But sometimes a fresh copy of a JSON value is needed. For example, if you need to modify an array, but still want to use the original afterwards, you should take a copy of it first.

Jansson supports two kinds of copying: shallow and deep. There is a difference between these methods only for arrays and objects. Shallow copying only copies the first level value (array or object) and uses the same child values in the copied value. Deep copying makes a fresh copy of the child values, too. Moreover, all the child values are deep copied in a recursive fashion.

Copying objects preserves the insertion order of keys.

json_t *json_copy(json_t *value)
Return value: New reference.

Returns a shallow copy of value, or NULL on error.

json_t *json_deep_copy(const json_t *value)
Return value: New reference.

Returns a deep copy of value, or NULL on error.

Custom Memory Allocation

By default, Jansson uses malloc() and free() for memory allocation. These functions can be overridden if custom behavior is needed.

json_malloc_t

A typedef for a function pointer with malloc()‘s signature:

typedef void *(*json_malloc_t)(size_t);
json_free_t

A typedef for a function pointer with free()‘s signature:

typedef void (*json_free_t)(void *);
void json_set_alloc_funcs(json_malloc_t malloc_fn, json_free_t free_fn)

Use malloc_fn instead of malloc() and free_fn instead of free(). This function has to be called before any other Jansson’s API functions to ensure that all memory operations use the same functions.

void json_get_alloc_funcs(json_malloc_t *malloc_fn, json_free_t *free_fn)

Fetch the current malloc_fn and free_fn used. Either parameter may be NULL.

New in version 2.8.

Examples:

Circumvent problems with different CRT heaps on Windows by using application’s malloc() and free():

json_set_alloc_funcs(malloc, free);

Use the Boehm’s conservative garbage collector for memory operations:

json_set_alloc_funcs(GC_malloc, GC_free);

Allow storing sensitive data (e.g. passwords or encryption keys) in JSON structures by zeroing all memory when freed:

static void *secure_malloc(size_t size)
{
    /* Store the memory area size in the beginning of the block */
    void *ptr = malloc(size + 8);
    *((size_t *)ptr) = size;
    return ptr + 8;
}

static void secure_free(void *ptr)
{
    size_t size;

    ptr -= 8;
    size = *((size_t *)ptr);

    guaranteed_memset(ptr, 0, size + 8);
    free(ptr);
}

int main()
{
    json_set_alloc_funcs(secure_malloc, secure_free);
    /* ... */
}

For more information about the issues of storing sensitive data in memory, see http://www.dwheeler.com/secure-programs/Secure-Programs-HOWTO/protect-secrets.html. The page also explains the guaranteed_memset() function used in the example and gives a sample implementation for it.

Changes in Jansson

Version 2.10

Released 2017-03-02

  • New features:
    • Add JSON_EMBED encoding flag allowing arrays and objects to be encoded into existing streams (#329).
    • Add json_dumpb() function for dumping to a pre-allocated buffer (#328).
    • Add json_dumpfd() and json_loadfd() functions for dumping to streaming file descriptors (#328).
    • Add support for parsing buffers larger than 2GB (#309).
  • Build:
    • Fix CMake build when LONG_LONG_INT is defined as “” (#321)
  • Other:
    • Internal code cleanup (#311, #314)

Version 2.9

Released 2016-09-18

  • New features:
    • Add json_auto_t to automatically decref a value that goes out of scope. Available only on GCC and Clang. (#301)
  • Build:
    • Fix CMake build (at least on Linux) by removing conflicting jansson_config.h from the distribution (#306)
    • Change CMake install target generation to be optional (#305)
  • Documentation:
    • Small documentation fixes.

Version 2.8

Released 2016-08-30

  • New features:
  • Bug fixes:
    • Fix a crash when parsing inputs consisting of very deeply nested arrays or objects (#282, #284).
    • Never convert numbers to integers in the parser when JSON_DECODE_INT_AS_REAL is set. This fixes error messages for overflowing numbers when JSON_DECODE_INT_AS_REAL is set (#212).
    • Fix a use-after-free in json_pack() error handling.
    • Fix subnormal number parsing on mingw32.
    • Handle out-of-memory situations gracefully in the hashtable implementation (#298).
  • Build:
    • Fix build with CMake on all versions of Visual Studio up to 2015 (#262, #289).
    • Fix pkgconfig libdir when using CMake (#268).
    • Fix CMake config for static CRT builds on Windows (#206).
    • Fix warnings on LLVM 6.0 targeting iOS arm64 (#208).
    • Add coverlls.io support via Travis for a nice test coverage badge (#211).
    • Don’t expect jansson_config.h to be in the compiler’s include path (#209).
    • Add a build-time option to set initial hashtable size (#213).
    • Use snprintf and strncpy in place of sprintf and strcpy to silence linker warnings on OpenBSD (#233).
  • Documentation:
    • Fix various typos in documentation, and a broken link (#258).
    • Add an example program in examples/ (#214, #217).
    • Fix building of documentation man pages (#207).
    • Document the fact that copying objects doesn’t preserve the insertion order of keys (#237).
  • Tests:
    • Don’t use the nonstandard __FUNCTION__ macro in tests.
    • Use expr instead of $((...)) in shell scripts for Solaris 10 compatibility.
    • Disable Visual Studio warning C4756 when triggered deliberately in tests (#216).
    • Other minor fixes (#221, #248).
  • Other changes:
    • List all unrecognized object keys when strict unpacking fails (#263).
    • Alter the order of the members of the hashtable_pair struct for easier debugging.
    • Minor performance improvement to json_dump() and friends (#234).
    • Minor style fixes (#255, #257).

Version 2.7

Released 2014-10-02

  • New features:
  • Bug fixes:
    • Some malformed \uNNNN escapes could crash the decoder with an assertion failure.
    • Avoid integer overflows with very long strings in UTF-8 decoder and hashtable.
    • Check for NULL key in json_object_get() and json_object_del() (#151).
    • Enhance hashtable seeding on Windows (#162).
    • json_unpack(): Allow mixing JSON_STRICT with optional keys (#162, #163).
    • Fix int/int32 mismatch (#142).
    • Parse subnormal numbers correctly (#202).
  • Build:
    • Remove VS2010 build files. CMake should be used on Windows instead (#165).
    • Fix CMake build flags for MinGW (#193).
    • Add CMake config files for find_package. Rename config.h to jansson_private_config.h (#157, #159).
    • Make Valgrind checks work with CMake (#160).
    • Fix feature checks to use correct __ATOMIC flags.
    • Fix CMake checks for uint16_t and uint8_t support (#177).
    • Make Jansson build on SmartOS/Solaris (#171).
    • Work around a GCC bug on Solaris (#175).
    • Fix autoreconf on Debian (#182).
    • Don’t use GNU make specific export for global AM_CFLAGS (#203, #204).
    • Fix building on Android using the supplied Android.mk (#166, #174).
    • Android.mk: Add -DHAVE_STDINT_H to LOCAL_CFLAGS (#200).
  • Documentation:
    • Document JANSSON_BUILD_SHARED_LIBS CMake option (#187).
  • Tests:
    • Close file handles correctly (#198).
  • Other changes:
    • \uNNNN escapes are now encoded in upper case for better readability.
    • Enable usage of AddressSanitizer (#180).

Version 2.6

Released 2014-02-11

  • Security:
    • CVE-2013-6401: The hash function used by the hashtable implementation has been changed, and is automatically seeded with random data when the first JSON object is created. This prevents an attacker from causing large JSON objects with specially crafted keys perform poorly.
  • New features:
  • Bug fixes:
    • Include CMake specific files in the release tarball.
  • Documentation:
    • Fix tutorial source to send a User-Agent header, which is now required by the GitHub API.
    • Set all memory to zero in secure_free() example.

Version 2.5

Released 2013-09-19

  • New features:
  • Bug fixes:
    • json_dumps() and friends: Don’t crash if json is NULL and JSON_ENCODE_ANY is set.
    • Fix a theoretical integer overflow in jsonp_strdup().
    • Fix l_isxdigit() macro (#97).
    • Fix an off-by-one error in json_array_remove().
  • Build:
    • Support CMake in addition to GNU Autotools (#106, #107, #112, #115, #120, #127).
    • Support building for Android (#109).
    • Don’t use -Werror by default.
    • Support building and testing with VPATH (#93).
    • Fix compilation when NDEBUG is defined (#128)
  • Tests:
    • Fix a refleak in test/bin/json_process.c.
  • Documentation:
    • Clarify the return value of json_load_callback_t().
    • Document how to circumvent problems with separate heaps on Windows.
    • Fix memory leaks and warnings in github_commits.c.
    • Use json_decref() properly in tutorial.
  • Other:
    • Make it possible to forward declare struct json_t.

Version 2.4

Released 2012-09-23

  • New features:
    • Add json_boolean() macro that returns the JSON true or false value based on its argument (#86).
    • Add json_load_callback() that calls a callback function repeatedly to read the JSON input (#57).
    • Add JSON_ESCAPE_SLASH encoding flag to escape all occurences of / with \/.
  • Bug fixes:
    • Check for and reject NaN and Inf values for reals. Encoding these values resulted in invalid JSON.
    • Fix json_real_set() to return -1 on error.
  • Build:
    • Jansson now builds on Windows with Visual Studio 2010, and includes solution and project files in win32/vs2010/ directory.
    • Fix build warnings (#77, #78).
    • Add -no-undefined to LDFLAGS (#90).
  • Tests:
    • Fix the symbol exports test on Linux/PPC64 (#88).
  • Documentation:
    • Fix typos (#73, #84).

Version 2.3.1

Released 2012-04-20

  • Build issues:
    • Only use long long if strtoll() is also available.
  • Documentation:
    • Fix the names of library version constants in documentation. (#52)
    • Change the tutorial to use GitHub API v3. (#65)
  • Tests:
    • Make some tests locale independent. (#51)
    • Distribute the library exports test in the tarball.
    • Make test run on shells that don’t support the export FOO=bar syntax.

Version 2.3

Released 2012-01-27

  • New features:
    • json_unpack() and friends: Add support for optional object keys with the {s?o} syntax.
    • Add json_object_update_existing() and json_object_update_missing(), for updating only existing keys or only adding missing keys to an object. (#37)
    • Add json_object_foreach() for more convenient iteration over objects. (#45, #46)
    • When decoding JSON, write the number of bytes that were read from input to error.position also on success. This is handy with JSON_DISABLE_EOF_CHECK.
    • Add support for decoding any JSON value, not just arrays or objects. The support is enabled with the new JSON_DECODE_ANY flag. Patch by Andrea Marchesini. (#4)
  • Bug fixes
    • Avoid problems with object’s serial number growing too big. (#40, #41)
    • Decoding functions now return NULL if the first argument is NULL. Patch by Andrea Marchesini.
    • Include jansson_config.h.win32 in the distribution tarball.
    • Remove + and leading zeros from exponents in the encoder. (#39)
    • Make Jansson build and work on MinGW. (#39, #38)
  • Documentation
    • Note that the same JSON values must not be encoded in parallel by separate threads. (#42)
    • Document MinGW support.

Version 2.2.1

Released 2011-10-06

  • Bug fixes:
    • Fix real number encoding and decoding under non-C locales. (#32)
    • Fix identifier decoding under non-UTF-8 locales. (#35)
    • json_load_file(): Open the input file in binary mode for maximum compatiblity.
  • Documentation:
    • Clarify the lifecycle of the result of the s fromat of json_unpack(). (#31)
    • Add some portability info. (#36)
    • Little clarifications here and there.
  • Other:
    • Some style fixes, issues detected by static analyzers.

Version 2.2

Released 2011-09-03

  • New features:
  • Bug fixes:
  • Other:
    • Documentation typo fixes and clarifications.

Version 2.1

Released 2011-06-10

  • New features:
    • json_loadb(): Decode a string with a given size, useful if the string is not null terminated.
    • Add JSON_ENCODE_ANY encoding flag to allow encoding any JSON value. By default, only arrays and objects can be encoded. (#19)
    • Add JSON_REJECT_DUPLICATES decoding flag to issue a decoding error if any JSON object in the input contins duplicate keys. (#3)
    • Add JSON_DISABLE_EOF_CHECK decoding flag to stop decoding after a valid JSON input. This allows other data after the JSON data.
  • Bug fixes:
    • Fix an additional memory leak when memory allocation fails in json_object_set() and friends.
    • Clear errno before calling strtod() for better portability. (#27)
  • Building:
    • Avoid set-but-not-used warning/error in a test. (#20)
  • Other:
    • Minor clarifications to documentation.

Version 2.0.1

Released 2011-03-31

  • Bug fixes:
    • Replace a few malloc() and free() calls with their counterparts that support custom memory management.
    • Fix object key hashing in json_unpack() strict checking mode.
    • Fix the parentheses in JANSSON_VERSION_HEX macro.
    • Fix json_object_size() return value.
    • Fix a few compilation issues.
  • Portability:
    • Enhance portability of va_copy().
    • Test framework portability enhancements.
  • Documentation:
    • Distribute doc/upgrading.rst with the source tarball.
    • Build documentation in strict mode in make distcheck.

Version 2.0

Released 2011-02-28

This release is backwards incompatible with the 1.x release series. See the chapter “Upgrading from older versions” in documentation for details.

  • Backwards incompatible changes:
    • Unify unsigned integer usage in the API: All occurences of unsigned int and unsigned long have been replaced with size_t.
    • Change JSON integer’s underlying type to the widest signed integer type available, i.e. long long if it’s supported, otherwise long. Add a typedef json_int_t that defines the type.
    • Change the maximum indentation depth to 31 spaces in encoder. This frees up bits from the flags parameter of encoding functions json_dumpf(), json_dumps() and json_dump_file().
    • For future needs, add a flags parameter to all decoding functions json_loadf(), json_loads() and json_load_file().
  • New features
  • Fix many portability issues, especially on Windows.
  • Configuration
    • Add file jansson_config.h that contains site specific configuration. It’s created automatically by the configure script, or can be created by hand if the configure script cannot be used. The file jansson_config.h.win32 can be used without modifications on Windows systems.
    • Add a section to documentation describing how to build Jansson on Windows.
    • Documentation now requires Sphinx 1.0 or newer.

Version 1.3

Released 2010-06-13

  • New functions:
  • New encoding flags:
    • JSON_PRESERVE_ORDER: Preserve the insertion order of object keys.
  • Bug fixes:
    • Fix an error that occured when an array or object was first encoded as empty, then populated with some data, and then re-encoded
    • Fix the situation like above, but when the first encoding resulted in an error
  • Documentation:
    • Clarify the documentation on reference stealing, providing an example usage pattern

Version 1.2.1

Released 2010-04-03

  • Bug fixes:
    • Fix reference counting on true, false and null
    • Estimate real number underflows in decoder with 0.0 instead of issuing an error
  • Portability:
    • Make int32_t available on all systems
    • Support compilers that don’t have the inline keyword
    • Require Autoconf 2.60 (for int32_t)
  • Tests:
    • Print test names correctly when VERBOSE=1
    • test/suites/api: Fail when a test fails
    • Enhance tests for iterators
    • Enhance tests for decoding texts that contain null bytes
  • Documentation:
    • Don’t remove changes.rst in make clean
    • Add a chapter on RFC conformance

Version 1.2

Released 2010-01-21

  • New functions:
  • New encoding flags:
    • JSON_SORT_KEYS: Sort objects by key
    • JSON_ENSURE_ASCII: Escape all non-ASCII Unicode characters
    • JSON_COMPACT: Use a compact representation with all unneeded whitespace stripped
  • Bug fixes:
    • Revise and unify whitespace usage in encoder: Add spaces between array and object items, never append newline to output.
    • Remove const qualifier from the json_t parameter in json_string_set(), json_integer_set() and json_real_set().
    • Use int32_t internally for representing Unicode code points (int is not enough on all platforms)
  • Other changes:
    • Convert CHANGES (this file) to reStructured text and add it to HTML documentation
    • The test system has been refactored. Python is no longer required to run the tests.
    • Documentation can now be built by invoking make html
    • Support for pkg-config

Version 1.1.3

Released 2009-12-18

  • Encode reals correctly, so that first encoding and then decoding a real always produces the same value
  • Don’t export private symbols in libjansson.so

Version 1.1.2

Released 2009-11-08

  • Fix a bug where an error message was not produced if the input file could not be opened in json_load_file()
  • Fix an assertion failure in decoder caused by a minus sign without a digit after it
  • Remove an unneeded include of stdint.h in jansson.h

Version 1.1.1

Released 2009-10-26

  • All documentation files were not distributed with v1.1; build documentation in make distcheck to prevent this in the future
  • Fix v1.1 release date in CHANGES

Version 1.1

Released 2009-10-20

  • API additions and improvements:
    • Extend array and object APIs
    • Add functions to modify integer, real and string values
    • Improve argument validation
    • Use unsigned int instead of uint32_t for encoding flags
  • Enhance documentation
    • Add getting started guide and tutorial
    • Fix some typos
    • General clarifications and cleanup
  • Check for integer and real overflows and underflows in decoder
  • Make singleton values thread-safe (true, false and null)
  • Enhance circular reference handling
  • Don’t define -std=c99 in AM_CFLAGS
  • Add C++ guards to jansson.h
  • Minor performance and portability improvements
  • Expand test coverage

Version 1.0.4

Released 2009-10-11

  • Relax Autoconf version requirement to 2.59
  • Make Jansson compile on platforms where plain char is unsigned
  • Fix API tests for object

Version 1.0.3

Released 2009-09-14

  • Check for integer and real overflows and underflows in decoder
  • Use the Python json module for tests, or simplejson if the json module is not found
  • Distribute changelog (this file)

Version 1.0.2

Released 2009-09-08

  • Handle EOF correctly in decoder

Version 1.0.1

Released 2009-09-04

Version 1.0

Released 2009-08-25

  • Initial release

Indices and Tables