Boost C++ Libraries of the most highly regarded and expertly designed C++ library projects in the world. Herb Sutter and Andrei Alexandrescu, C++ Coding Standards


Quick Look

Here we highlight important features through example code to help convey the style of the interface. We begin by including the library header file which brings all the symbols into scope. Alternatively, individual headers may be included to obtain the declarations for specific types:

#include <boost/json.hpp>

In order to link your program you will need to link with a built library. Alternatively you can use the header-only configuration simply by including this header file in any one of your new or existing source files:

#include <boost/json/src.hpp>
[Note] Note

Sample code and identifiers used throughout are written as if the following declarations are in effect:

#include <boost/json.hpp>
using namespace boost::json;

Say you want to recreate this JSON object in a container:

  "pi": 3.141,
  "happy": true,
  "name": "Boost",
  "nothing": null,
  "answer": {
    "everything": 42
  "list": [1, 0, 2],
  "object": {
    "currency": "USD",
    "value": 42.99

In this library the types array, object, and string hold JSON arrays, objects, and strings respectively while the type value is a special variant which can hold any JSON element. Here we construct an empty object and then insert the elements above:

object obj;                                                     // construct an empty object
obj[ "pi" ] = 3.141;                                            // insert a double
obj[ "happy" ] = true;                                          // insert a bool
obj[ "name" ] = "Boost";                                        // insert a string
obj[ "nothing" ] = nullptr;                                     // insert a null
obj[ "answer" ].emplace_object()["everything"] = 42;            // insert an object with 1 element
obj[ "list" ] = { 1, 0, 2 };                                    // insert an array with 3 elements
obj[ "object" ] = { {"currency", "USD"}, {"value", 42.99} };    // insert an object with 2 elements

While keys are strings, the mapped type of objects and the element type of arrays is a special type called value which can hold any JSON element, as seen in the previous assignments. Instead of building the JSON document using a series of function calls, we can build it in one statement using an initializer list:

value jv = {
    { "pi", 3.141 },
    { "happy", true },
    { "name", "Boost" },
    { "nothing", nullptr },
    { "answer", {
        { "everything", 42 } } },
    {"list", {1, 0, 2}},
    {"object", {
        { "currency", "USD" },
        { "value", 42.99 }
            } }

When a value, array, or object is assigned or constructed from an initializer list, the creation of the new value happens only once. This makes initializer lists equally efficient as using the other ways to create a value. The types in this library are first-class, supporting copy and move construction and assignment:

array arr;                                          // construct an empty array
arr = { 1, 2, 3 };                                  // replace the contents with 3 elements
value jv1( arr );                                   // this makes a copy of the array
value jv2( std::move(arr) );                        // this performs a move-construction

assert( arr.empty() );                              // moved-from arrays become empty
arr = { nullptr, true, "boost" };                   // fill in the array again

To permit custom memory allocation strategies, these containers all allow construction with a storage_ptr which is a smart pointer to a memory_resource. The constructor signatures have the same ordering as their std equivalents which use Allocator parameters. Once a container is constructed its memory resource can never change. Here we create an array without performing any dynamic allocations:

    unsigned char buf[ 4096 ];                      // storage for our array
    static_resource mr( buf );                      // memory resource which uses buf
    array arr( &mr );                               // construct using the memory resource
    arr = { 1, 2, 3 };                              // all allocated memory comes from `buf`

The containers in this library enforce the invariant that every element of the container will use the same memory resource:

    monotonic_resource mr;                          // memory resource optimized for insertion
    array arr( &mr );                               // construct using the memory resource
    arr.resize( 1 );                                // make space for one element
    arr[ 0 ] = { 1, 2, 3 };                         // assign an array to element 0
    assert( *arr[0].storage() == * );  // same memory resource

When a library type is used as the element type of a pmr container; that is, a container which uses a polymorphic_allocator, the memory resource will automaticaly propagate to the type and all of its children:

    monotonic_resource mr;
    boost::container::pmr::vector< value > vv( &mr );
    vv.resize( 3 );

    // The memory resource of the container is propagated to each element
    assert( *vv.get_allocator().resource() == *vv[0].storage() );
    assert( *vv.get_allocator().resource() == *vv[1].storage() );
    assert( *vv.get_allocator().resource() == *vv[2].storage() );

Up until now we have shown how values may be constructed from a memory resource pointer, where ownership is not transferred. In this case the caller is responsible for ensuring that the lifetime of the resource is extended for the life of the container. Sometimes you want the container to acquire shared ownership of the resource. This is accomplished with make_shared_resource:

value f()
    // create a reference-counted memory resource
    storage_ptr sp = make_shared_resource< monotonic_resource >();

    // construct with shared ownership of the resource
    value jv( sp );

    // assign an array with 3 elements, the monotonic resource will be used
    jv = { 1, 2, 3 };

    // The caller receives the value, which still owns the resource
    return jv;

A counted memory resource will not be destroyed until every container with shared ownership of the resource is destroyed.


JSON can be parsed into the value container in one step using a free function. In the following snippet, a parse error is indicated by a thrown exception:

value jv = parse( "[1, 2, 3]" );

Error codes are also possible:

boost::system::error_code ec;
value jv = parse( R"( "Hello, world!" )", ec );

By default, the parser is strict and only accepts JSON compliant with the standard. However this behavior can be relaxed by filling out an options structure enabling one or more extensions. Here we use a static buffer and enable two non-standard extensions:

unsigned char buf[ 4096 ];
static_resource mr( buf );
parse_options opt;
opt.allow_comments = true;
opt.allow_trailing_commas = true;
value jv = parse( "[1, 2, 3, ] // array ", &mr, opt );

The parser in this library implements a streaming algorithm; it can process JSON piece-by-piece, without the requirement that the entire input is available from the start. The parser uses a temporary memory allocation to do its work. If you plan on parsing multiple JSONs, for example in a network server, keeping the same parser instance will allow re-use of this temporary storage, improving performance.

stream_parser p;
boost::system::error_code ec;
p.write( "[1, 2 ", ec );
if( ! ec )
    p.write( ", 3]", ec );
if( ! ec )
    p.finish( ec );
if( ec )
value jv = p.release();

With strategic use of the right memory resources, parser instance, and calculated upper limits on buffer sizes, it is possible to parse and examine JSON without any dynamic memory allocations. This is explored in more detail in a later section.


Simple free functions are provided for serializing a value to a std::string containing JSON:

value jv = { 1, 2, 3 };
std::string s = serialize( jv );                // produces "[1,2,3]"

The serializer in this library implements a streaming algorithm; it can output JSON a piece at a time, without the requirement that the entire output area is allocated at once:

serializer sr;
sr.reset( &jv );                                // prepare to output `jv`
    char buf[ 16 ];
    std::cout << buf );
while( ! sr.done() );
Value Conversion

Given a user-defined type:

namespace my_app {

struct customer
    int id;
    std::string name;
    bool current;

} // namespace my_app

We can define a conversion from the user-defined type to a value by defining an overload of tag_invoke in the same namespace. This maps customer to a JSON object:

namespace my_app {

void tag_invoke( value_from_tag, value& jv, customer const& c )
    jv = {
        { "id" , },
        { "name", },
        { "current", c.current }

} // namespace my_app

This allows us to use the library function value_from to produce a value from our type:

my_app::customer c{ 1001, "Boost", true };
std::cout << serialize( value_from( c ) );

The library knows what to do with standard containers. Here we convert an array of customers to a value:

std::vector< my_app::customer > vc;
value jv = value_from( vc );

To go from JSON to a user-defined type we use value_to, which uses another overload of tag_invoke. This converts a JSON value to a customer. It throws an exception if the contents of the value do not match what is expected:

namespace my_app {

// This helper function deduces the type and assigns the value with the matching key
template<class T>
void extract( object const& obj, T& t, string_view key )
    t = value_to<T>( key ) );

customer tag_invoke( value_to_tag< customer >, value const& jv )
    customer c;
    object const& obj = jv.as_object();
    extract( obj,, "id" );
    extract( obj,, "name" );
    extract( obj, c.current, "current" );
    return c;

} // namespace my_app

The code above defines the convenience function extract, which deduces the types of the struct members. This works, but requires that the struct is DefaultConstructible. An alternative is to construct the object directly, which is a little more verbose but doesn't require default construction:

namespace my_app {

customer tag_invoke( value_to_tag< customer >, value const& jv )
    object const& obj = jv.as_object();
    return customer {
        value_to<int>( "id" ) ),
        value_to<std::string>( "name" ) ),
        value_to<bool>( "current" ) )

} // namespace my_app

Now we can construct customers from JSON:

json::value jv;
customer c( value_to<customer>(jv) );

The library's generic algorithms recognize when you are converting a value to a container which resembles an array or object, so if you wanted to turn a JSON array into a vector of customers you might write:

std::vector< customer > vc = value_to< std::vector< customer > >( jv );