Json Handling

JSON is a simple textual description of a data structure. An example of some JSON would be;

[ "apple", "orange", { "drink": "tonic water", "count" : 123 } ]

This example is a list that contains two strings followed by an "object". The term object refers to a construct akin to a "dictionary" or a "map". It is also possible for top-level objects to be non-collection types such as strings. The following is also valid JSON;


This page details how Haiku provides facilities for both parsing as well as writing data encoded as Json.

Parsing with Generic In-Memory Model

For some applications, parsing to an in-memory data structure is ideal. In such cases, the BJson class provides static methods for parsing a block of JSON data into a BMessage object. The application logic is then able to introspect the BMessage to obtain values.

BMessage Structure

The BMessage class has the ability to carry a collection of key-value pairs. In the case of a Json object type, the key-value pairs correlate to a JSON object or array. In the case of a JSON array type, the key-value pairs are the index of the elements in the JSON array represented as strings.

For example, the following JSON array...

[ "a", "b", "c" ]

...would be represented by the following BMessage ;

Key Value
"0" "a"
"1" "b"
"2" "c"

A Json object that, in its entirety, consists of a non-collection type such as a simple string or a boolean is not able to be represented by a BMessage ; at the top level there must be an array or an object for the parse to be successful.

Stream-based Parsing

Streaming is useful in many situations;

This architecture is sometimes known as an event-based parser or a "SAX" parser.

The BJson class provides a static method that accepts a stream of Json data in the form of a BDataIO. A BJsonEventListener sub-class is also supplied and as each Json token is read-in from the stream, it will be provided to the listener. The listener must implement three callback methods to handle the Json tokens;

Method Description
Handle(..) Provides JSON events to the listener
HandleError(..) Signals parse or processing errors to the listener
Complete(..) Informs the listener that parsing has completed

Events are embodied in instances of the BJsonEvent class and each of these has a type. Example types are;

In this way, the listener is able to interpret the incoming stream of data as Json and handle it in some way.

The following Json...

{"color": "red", "alpha": 0.6}

Would yield the following stream of events;

Event Type Event Data

Number Handling

The Json number literal format does not specify a numeric type such as int32 or double. To cope with the widest range of possibilities, the B_JSON_NUMBER event type captures the content as a string and then the BJsonEvent object is able to provide the original string for specific handling as well as convenient accessors for parsing to double or int64 types. This provides a high level of flexibility for the client.

Stacked Listeners

One implementation approach for a listener implement that might be used to read a data-transfer-object (DTO) is to create "sub-listeners" that mirror the structure of the Json data.

In the following example, a nested data structure is being parsed.


A primary-listener is employed called ColorGradientsListener. The primary-listener accepts Json parse events and will relay them to a sub-listener. The sub-listener is implemented to specifically deal with one tier of the inbound data. The sub-listeners are structured in a stack where the sub-listener at the head of the stack has a pointer to it's parent. The primary-listener maintains a pointer to the current head of the stack and will direct events to that sub-listener.

In response to events, the sub-listener can take-up the data, pop itself from the stack or push additional sub-listeners from the stack.

The same approach has been used in the following classes in a more generic manner;

The intention with this approach is that the structure of the event handling code in the sub-listeners mirrors that of the data-structure being parsed. Hopefully this makes creating the filling of a specific data-model easier even when very specific behaviours are required.

From a schema of the data structure it is probably also possible to create these sub-listeners and in this way automatically generate the C++ parse code as event listeners.


In order to render a data-structure as textual Json data, the opposite flow occurs; events are emitted by the client software into a class BJsonTextWriter. This class supports public methods such as WriteFalse() , WriteObjectStart() and WriteString(...) that control the outbound Json stream.

End to End

Because BJsonTextWriter is accepting JSON parse events, it is also a JsonEventListener and so can be used as a listener with the stream parsing; producing Json output from Json input. The output will however not include inbound whitespace because whitespace is not grammatically significant in Json.