The term "SAX" originated from Simple API for XML. We borrowed this term for JSON parsing and generation.
In RapidJSON, Reader (typedef of GenericReader<...>) is the SAX-style parser for JSON, and Writer (typedef of GenericWriter<...>) is the SAX-style generator for JSON.
Reader parses a JSON from a stream. While it reads characters from the stream, it analyze the characters according to the syntax of JSON, and publish events to a handler.
For example, here is a JSON.
While a Reader parses this JSON, it publishes the following events to the handler sequentially:
These events can be easily matched with the JSON, except some event parameters need further explanation. Let's see the simplereader example which produces exactly the same output as above:
Note that, RapidJSON uses template to statically bind the Reader type and the handler type, instead of using class with virtual functions. This paradigm can improve the performance by inlining functions.
As the previous example showed, user needs to implement a handler, which consumes the events (function calls) from Reader. The handler must contain the following member functions.
Null() is called when the Reader encounters a JSON null value.
Bool(bool) is called when the Reader encounters a JSON true or false value.
When the Reader encounters a JSON number, it chooses a suitable C++ type mapping. And then it calls one function out of Int(int), Uint(unsigned), Int64(int64_t), Uint64(uint64_t) and Double(double). If kParseNumbersAsStrings is enabled, Reader will always calls RawNumber() instead.
String(const char* str, SizeType length, bool copy) is called when the Reader encounters a string. The first parameter is pointer to the string. The second parameter is the length of the string (excluding the null terminator). Note that RapidJSON supports null character ‘’\0'inside a string. If such situation happens,strlen(str) < length. The lastcopyindicates whether the handler needs to make a copy of the string. For normal parsing,copy = true. Only when *insitu* parsing is used,copy = false`. And beware that, the character type depends on the target encoding, which will be explained later.
When the Reader encounters the beginning of an object, it calls StartObject(). An object in JSON is a set of name-value pairs. If the object contains members it first calls Key() for the name of member, and then calls functions depending on the type of the value. These calls of name-value pairs repeats until calling EndObject(SizeType memberCount). Note that the memberCount parameter is just an aid for the handler, user may not need this parameter.
Array is similar to object but simpler. At the beginning of an array, the Reader calls BeginArary(). If there is elements, it calls functions according to the types of element. Similarly, in the last call EndArray(SizeType elementCount), the parameter elementCount is just an aid for the handler.
Every handler functions returns a bool. Normally it should returns true. If the handler encounters an error, it can return false to notify event publisher to stop further processing.
For example, when we parse a JSON with Reader and the handler detected that the JSON does not conform to the required schema, then the handler can return false and let the Reader stop further parsing. And the Reader will be in error state with error code kParseErrorTermination.
As mentioned before, Reader is a typedef of a template class GenericReader:
The Reader uses UTF-8 as both source and target encoding. The source encoding means the encoding in the JSON stream. The target encoding means the encoding of the str parameter in String() calls. For example, to parse a UTF-8 stream and outputs UTF-16 string events, you can define a reader by:
Note that, the default character type of UTF16 is wchar_t. So this readerneeds to call String(const wchar_t*, SizeType, bool) of the handler.
The third template parameter Allocator is the allocator type for internal data structure (actually a stack).
The one and only one function of Reader is to parse JSON.
If an error occurs during parsing, it will return false. User can also calls bool HasParseEror(), ParseErrorCode GetParseErrorCode() and size_t GetErrorOffset() to obtain the error states. Actually Document uses these Reader functions to obtain parse errors. Please refer to DOM for details about parse error.
Reader converts (parses) JSON into events. Writer does exactly the opposite. It converts events into JSON.
Writer is very easy to use. If your application only need to converts some data into JSON, it may be a good choice to use Writer directly, instead of building a Document and then stringifying it with a Writer.
In simplewriter example, we do exactly the reverse of simplereader.
There are two String() and Key() overloads. One is the same as defined in handler concept with 3 parameters. It can handle string with null characters. Another one is the simpler version used in the above example.
Note that, the example code does not pass any parameters in EndArray() and EndObject(). An SizeType can be passed but it will be simply ignored by Writer.
You may doubt that, why not just using sprintf() or std::stringstream to build a JSON?
There are various reasons:
Writer must output a well-formed JSON. If there is incorrect event sequence (e.g. Int() just after StartObject()), it generates assertion fail in debug mode.Writer::String() can handle string escaping (e.g. converting code point U+000A to \n) and Unicode transcoding.Writer handles number output consistently.Writer implements the event handler concept. It can be used to handle events from Reader, Document or other event publisher.Writer can be optimized for different platforms.Anyway, using Writer API is even simpler than generating a JSON by ad hoc methods.
Writer has a minor design difference to Reader. Writer is a template class, not a typedef. There is no GenericWriter. The following is the declaration.
The OutputStream template parameter is the type of output stream. It cannot be deduced and must be specified by user.
The SourceEncoding template parameter specifies the encoding to be used in String(const Ch*, ...).
The TargetEncoding template parameter specifies the encoding in the output stream.
The Allocator is the type of allocator, which is used for allocating internal data structure (a stack).
The writeFlags are combination of the following bit-flags:
| Parse flags | Meaning |
|---|---|
kWriteNoFlags | No flag is set. |
kWriteDefaultFlags | Default write flags. It is equal to macro RAPIDJSON_WRITE_DEFAULT_FLAGS, which is defined as kWriteNoFlags. |
kWriteValidateEncodingFlag | Validate encoding of JSON strings. |
kWriteNanAndInfFlag | Allow writing of Infinity, -Infinity and NaN. |
Besides, the constructor of Writer has a levelDepth parameter. This parameter affects the initial memory allocated for storing information per hierarchy level.
While the output of Writer is the most condensed JSON without white-spaces, suitable for network transfer or storage, it is not easily readable by human.
Therefore, RapidJSON provides a PrettyWriter, which adds indentation and line feeds in the output.
The usage of PrettyWriter is exactly the same as Writer, expect that PrettyWriter provides a SetIndent(Ch indentChar, unsigned indentCharCount) function. The default is 4 spaces.
A Writer can only output a single JSON, which can be any JSON type at the root. Once the singular event for root (e.g. String()), or the last matching EndObject() or EndArray() event, is handled, the output JSON is well-formed and complete. User can detect this state by calling Writer::IsComplete().
When a JSON is complete, the Writer cannot accept any new events. Otherwise the output will be invalid (i.e. having more than one root). To reuse the Writer object, user can call Writer::Reset(OutputStream& os) to reset all internal states of the Writer with a new output stream.
Document's parsing capability is completely based on Reader. Actually Document is a handler which receives events from a reader to build a DOM during parsing.
User may uses Reader to build other data structures directly. This eliminates building of DOM, thus reducing memory and improving performance.
In the following messagereader example, ParseMessages() parses a JSON which should be an object with key-string pairs.
The first JSON (json1) was successfully parsed into MessageMap. Since MessageMap is a std::map, the printing order are sorted by the key. This order is different from the JSON's order.
In the second JSON (json2), foo's value is an empty object. As it is an object, MessageHandler::StartObject() will be called. However, at that moment state_ = kExpectValue, so that function returns false and cause the parsing process be terminated. The error code is kParseErrorTermination.
As mentioned earlier, Writer can handle the events published by Reader. condense example simply set a Writer as handler of a Reader, so it can remove all white-spaces in JSON. pretty example uses the same relationship, but replacing Writer by PrettyWriter. So pretty can be used to reformat a JSON with indentation and line feed.
Actually, we can add intermediate layer(s) to filter the contents of JSON via these SAX-style API. For example, capitalize example capitalize all strings in a JSON.
Note that, it is incorrect to simply capitalize the JSON as a string. For example:
Simply capitalizing the whole JSON would contain incorrect escape character:
The correct result by capitalize:
More complicated filters can be developed. However, since SAX-style API can only provide information about a single event at a time, user may need to book-keeping the contextual information (e.g. the path from root value, storage of other related values). Some processing may be easier to be implemented in DOM than SAX.