wxJSON - A brief tutorial

This page is a simple tutorial that describes how to use the wxJSON library in order to store data in JSON format, reading them, changing the values and how to generate JSON text that can be saved to a file, a stream or sent over a network connection.

Table of contents

Introduction

JSON is a text format and it is completely platform-independent. You do not need to worry about computer's architecture, endianness or operating system. It is very easy for humans to read and write and easy for machines to parse and generate. It is also light-weight and much more compact than any other text-based data-interchange formats.

We will learn how to use the wxJSON library by examining some simple examples. JSON is a data format; it is not suitable for describing complex documents or to store images, audio and video streams. For these purposes, there are many other formats which are by far more appropriate than JSON.

If you are new to JSON, it is better that you first read the following page, which describe the JSON syntax, its advantages and its disadvantages:

Example 1: a simple example

Let's look an example on how to read and write JSON data using the wxWidgets implementation of the JSON value class. Suppose we have the following text stored in a wxString object (but you can also read it from a stream class):

 /***************************
  This is a C-style comment
 ***************************/
 {
   // this is a comment line in C++ style
   "wxWidgets" :
   {
     "Description" : "Cross-platform GUI framework",
     "License" : "wxWidgets",
     "Version" :
     {
       "Major" : 2,
       "Minor" : 8,
       "Stable" : true
     },
     "Languages" :
     [
       "C++",
       "Phyton",
       "Perl",
       "C#/Net"
     ]
   }
 }

We can retrieve the values using several access methods as explained in the following code fragment:

  // the JSON text, stored in a wxString object
  wxString document( _T( "/************\n  This is a ...... "));

  // construct the JSON root object
  wxJSONValue  root;

  // construct a JSON parser
  wxJSONReader reader;

  // now read the JSON text and store it in the 'root' structure
  // check for errors before retreiving values...
  int numErrors = reader.Parse( document, &root );
  if ( numErrors > 0 )  {
    cout << "ERROR: the JSON document is not well-formed" << endl;
    const wxArrayString& errors = reader.GetErrors();
    // now print the errors array
    ...
    return;
  }

  // get the 'License' value
  wxString license  = root["wxWidgets"]["License"].AsString();

  // check if a 'Version' value is present
  bool hasMember = root["wxWidgets"].HasMember( "Version" );

  // get the major version value as an integer
  int majorVer = root["wxWidgets"]["Version"]["Major"].AsInt();

  // get the minor version; if the value does not exists, the
  // default value of ZERO is returned
  wxJSONValue defaultValue( 0 );
  int minorVer = root["wxWidgets"]["Version"].Get( "Minor", defaultValue).AsInt();

  // the same as above, but directly constructing the default value
  int minorVer = root["wxWidgets"]["Version"].Get( "Minor", wxJSONValue( 0)).AsInt();

  // now retrive the array of supported languages
  wxJSONValue languages = root["wxWidgets"]["Languages"];

  // before obtaining the array of strings, we check that the type
  // of the 'language' object is an array
  // NOTE: this is not strictly necessary.
  bool isArray = languages.IsArray();

  wxArrayString supportedLanguages;
  for ( int i = 0; i < languages.Size() i++ ) {
    supportedLanguages.Add( languages[i].AsString());
  }

  // finally, we get an array of all member's names of the 'wxWidgets'
  // item. The string array will contain (maybe not in this order):
  // 
  //   Description
  //   License
  //   Version
  //   Languages
  //
  wxArrayString memberNames = root["wxWidgets"].GetMemberNames();
  
  // starting from version 1.1 you can also get the value and check
  // if it is of the expected type in only one call:
  int i;
  if ( !root["wxWidgets"]["Version"]["Major"].AsInt( i ))  {
     cout << "Error: major version is not of type INT";
  }

You can also use iterators to get all key/value objects in a JSON value, provided that the value (that may also be the root value) is of type key/value. This is an example:

{
    wxJSONValue root;
    root[_T("first")] = 1;
    root[_T("second")] = 2;
    root[_T("third")] = 3;
    root[_T("fourth")] = 4;
    
    // the following works in both enabled and disabled STL
    
    const wxJSONInternalMap* map = root.AsMap();
    wxASSERT( map );    // check that the type is correct
    
    wxJSONInternalMap::const_iterator iter;
    for ( iter = map->begin(); iter != map->end(); iter++ ) {
        const wxJSONInternalMap::value_type& v = *iter;
        // 'v.first' is the key
        // 'v.second' is the wxJSONValue object
    }
}

The wxJSONReader class's constructor has some parameters that control how much error-tolerant should the parser be. By default, the parser is very tolerant about non fatal errors which are reported as warnings. For more information see the wxJSONReader class's description. In addition to reading from a wxString object, you can also read the JSON text from a wxInputStream object. The difference is that the text read from a stream must be encoded in UTF-8 format while the text stored in a string object is encoded in different ways depending on the platform and the build mode: in Unicode builds, strings are stored in UCS-2 format on Windows and UCS-4 on GNU/Linux; in ANSI builds, the string object contains one-byte, locale dependent characters. To know more about Unicode / ANSI read Unicode support in wxJSON

Adding new values or changing the value of existing JSON-value objects is also very simple. The following code fragment shows some examples:

  // upgrade the minor version
  root["wxWidgets"]["Version"]["Minor"] = 9;

  // create the new 'URL' item
  root["wxWidgets"]["URL"] = "http://www.wxwidgets.org";

  // append a new supported language in the 'Language' array
  root["wxWidgets"]["Languages"].Append( "Java" );

  // creates the new 'Authors' array.
  // creating an array is just as simple as using the 'Append()'
  // member function.
  root["wxWidgets"]["Authors"].Append( "J. Smart" );
  root["wxWidgets"]["Authors"].Append( "V. Zeitling" );
  root["wxWidgets"]["Authors"].Append( "R. Roebling" );
  ... and many others...

  // you can also use subscripts to obtain the same result:
  root["wxWidgets"]["Authors"][0] = "J. Smart";
  root["wxWidgets"]["Authors"][1] = "V. Zeitling";
  root["wxWidgets"]["Authors"][2] = "R. Roebling";
  ... and many others...

  // after the changes, we have to write the JSON object back
  // to its text representation
  wxJSONWriter writer;
  wxString     str;
  writer.Write( root, str );

  // if you use the default writer constructor the JSON text
  // output is human-readable (indented) but does not contain
  // the comment lines
  // if you want to keep the comment lines you have to pass
  // some parameters to the wxJSONWriter constructor
  wxJSONWriter writer2( wxJSONWRITER_STYLED | wxJSONWRITER_WRITE_COMMENTS );
  wxString     str2;
  writer2.Write( root, str2 );

The writer class's constructor has some parameters that allow you to control the style of the output. By default, the writer produces human-readable output with a three-space indentation for objects / arrays sub-items (as shown in the example text above) but it does not write comment lines. You can suppress indentation if, for example, the JSON text has to be sent over a network connection.

Also note that in order to actually have comment lines written back to the JSON text document, you also have to store comments when reading the JSON text document. By default, the parser is error-tolerant and recognizes C/C++ comments but it ignores them. This means that you cannot rewrite them back regardless the flags used in the writer class. To know more about comment lines in JSONvalue objects, see Using comment lines in wxJSON.

In addition to writing to a wxString object, you can also write to a wxOutputStream object. The difference is that the text written to streams is always encoded in UTF-8 format in both Unicode and ANSI builds while the text written to a string object is encoded in different ways depending on the platform and the build mode: in Unicode builds, strings are stored in UCS-2 format on Windows and UCS-4 on GNU/Linux; in ANSI builds, the string object contains one-byte, locale dependent characters. To know more about Unicode / ANSI read Unicode support in wxJSON

Also note that the wxJSONWriter::Write() function does not return a status code. This is OK for writing to a string object but when writing to streams, you have to check for errors. Because the wxJSON writer does not return error codes, you have to check for errors using the stream's memberfunctions, as in the following example code:

   // construct the JSON value object and add values to it
   wxJSONValue root;
   root["key1"] = "some value";

   // write to a stream
   wxMemoryOutputStream mem;
   wxJSONWriter writer;
   writer.Write( root, mem );

   // use the stream's 'GetLastError()' function to know if the
   // write operation was successfull or not
   wxStreamError err = mem.GetLastError();
   if ( err != wxSTREAM_NO_ERROR )  {
     MessageBox( _T("ERROR: cannot write the JSON text output"));
   }

The power and simplicity of JSON

I do not know much about XML but I think that JSON is really a valid alternative to it if you just need a simple format for data interchange. JSON is not suitable for describing complex documents: it can only handle data and it is specialized in handling progamming language's variables.

I only would like to let you know how much simple is wxJSON: the subscript operators used to access JSON values returns a reference to the JSON value itself thus allowing to have multiple subscripts. Moreover, if the accessed value does not exists, it will be created and a reference to the newly created value is returned. This feature lets you use constructs such as the following:

  wxJSONValue value;
  value["key-1"]["key-2"]["key-3"][4] = 12;
  
  // now write to JSON text
  wxJSONWriter writer;
  wxString     jsonText;
  writer.Write( root, jsonText );

Because value does not contain any of the specified keys (for objects) and elements (for the array), they will be created. The JSON text output of the code fragment seen above is as follows:

  {
    "key-1" :  {
       "key-2" :  {
          "key-3" : [
             null,
             null,
             null,
             null,
             12
          ]
        }
     }
  }

Example 2: a configuration file

We start by using JSON for an application's configuration file. There are many formats for storing application's configuration data. I remember when there was MS-DOS: each application used its own, unreadable and proprietary format (it was a nightmare). Next came Windows 3: it had a better way for storing application's configuration data; they were kept in an .INI file which contains simple ASCII text. This was an improvement because it was easier for humans to fine-tuning application's behaviour.

In this example we use JSON to store the configuration data of a simple web server application. If you take a look at the Apache config file you will notice that our example looks very similar (but much more human readable).

Our server is a neverending application and it is not interactive: it reads its configuration at startup and when a signal is sent to it. Using JSON for the configuration data is a good choice because it is easy for humans to write the JSON text document. Below we find our webserver's configuration file:

 {
   // global configuration
   "Global" :  {
     "DocumentRoot"  : "/var/www/html",
     "MaxClients"    : 250,
     "ServerPort"    : 80,
     "ServerAddress" : 0.0.0.00
     "MaxRequestsPerClient"  : 1000
   }

   // an array of objects that describes the modules that has to
   // be loaded at startup
   "Modules" : [
      {
        "Name"    : "auth_basic",
        "File"    : "modules/mod_auth_basic.so",
        "OnStart" : true
      },
      {
        "Name"    : "auth_digest",
        "File"    : "modules/mod_auth_digest.so",
        "OnStart" : true
      },
      {
        "Name"    : "auth_file",
        "File"    : "modules/mod_auth_file.so",
        "OnStart" : false
      },
   ]

   // Main server configuration
   "Server" :       {
      "Admin" : "root@localhost.localdomain"
      "Name"  : "www.example.com"
   },

   // The description of directories and their access permissions
   "Diretory"  : [
      {
         "Path"       : "/var/www/html",
         "AllowFrom"  : "ALL",
         "DenyFrom"   : null,
         "Options" :     {
            "Multiviews"    : false,
            "Indexes"       : true,
            "FollowSymLink" : false
         }
      }
   ]
 }

I think that the file is self-explanatory. I do not want to write the code of the whole web-server application: I only want to show you how to read the configuration data.

When the application starts, it calls a function that reads the configuration file and returns ZERO if there was no error or an exit status code if the file is not correct. The function may be similar to the following one:

  int ReadConfig( wxInputStream& jsonStream )
  {
    // comment lines are recognized by the wxJSON library and
    // can also be stored in the JSON value objects they refer to
    // but this is not needed by our application because the
    // config file is written by hand by the website admin
    // so we use the default ctor of the parser which recognizes
    // comment lines but do not store them
    wxJSONReader reader;
    wxJSONvalue  root;
    int numErrors = reader.Parse( jsonStream, root );
    if ( numErrors > 0 )  {
      // if there are errors in the JSON document, print the
      // errors and return a non-ZERO value
      const wxArrayString& errors = reader.GetErrors();
      for ( int i = 0; i < numErrors; i++ )  {
        cout << errors[i] << endl;
      }
      return 1;
    }

    // if the config file is syntactically correct, we retrieve
    // the values and store them in application's variables
    gs_docRoot = root["Global"]["DocumentRoot"].AsString();

    // we use the Get() memberfunction to get the port on which
    // the server listens. If the parameter does not exist in the
    // JSON value, the default port 80 is returned
    wxJSONvalue defaultPort = 80;
    gs_serverPort = root["Global"].Get( "ServerPort", defaultPort ).AsInt();

    // the array of modules is processed in a different way: for
    // every module we print its name and the 'OnStart' flag.
    // if the flag is TRUE, we load it.
    wxJSONValue modules = root["Modules"];

    // check that the 'Modules' value is of type ARRAY
    if ( !modules.IsArray() ) {
      cout << "ERROR: \'modules\' must be a JSON array" << endl;
      return 1;
    }

    for ( int i = 0; i < modules.Size(); i++ )  {
      cout << "Processing module: " << modules[i]["Name"].AsString() << endl;
      bool load =  modules[i]["OnStart"].AsBool();
      cout << "Load module? " << ( load ? "YES" : "NO" ) << endl;
      if ( load )  {
        LoadModule( modules[i]["File"].Asstring());
      }
    }
    // return a ZERO value: it means success.
    return 0;
  }

Example 3: Describing a table

How many times did you use a table in your application? I know the answer: many times. So the best thing would be to write a general-purpose panel window that is capable to show every possible table and table's format.

It is not hard to write a similar panel: we only need to implement some basic data that will be passed as parameters to the panel window:

A more complex work is to define a good data structure that holds this informations and the table's data themselves. Because we have to show many different tables in our application, there is not a general structure suitable for our needs because the data type of each column may vary from table to table. But we need something to pass as a parameter to our general-purpose table-viewer panel.

The answer could be: use a JSON formatted wxString object. We define a JSON object that contains two main objects:

Below you find the format of the JSON text that describes a table containing three columns and three rows:

 {
   "Border"  : 1,

   "Columns" : [ 
     {
       "Name"   : "City",
       "Width"  : 50,
       "Unit"   : "Percentage"
     },
     {
       "Name"      : "Temperature",
       "Width"     : 20,
       "Unit"      : "Percentage"
     },
     {
       "Name"      : "Date",
       "Width"     : 30,
       "Unit"      : "Percentage"
       "Alignment" : "center"
     }
  ]

   "Rows" : [
     [ "Baltimora", 20, "20 july" ],
     [ "New York", 25, "23 july" ],
     [ "Los Angeles", 29, "25 july" ]
   ]
 }

Note that there is no need to specify the type of the data contained in each column because the JSON value object already carries it.

The code for displaying a table that is described in the above JSON text is similar to this one:

 void DisplayTable( const wxString& jsonText )
 {
   wxJSONReader reader;
   wxJSONvalue  root;
   int numErrors = reader.Parse( jsonText, root );
   if ( numErrors > 0 )  {
     // if there are errors in the JSON document return
     return;
   {

   // now display the column names
   wxJSONvalue columns = root["Columns"];
   int border = root["Border"].AsInt();
   int width; string align;
   for ( int i = 0; i < columns.Size(); i++ )  {
     width = columns[i]["Width"].AsInt();
     DisplayColName( columns[i]["Name"].AsString(), width );
   }

   // and now we display the data in the rows
   // note that we use a predefined alignment for data
   // unless a specific alignment is set:
   //
   //  left for strings
   //  right for numbers

   // the bidimensional array
   wxJSONValue rows = root["Rows"];

   // the string that has to be displayed in the table's cell
   string valueString;

   // the default alignment: it is set depending on the data type
   wxJSONValue defaultAlign;

   // for all rows ...
   for ( int x = 0; x < rows.Size(); x++ )  {

     // .. and for all columns in the row
     for ( int y = 0; y < rows[x].Size(); y++ )  {
       // get the width of the column
       width = columns[y]["Width"].AsInt();

       // get the value object
       wxJSONValue value = rows[x][y];

       // check the type of the data
       wxJSONValue::wxJSONType type = value.GetType();
       switch ( type )  {
         case wxJSONTYPE_NULL :
           // display an empty string
           valueString.clear();;
           break;
         case wxJSONTYPE_INT :
         case wxJSONTYPE_UINT :
         case wxJSONTYPE_DOUBLE :
           // numeric values are right-aligned
           defaulAlign = "right";
           align = columns[y].Get( "Align", defaultAlign ).AsString();
           valueString = value.AsString();
           break;
         case wxJSONTYPE_STRING :
         case wxJSONTYPE_CSTRING :
           defaulAlign = "left";
           align = columns[y].Get( "Align", defaultAlign ).AsString();
           valueString = value.AsString();
           break;
         case wxJSONTYPE_BOOL :
           defaulAlign = "center";
           align = columns[y].Get( "Align", defaultAlign ).AsString();
           valueString = value.AsString();
           break;
       }
       // now that we have the alignment, the column's width and the 
       // value of the data as a string:
       // note that numeric data are converted to a decimal string
       // and boolean values are converted to 'TRUE' or 'FALSE' when you
       // use the wxJSONValue::AsString() memberfunction
       DisplayValue( valueString, width, align );

     }   // continue for all columns
   }     // continue for all rows
 }

JSON format is very flexible: in future we can add new features to the application. For example we may decide that our general-purpose table-viewer panel will let the user to change the values in the table rows but only for some specific columns.

We add a new item in the Columns array descriptors: the Editable flag which is a boolean type. Example:

   "Columns" : [ 
     {
       "Name"   : "Temperature",
       "Width"  : 50,
       "Unit"   : "Percentage",
       "Editable" : true
     },

Note that this new format of our table description is compatible in both directions: it is normal that a new version of the application can read and handle old-style data but it is not very easy to maintain the old application compatible with a new data format that was introduced in a new version.

In our example, the simplicity and flexibility of JSON make the old application capable of reading the new format of JSON data. Of course, the data are not editable because the old application does not permit this operation. The old version of the application simply ignores the existance of the new Editable flag so that the JSON text can be read and processed as in the previous version.

Obtaining values from JSON value class

The wxJSONValue class defines functions for checking what type of data is stored in the JSON object and for getting the value stored in the class. This topic has totally changed from versions 0.x to versions 1.x.

In older 0.x versions you can get a value stored in the JSON value object in a type that is different from the one that is actually stored provided that the two types are compatible. For example, if a value contains an integer type of value -1, you can get the value as an integer or as a double, or as a string:

  wxJSONValue v( -1 );

  int i      = v.AsInt();    // returns -1
  double d   = v.AsDouble(); // returns -1.0
  wxString s = v.AsString(); // returns "-1"

This is no longer supported in new 1.x versions: all AsXxxxxx() functions return the stored value without reinterpreting the bits stored in memory so when you call the AsDouble() function on a value object that contains the integer -1 (all bits set), you get a NaN because all bits set in a double type represent a NaN:

  wxJSONValue v( -1 );

  int i      = v.AsInt();    // returns -1
  double d   = v.AsDouble(); // returns NaN
  wxString s = v.AsString(); // returns "-1"

The only exceptions to this rule are the functions:

The first function returns a string representation of all types included arrays and objects. The second function returns:

The new AsXxxxxx(T&) function.

In order to get the correct value from a JSON value object you should always call the IsXxxxxx() function before getting the value and, if it is of the wrong type you should notify the user:

  wxJSONValue v( _T("100"));
  int i = 10;           // the default value
  if ( v.IsInt() )  {
    i = v.AsInt();
  }
  else  {
    wxMessageBox( _T("Warning: parameter is of the wrong type - using default")) ;
  }

Starting from version 1.1 you can get the value of a wxJSONValue object and check if it is of the expected type in only one call using an overloaded version of the AsXxxxxxx functions which takes the expected type as a parameter and returns TRUE if the value actually stored in the wxJSONValue object is of that type (see wxJSONValue::AsInt(int&) for details). For example:

  wxJSONValue v( _T("100"));
  int i = 10;           // the default value
  
  if ( !value.AsInt( i ) ) {
    wxMessageBox( _T("Warning: parameter is of the wrong type - using default")) ;
  }

You can also use the wxJSONValue::GetType() function to get directly the type of the value which returns a wxJSONType type that can be used in a switch statement:

  wxJSONValue value( 100 );
  wxJSONType type = value.GetType();
  switch ( type )  {
    // use 64-bit int for all signed integers
    case wxJSONTYPE_INT :
    case wxJSONTYPE_SHORT :
    case wxJSONTYPE_LONG :
    case wxJSONTYPE_INT64 :
      wxInt64 i64 = value.AsInt64();
      break;
      
    // use 64-bit uint for all unsigned integers
    case wxJSONTYPE_UINT :
    case wxJSONTYPE_USHORT :
    case wxJSONTYPE_ULONG :
    case wxJSONTYPE_UINT64 :
      wxUint64 i64 = value.AsUInt64();

    case wxJSONTYPE_DOUBLE :
      double d = value.AsDouble();

    ... always specify ALL the possible types ...

    default :
      wxFAIL_MSG( _T("Unexpected value type"));
  }

What happens if I get the wrong type?

As explained earlier, the wxJSONValue::AsXxxxxx() functions just return the content of the memory area as it apears without trying to promote the stored type to another type even when they are compatible. The conseguence is that the function returns a wrong value. Example:

  wxJSONValue v( -1 );

  // returns -1 (OK): the IsInt() returns TRUE
  int i      = v.AsInt();

  // returns NaN (wrong): IsDouble() returns FALSE
  double d   = v.AsDouble();

  // returns 4294967295 (wrong) IsUInt() returns FALSE
  unsigned u = v.AsUInt();

  // returns 65535 (wrong) IsUShort() returns FALSE
  unsigned short h  = v.AsUShort();

  // returns "-1" (OK) IsString() returns FALSE; this is an exception
  wxString s  = v.AsString();

  // returns a NULL pointer IsCString() returns FALSE
  wxChar* c  = v.AsCString();

The above return values are just examples which demonstrate the wrong results. When you store something in the m_value data member of the JSON value object, the actual return value is undefined if incorrectly accessed. For example, if a positive integer value of 100,000 (0x186A0) is stored in a JSON value object and we try to access it as a SHORT we get:

However, in debug builds all AsXxxxxx() functions ASSERT that the corresponding IsXxxxx functions return TRUE so avoiding type mismatch errors. The ASSERTION failures does not occur in release builds.

So now I cannot get the value as a compatible type? Oh no, you can get it but you have to access it using the correct type. Example:

  wxJSONValue v1( 10 );    // a SHORT
  wxJSONValue v2( 100.00); // a DOUBLE
  
  double d = v1.AsInt();   // OK, a SHORT accessed as INT
  int    i = v2.AsInt();   // WRONG! a double accessed as INT
  int  i2 = v2.AsDouble(); // OK, a double accessed as DOUBLE, but the compiler warns you
  
  int  i3 = (int) v2.AsDouble();  // OK, no warning, if you know what you are doing

Starting from version 0.5 the wxJSON library supports 64-bits integers on platforms that have native support for this. If the library is compiled with the 64-bits support enabled, the JSON value class defines functions in order to let the user know the storage needed by the integer value (16-, 32-bits or 64-bits). To know more about this topic read 64-bits and 32-bits integers.

How numbers are read from JSON text.

If 64-bits support is enabled in the wxJSON library, reading values from JSON text may have different results. If a numeric value is too large for 32-bits integer storage:

When the wxJSON parser reads a token that start with a digit or with a minus or plus sign, it assumes that the value read is a number and it tries to convert it to:

Note that if 64-bits storage is enabled, the wxJSON parser, does not try to convert numeric tokens in a 32-bits integer but immediatly tries 64-bits storage. This has the conseguence that numeric values that are between LONG_MAX + 1 and ULONG_MAX are stored as signed integers in a 64-bits environment and as unsigned integers in a 32-bits environment.

Examples:

  {
    // in a 32 bits integer environment
    2147483647  // INT_MAX: read as a signed long integer
    2147483648  // INT_MAX + 1: read as a unsigned long

   -2147483648  // INT_MIN: read as a signed long integer
    4294967295  // UINT_MAX: read as a unsigned long integer
    4294967296  // UINT_MAX + 1: read as a double

    // in a 64 bits integer environment
    2147483648  // INT_MAX + 1: read as a signed wxInt64
    4294967295  // UINT_MAX: read as a signed wxInt64
    4294967296  // UINT_MAX + 1: read as a signed wxInt64
  }

Also note that if a number is between INT64_MIN and INT64_MAX (or LONG_MIN and LONG_MAX, depending on the platform) it is always read as a signed integer regardless its original type. You can use a special writer's flag in order to force the wxJSON library to recognize unsigned JSON values written to JSON text. See The wxJSONWRITER_RECOGNIZE_UNSIGNED flag for more info.

Writing doubles to JSON text output

There is an issue when you have to write double values to a decimal representation. You know that some fractional values such as, for example, 1/10 cannot be exactly represented in binary so when you write such values to a text representation there are rounding and/or truncating problems. The output you have depends on the format string used. This is an example using variuos format strings for printing the value of 1/10.
  Format string      Output
  -------------      ------
  %f                  0.100000
  %g                  0.1
  %.0g                0.1
  %.g:                0.1
  %.20g               0.10000000000000000555
  %.10g               0.1

The wxJSON writer class has a function that allows you to set the format string for prinitng double values. The function is called wxJSONWriter::SetDoubleFmtString and sets the format string for the JSON writer object it is called for. If not specified, the default format string for the writer is:

  %.10g
which specifies a precision of 10 decimal digits and suppress trailing ZEROes.

Using comment lines in wxJSON

Comments are not supported by the JSON syntax specifications but many JSON implementations do recognize and store comment lines in the JSON value objects. Starting by version 0.2, the wxJSON library do recognize and store C/C++ comment lines in the JSON input text and can also write comments to the JSON output text.

Why should we use comments in JSON formatted text?

There are several reasons: in an application's configuration file like the one we have seen in Example 2: a configuration file comments are very usefull to help the user to understand the meaning of each configuration option.

On the other hand, if a data structure is sent over a network connection, it is most likely that comments are not really needed but they may still be usefull for debugging purposes or for explaining the value, as in the following example:

 {
   "Person" :  {
   {
     "Name"    : "John Smith",
     "Height"  : 190,   // expressed in centimeters
     "Birthday" :  {
       "Year"  : 1965,
       "Month" : 8,
       "Day"   : 18
     }
   }
 }

Adding comments to JSON values

The wxJSONValue class defines some functions for adding and retrieving comment lines in a JSON value. The function for adding comments is the wxJSONValue::AddComment() function which takes two parameters:

The possible values for the position parameter are:

Here is an example:

{
  // comment before 'key-1'
  "key-1" : "value-1",
  "key-2" : "value-2", // comment inline 'key-2'
  "key-3" : "value-3"
  // comment after 'key-3'
}

To get the above output use the following code fragment:

  wxJSONValue root;
  root["key-1"] = "value-1";
  root["key-2"] = "value-2";
  root["key-3"] = "value-3";

  root["key-1"].AddComment( "// comment before", wxJSONVALUE_COMMENT_BEFORE );
  root["key-2"].AddComment( "// comment inline", wxJSONVALUE_COMMENT_INLINE );
  root["key-3"].AddComment( "// comment after", wxJSONVALUE_COMMENT_AFTER );

You have to note that comment lines are kept in an array of strings in a data member of the wxJSONValue object: this means that you can add more than one comment line to a JSON value object but remember that there is only one data member for storing the position of all comment lines. In other words, the position at which the comment lines are written in the JSON output text in the same position as the one specified in the last call to the wxJSONValue::AddComment() function.

In order to prevent that the comment's position have to be set in the last call to the AddComment() function, you can specify the wxJSONVALUE_COMMENT_DEFAULT constant as the position parameter. This constant causes the function to not modify the actual position value. If you use this constant in the first call to the AddComment() function, it is interpreted as wxJSONVALUE_COMMENT_BEFORE. Below you find an example:

  wxJSONValue root;
  root.AddComment( "// comment for root (line 1)", wxJSONVALUE_COMMENT_BEFORE );

  // no need to specify the comment position in subsequent calls to AddComment()
  // the old position is not modified
  root.AddComment( "// comment for root (line 2)" );

  // set the value for 'key-1'
  root["key-1"] = "value1";

  // now we add a comment line for 'key-1'. We do not specify the comment
  // position so it defaults to wxJSONVALUE_COMMENT_DEFAULT which cause
  // the AddCommnen() function to maintan the old position.
  // As the comment position was never set before, the wxJSONVALUE_COMMENT_BEFORE
  // will be set
  root["key-1"].AddComment( "// comment before key-1" );

  // set the value of 'key-4' an an empty object.
  // note that we cannot use the default wxJSONValue constructor to get an
  // empty object type: the default ctor constructs a NULL value object.
  root["key-4"] = wxJSONValue( wxJSONTYPE_OBJECT );

  // now we add an inline comment to 'key-4'
  root["key-4"].AddComment( "// comment inline key-4",
                     wxJSONVALUE_COMMENT_INLINE );

  // now we write the JSON 'root' value to a JSON formatted string
  // object. Note that we have to specify some flags in the wxJSONWriter
  // constructor
  wxJSONWriter writer( wxJSONWRITER_STYLED | wxJSONWRITER_WRITE_COMMENTS );
  wxString  jsonText;
  writer.Write( root, jsonText );

Below is the output text:

 // comment for root (line 1)
 // comment for root (line 2)
 {
   // comment before 'key2'
   "key-1" : "value1",
   "key-4" : {   // comment inline key-4
   }
 }

Adding inline comments

You should be carefull when adding inline comments. Comment lines are stored in an array of string, thus allowing to have more than one line of comments. This is good for comments that apear before or after the value they refer to but for inline comments, the output is not easy to read. Look at the following example:

  wxJSONValue root;
  root["key-1"] = "value1";
  root["key-1"].AddComment( " // comment inline (1)", wxJSONVALUE_COMMENT_INLINE );
  root["key-1"].AddComment( " // comment inline (2)" );
  root["key-1"].AddComment( " // comment inline (3)" );

  // this is the JSON formatted output:

{
   "key-1" : "value1", // comment inline (1)
// comment inline (2)
// comment inline (3)
}

Note that only the first line is really printed inline. The other two lines are printed after the value they refer to and without indentation: this is not very readable. For this reason, you should use inline comments only when you have only one line of comments. If you need more than one line of comment use the before or the after comment's position.

Syntax checks for comment lines

The wxJSONValue::AddComment() function checks that the string that you are adding as a comment to the JSONvalue object is a correct C/C++ comment. In other words, if you want to add a C++ comment string, the string passed as a parameter to the wxJSONValue::AddComment() function must start with two slash characters and must end with a LF. If the LF character is missing, the function adds it for you. The following code fragment shows some examples:

  wxJSONValue v1( 10 );
  v1.AddComment( "// A C++ comment line\n" );     // this is OK

  v1.AddComment( "// Another C++ comment line" ); // this is OK

  v1.AddComment( "/*  A C-style comment */");     // OK

  wxJSONValue v2( 20 );
  v2.AddComment( "A C++ comment line\n" );   // Error: does not start with '//'

  v2.AddComment( "/ A C++ comment line\n" ); // Error: does not start with '//'

  v2.AddComment( "/*** comment **" );        // Error: the close-comment is missing

  // the following is OK: new-line characters may follow
  // the end-comment characters of a C-style comment
  wxJSONValue v3( 30 );
  v2.AddComment( "/*** C comment ***/\n\n\n" );

Note that the function cannot trap all possible errors because the checks that are done by the function are very simple:

Note that the following examples are considered OK by the function but if you add those strings to some values and write them to a JSON text stream you end up with a incorrect JSON text.

  // the following is not correct: the AddComment() function only
  // appends the final LF char 
  wxJSONValue v1( 10 );
  v1.AddComment( "// Line 1\nLine2" );

  // this is the JSON output (it is not valid JSON text)
  ...
  // Line 1
  Line 2
  10
  ...

You would have to write:

  wxJSONValue v1( 10 );
  v1.AddComment( "// Line 1" );
  v1.AddComment( "// Line 2" );

Nested C-style comments are not handled correctly by the wxJSON parser:

  wxJSONValue v2( 20 );
  v2.AddComment( "/* comment1 /* comment2 */ */" );

  // this is the JSON text output:
  ...
  /* comment1 /* comment2 */ */
  20
  ...

The parser will report an error when it reads the last close-comment characters because when a C-style comment starts, all characters until the first close-comment chars are ignored by the parser.

Reading comment lines from JSON text

As already written above, comment lines are a wxJSON extension to the JSON format specification. Comments may be directly added to wxJSONValue objects using memberfunctions or they can be stored in the values when a JSON formatted text input is read by the parser. Note that by default the wxJSONReader class recognizes C/C++ comments in the input text but simply ignores them: if you want to store the comments in the value they refer to you have to pass some flags to the parser's constructor:

  // this ctor is error tolerant and stores comments
  wxJSONReader reader1( wxJSONREADER_TOLERANT | wxJSONREADER_STORE_COMMENTS );

  // this ctor is not error tolerant: wxJSON extensions are off
  // the parser does not recognize comments: they are reported as errors
  wxJSONReader reader2( wxJSONREADER_STRICT );

  // this ctor is error tolerant but does not store comments
  wxJSONReader reader3;

  // this ctor recognizes all wxJSON extensions except the
  // 'multiline string' feature which is reported as an error
  // the parser also stores comments
  wxJSONReader reader1( wxJSONREADER_ALLOW_COMMENTS
                      | wxJSONREADER_CASE
                      | wxJSONREADER_MISSING
                      | wxJSONREADER_STORE_COMMENTS  );

  // parser is tolerant and stores comments but comments apear AFTER
  // the value they refer to
  wxJSONReader reader1( wxJSONREADER_TOLERANT | wxJSONREADER_STORE_COMMENTS );
                      | wxJSONREADER_COMMENTS_AFTER  );

See the wxJSONReader class's description for more informations about the wxJSON parser's extensions. Also note that the constructor's flags related to comments are only meaningfull if the main flags are also specified. In other words, the wxJSONREADER_STORE_COMMENTS flag is only meaningfull if wxJSONREADER_ALLOW_COMMENTS is also set (or the wxJSONREADER_TOLERANT constant which includes it). Also, the wxJSONREADER_COMMENTS_AFTER is only meaningfull if wxJSONREADER_STORE_COMMENTS is also set: if comments are not stored, there is no need for the parser to know the position of the comments with respect to the value.

Below you find a JSON text with many comment lines and the description of which value the comments refer to. The parser is constructed with the wxJSONREADER_STORE_COMMENT flag set, thus the parser assumes that comments apear before the value they refer to.

// comment for root (line 1)
// comment for root (line 2)
{
   "key-1" : "value1",

   // comment before 'key2'
   "key-2" : "value2",
   // comment before 'key3' (1)
   // comment before 'key3' (2)

   "key-3" : {
      "key3-1" : "value3-1",

      // comment before key3-2
      "key3-2" : "value3-2"
   },

   "key-4" : {   // comment inline key4
      // this comment does not refer to anything
   }

   "key-5" : [ // comment inline key5

      // comment before item 5-1
      "item5-1",
      "item5-2", // comment inline 5-2
      "item5-3"  // comment inline 5-3

      // this comment does not refer to anything
   ],

   "key-6"
      :        // comment inline key-6
        "value",

   "key-7" : {
      "key-7-1" : "value-7-1"
   },        // comment inline key-7

   "key-8"     // comment inline key-8(1)
      :        // comment inline key-8(2)
      "value", // comment inline key-8(3)

   "key-9" : {
      "key9-1" : 91,
      "key9-2" : 92
   }


   "key-10" : [
   ]            // comment inline key-10

   // this comment does not refer to anything
}
// this comment does not refer to anything
// if comments apear before the value

This non-JSON text is ignored by the parser because
it apears after the top-level close-object character

Unicode support in wxJSON

The JSON syntax states that JSON string values are stored in Unicode format and the encoding of a JSON text is by default UTF-8; UCS-2 (AKA UTF-16) and UCS-4 (AKA UTF-32) are also allowed (but, in fact, not much used). The wxJSON library follows this rules but because wxJSON (and wxWidgets itself) may be compiled in two different modes (by now) we have to distinguish two situations:

Also note that JSON text may be written to / read from two different kind of objects: a wxString or a stream. These two objects are very different in the way they encode strings.

Unicode support: Unicode builds

When wxJSON is built in Unicode mode, there is no problem at all. JSON text is read from / written to a wxString object or a stream in the same way because:

The conversion of wxString objects to temporary UTF-8 streams is convenient for several reasons:

When using Unicode builds you can directly hardcode JSON values in all character sets as, for example:

  wxJSONValue  value;
  value[_T("us-ascii")] = _T(""abcABC");
  value[_T("latin1")]   = _T("àèì©®");
  value[_T("greek")]    = _T("αβγδ");
  value[_T("cyrillic")] = _T("ФХЦЧ");

  wxMemoryOutputStream os;
  wxJSONWriter writer;
  writer.Write( value, os );

The above code fragment contains characters from various european languages which are incompatible in a locale dependant environment. The output memory stream contains a UTF-8 encoded text. WARNING: the possibility to directly hardcode Unicode strings such as the one I wrote before depends on your editor which has to be able to save the source file in a format that support Unicode (for example, UTF-8). Also, your compiler must be able to read and compile UTF-8 sources. This is the case of GCC on my linux machine but when I tried to run a test on windows using Borland BCC 5.5 the output was not as expected.

Note that I/O from / to stream objects is always encoded in UTF-8 format and no other format is supported by this library. If you want to encode the JSON text to a different format you can use the wxMBConv -derived classes to do the job. For example, if you want to encode the output JSON text in UCS-4, you may use the following code:

  wxJSONValue  value;
  value[_T("us-ascii")] = _T(""abcABC");
  value[_T("latin1")]   = _T("àèì©®");
  value[_T("greek")]    = _T("αβγδ");
  value[_T("cyrillic")] = _T("ФХЦЧ");

  wxMemoryOutputStream os;
  wxJSONWriter writer;
  writer.Write( value, os );
  
  // get the address of the UTF-8 buffer
  wxStreamBuffer* osBuff = os.GetOutputStreamBuffer();
  void* buffStart = osBuff->GetBufferStart();

  // first convert the UTF-8 stream to wide-chars
  wchar_t *wcBuffer;
  size_t len = wxMBConvUTF8.ToWChar(
                        0,           // wchar_t* the destination buffer
                        0,           // size_t: the destination length
                        buffStart,   // char*: the source buffer
                        os.GetLength());   // size_t: the source length

  // the conversion should never fail in unicode builds
  wxASSERT( len != wxCONV_FAILED);
  
  // allocate the destination buffer and perform conversion
  wcBuffer = new wchar_t[len + 1];
  wxMBConvUTF8.ToWChar( wcBuffer,
                        len + 1,
                        buffStart,
                        os.GetLength());

  // now convert the wide character buffer to UCS-4
  char* ucs4Buffer;
  ucs4Len = wxMBConvUTF32.FromWChar(
                        0,           // char*: the destination buffer
                        0,           // size_t: the destination length
                        buffStart,   // w_chart*: the source buffer
                        len);        // size_t: the source length
  
  // the conversion should never fail in unicode builds
  wxASSERT( ucs4Len != wxCONV_FAILED);

  // allocate the destination buffer and perform conversion
  ucs4Buffer = new char[ucs4Len + 1];
  wxMBConvUTF8.FromWChar( wcBuffer,
                            ucs4Len + 1,
                            wcBuffer,
                            len);

We have a UCS-4 buffer ready to be written to a file or trasmitted across the network. Note that UCS-4 do have endianness (byte order) issues so you should prepend a BOM (Byte Order Mark) to the stream. To know more about this topic read the wxMBConvUTF32 class's docs.

Note that the above code fragment should work fine because the selected encoding format has full support for Unicode characters. On the other hand, you cannot convert the output stream to whatever format you want; in other words, you cannot convert it to a locale dependent charset because the string contains characters from different languages which are encoded in different locale dependent charsets (by the way, the charsets that have to be used are ISO-8859-1 for Latin1, ISO-8859-7 for greek and ISO-8859-5 for cyrillic). For example, the following code does not work:

  wxJSONValue  value;
  ...
  
  // after converting to wide char we want now to convert to Latin-1
  wxCSConv latin1Conv( _T("ISO-8859-1"));
  
  // compute the length of the needed buffer
  // what we get is the wxCONV_FAILED error code
  size_t targetLen = latin1Conv.FromWChar(
                        0,           // char*: the destination buffer
                        0,           // size_t: the destination length
                        buffStart,   // w_chart*: the source buffer
                        len);        // size_t: the source length

The better solution, however, is to get a UTF-8 encoded buffer which is compatible with all other JSON implementations. To have a UTF-8 encoded JSON text just use a stream as output. You can also use a wxString object as JSON text output and then convert it to UTF-8.

  wxJSONValue  value;
  wxString     jsonText;
  wxJSONWriter writer;

  // write to a string object
  writer.Write( value, jsonText );

  // convert to UTF-8
  wxCharBuffer buffer = jsonText.ToUTF8();

However, you have to note that the output to a wxString object is obtained by writing to a UTF-8 temporary memory stream and converting it to wxString so it is very inefficient to use a string output object to get a UTF-8 encoded text (just use a stream for the output).

Is there a mean to exchange JSON data from a Unicode mode application to an ANSI one (and viceversa)? The answer is: YES, but there are several limitations.

Unicode support: ANSI builds

When wxJSON is compiled in ANSI mode both the wxJSONReader and the wxJSONWriter give you a limited Unicode support but remember that the wxJSONValue class cannot store wide char string values because the wxString class only contains one-byte character strings and the actual characters represented are locale dependent. We have to distinguish the following situations:

ANSI builds: writing to string objects

This operation does not have issues. Unlike in Unicode builds, the writer does not use a temporary UTF-8 stream to store the JSON text output. A stream is used for the actual output but it is not encoded in UTF-8: the temporary stream just contains ANSI characters which are just copied to the wxString output object.

ANSI builds: writing to stream objects

When output is sent to a stream, the default behaviour of the wxJSON library is to use a UTF-8 encoded text. This permits the JSON text to be sent to other JSON implementations and also to wxWidgets applications (that use wxJSON) built in Unicode mode.

Note that if the JSON text is to be read from the same ANSI application, you can suppress UTF-8 encoding by specifying the

    wxJSONWRITER_NOUTF8_STREAM

flag in the wxJSONWriter's constructor. This flag suppress UTF-8 encoding in the outputted JSON stream; thus, it will constain ANSI characters. Note that the JSON text produced in this way is not compatible with Unicode mode applications and is not valid JSON text so that other JSON implementations will fail to read it. Also note that the same ANSI application that wrote such ANSI text may be used in a different locale using a different character set: in this case all characters outside the US-ASCII charset (0x80..0xFF) will be misunderstood.

There is only an exception to this: wxJSONValue strings that contain UTF-8 code units that have been read from the wxJSONReader using a specific reader's flag. To know more about this read the subsection related to reading from streams.

ANSI builds: reading from string objects

This operation does not have issues. Unlike in Unicode builds, the reader does not use a temporary UTF-8 stream to read the JSON text input. A stream is used for the actual input operation but it is not encoded in UTF-8: the temporary stream just contains ANSI characters which are just copied to a wxString when the reader reads a JSON string value.

ANSI builds: reading from stream objects

When input is from a stream, the default behaviour of the wxJSON library is to read UTF-8 encoded text. When a JSON string is encontered, the reader stores all UTF-8 code-units in a temporary memory buffer. The reader tries to convert the temporary buffer to a wxString object using the wxString::FromUTF8() static function. If the conversion succeeds the string is stored in the wxString object but the conversion may fail because the UTF-8 input text may contain characters that cannot be represented in the current locale. If the UTF-8 input text cannot be converted in a wxString object then a char-by-char conversion takes place: every unrepresentable character is stored in the wxString object as a unicode escape sequence.

For example, the following UTF-8 file:

{
  "us-ascii" : "abcABC",
  "latin1"   : "àèì©®",
  "greek"    : "αβγδ",
  "cyrillic" : "ФХЦЧ"
}

is read by an ANSI application which is localized in West Europa thus using the ISO-8859-1 (Latin-1) character set. The last two string values cannot be represented in ISO-8859-1 charset so the wxJSONReader class stores unicode sequences. The following is the representation of the root wxJSONValue object:

{
  "us-ascii" : "abcABC",
  "latin1"   : "àèì©®",
  "greek"    : "\u03B1\u03B2\u03B3\u03B4",
  "cyrillic" : "\u0424\u0425\u0426\u0427"
}

In this way the original meaning of the UTF-8 JSON text is preserved and it may also be rewritten to a UTF-8 stream but there are some issues:

Someone may also think: "who cares about characters that cannot be represented in the actual locale charset? Just copy the UTF-8 text input in the wxString object".

This is simple, fast, convenient! Also, we can use the wxJSONWRITER_NOUTF8_STREAM flag to rewrite the wxJSONValue objects as they apear, without any conversion, thus producing the same, valid UTF-8 text as we read it!!

For this reason, the wxJSONReader class can be constructed with a special reader's flag, the

    wxJSONREADER_NOUTF8_STREAM

which forces the reader to not try any conversion from UTF-8 streams and just store the UTF-8 code-units in the wxJSONValue object.

WARNING: please note that both the writer's and reader's _NOUTF8_STREAM flag only have effect in ANSI builds. In Unicode builds UTF-8 streams are always converted to the native encoding of wxString objects. You cannot use the NOUTF8_STREAM flag to directly store UTF-8 code units in a wxString object in Unicode mode.

The Byte Order Mark (BOM)

A few wxJSON users wrote to me asking why the UTF-8 stream (or file) does not have a BOM such as the one that we can find at the beginning of text documents saved as UTF-8 on Windows platform.

The answer is taken from the Wikipedia (http://en.wikipedia.org/wiki/Byte_Order_Mark):

While UTF-8 does not have byte order issues, a BOM encoded in UTF-8 may
nonetheless be encountered. A UTF-8 BOM is explicitly allowed by the Unicode
standard, but is not recommended, as it only identifies a file as UTF-8 and
does not state anything about byte order. Many Windows programs (including
Windows Notepad) add BOM's to UTF-8 files by default.

As UTF-8 is the only recognized encoding of a stream we do not need a UTF-8 signature at the beginning of the stream:

The wxJSON writer's styles

The JSON text generator - wxJSONWriter class - normally generates strict JSON text which is very hard to read by humans. In order to facilitate this job, mainly for debugging purposes, the writer can write styled text using some flags in the writer's constructor. Note that some flags depend on other ones in order to take effect. The following are some examples of the text output using various flags.

The wxJSONWRITER_STYLED flag

This flag cause the wxJSON writer to produce human-readable text output. Every value is separated from the previous one by a line-feed character and sub-objects are indented by a three space characters. Here is an example:

{
   "key1" : "value1",
   "key2" : {
      "key2-1" : 12
   }
}

The wxJSONWRITER_TAB_INDENT flag

This is the same as above but the indentation is done using TABs and not the space character.This produces a more compact text and it is preferable than the normal styled output. This flag only has effect if wxJSONWRITER_STYLED is also set. Example:

{
    "key1" : "value1",
    "key2" : {
        "key2-1" : 12
    }
}

The wxJSONWRITER_NONE flag

This flag cause the writer to only produce strict JSON text without any formatting: LF characters between values are not written and no indentation is performed. Also, no comment strings are written. This is a good choice if the text has to be sent over a network connection because it produces the most compact text. Example:

{"key1" : "value1","key2" : { "key2-1" : 12 }}

Note that the JSON text output is a bit different from that of 0.x versions because in the prevoious versions this style adds LF characters between values. Linefeeds are not added in the new 1.0 version of wxJSON. The JSON output text is syntactically equal in both implementations: they only differ for the LF character. If you rely on the old text output, you can have the same results by specifying the following flags:

  wxJSONWRITER_STYLED | wxJSONWRITER_NO_INDENTATION

which actually produce the same result as the old wxJSONWRITER_NONE flag; adds LF between values but suppress indentation.

The wxJSONWRITER_NO_LINEFEEDS flag

The wxJSONWRITER_NO_INDENTATION flag

These two flags are only meaningfull if the wxJSONWRITER_STYLED was specified and cause the writer to not add line-feed characters between values and / or to remove the indentation. Because wxJSONWRITER_STYLED adds LF and performs indentation, specifying all the three flags is equal as wxJSONWRITER_NONE:

  wxJSONWriter writer( wxJSONWRITER_STYLED | 
            wxJSONWRITER_NO_INDENTATION |
            wxJSONWRITER_NO_LINEFEEDS );

  // is the same as:
  wxJSONWriter writer( wxJSONWRITER_NONE );

If the styled flag is not specfified, these two flags are useless and has no effect: it is not an error to specify them because the writer simply ignores them. The following is a text output of a value written using the wxJSONWRITER_STYLED and wxJSONWRITER_NO_LINEFEEDS flags: as you can see, LF are not printed but indentation is performed.

{   "key1" : "value1",   "key2" : {      "key2-1" : 12      }   }

Also note that the LF characters are only suppressed between values and not for comments. If you include C++ comments in the JSON text output, they need LF characters in order to mark the end of the comment. Note that C-style comments do not need a LF character to mark their end so this flag cause the LF character to be suppressed between values. LF characters that are contained in the C-style comment are not suppressed.

Example using the wxJSONWRITER_NO_LINEFEEDS and wxJSONWRITER_WRITE_COMMENTS:

{   "key1" : "value1", // C++ comment cause the LF char to be written
      "key2" : {      "key2-1" : 12   }}

The wxJSONWRITER_WRITE_COMMENTS flag

Comments are not supported by the JSON syntax specifications although many JSON implementation do recognize C/C++ comment strings. The wxJSON library also recognizes and stores C/C++ comments when a JSON text is read and is also capable to write comments. To know more about comments see Using comment lines in wxJSON.

Unlike in older 0.x versions, this flag does not depend on the wxJSONWRITER_STYLED flag. Comments may be added to the JSON text output even if indentation is suppressed. Because C++ comments rely on the LF character to mark their end, a LF character is always added to the end of a C++ comment and ti cannot be suppressed even if you specify the wxJSONWRITER_NO_LINEFEEDS flag.

Example using only this flag:

{ "key1" : "value1", // C++ comment cause the LF char to be written
"key2" : { "key2-1" : 12 } /* C-style comment */ }

Comments in JSON text are normally used for debugging purposes and, in general, because a human has to read them so the most common use of this flag is toghether with wxJSONREADER_STYLED:

{
    "key1" : "value1", // C++ comment always include a trailing LF
    "key2" : {
        "key2-1" : 12 /* C-style comment */
    }
}

Note that the wxJSON writer only writes one LF character between the first value and the second one. Because wxJSONWRITER_STYLED prints a LF between values and the C++ comment already terminates with a LF, the writer checks that a LF char was already written before writing the final LF that separates values.

The same applies for C-style comments: they do not need a terminating LF character but you can store it in the comment string: moreover, you can store more than one of them. Before writing the LF character that separates the values, the writer checks if the last character written is a LF: it is it, the final LF is not needed so it is omitted.

The wxJSONWRITER_COMMENT_BEFORE flag

The wxJSONWRITER_COMMENT_AFTER flag

The two syles are mutually exclusive and only have effect if wxJSONWRITER_WRITE_COMMENTS is also specified. If not, they are simply ignored. The flags cause the writer to always write comments before or after the value they refer to regardless the position of the comments specified in the wxJSONValue object. The following is an example using wxJSONWRITER_COMMENTS_BEFORE and the styled flag:

{
    // C++ comment always include a trailing LF
    "key1" : "value1",
    "key2" : {
         /* C-style comment */
        "key2-1" : 12
    }
}

Note that comments that are not written inline are indented using the same number of TABs (or spaces) of the value they refer to.

The wxJSONWRITER_SPLIT_STRINGS flag

This feature allows a JSON string to be splitted in two or more lines when a string value is written to JSON text. Consider the following example:

 {
    "copyright" : "This library is distributed\nunder the GNU Public License\n(C) 2007 XYZ Software"
 }

It would be by far more human readable if it would be written in the following way:

 {
    "copyright" : "This library is distributed\n"
                  "under the GNU Public License\n"
                  "(C) 2007 XYZ Software"
 }

Note that the escaped LF characters at the end of a line is enforced by splitting the string value into two or more indented lines of text as we would normally do in C/C++ sources. The wxJSONReader is capable to read such strings but you have to be aware that this is not valid JSON text and other parser may report an error. This flag has no effect if wxJSONWRITER_STYLED is not set. Note that the output text is not strict JSON: the wxJSONReader class is capable to read strings that are splitted in two or more lines but other JSON implementations may fail to recognize the generated text. Also note that the wxJSON reader class has to be constructed with a flag that does not consider splitted strings as errors.

In order to recognize such strings, the reader concatenates string values (that is values enclosed in double quotes) if a comma separator is not present. Note that only strings can be splitted into more than one line: numbers and literals cannot be splitted. The drawback is that this feature is error-prone. Consider the following example:

  {
    "seasons" :  [
      "spring",
      "summer"
      "autumn",
      "winter"
     ]
  }

The array has four elements but I forgot the comma character between the second and the third element. What I get in this case is a three-elements array where the second element is the concatenation of the two strings "summer" and "autumn" which is not what I wanted. Worse, the parser does not consider this an error: only a warning is reported if the parser is constructed with the wxJSONREADER_MULTISTRING flag.

The wxJSONWRITER_MULTILINE_STRING flag

This is a multiline string mode where newlines and tabs are not escaped. This is not valid JSON text but it helps immensely when manually editing JSON files that contains multiline strings. In the example above the JSON text output will be as follows:

 {
    "copyright" : "This library is distributed
under the GNU Public License
(C) 2007 XYZ Software"
 }

The wxJSONWRITER_RECOGNIZE_UNSIGNED flag

When the wxJSON parser reads a JSON text and enconters a string that apears as a number which value is between INT64_MIN and INT64_MAX ( or LONG_MIN and LONG_MAX if 64-bits support is disabled) it stores the value as a signed integer. For example, the numeric value 100 is read as a signed integer data type even if it was written from a wxJSONValue object that held a unsigned integer (for example a counter). This is the code:

  wxJSONValue v[ _T("counter")] = (unsigned int) 100;
  wxJSONWriter writer;
  wxString     jsonText;
  writer.Write( v, jsonText );

  // the output is:
  {
    "counter" : 100
  }

The reader cannot know that the variable that generated the value was of type unsigned and it stores the value as a signed integer. In order to force the reader use a unsigned integer in this case, the wxJSONWriter prepends a plus sign to the integer value:

  wxJSONValue v[ _T("counter")] = (unsigned int) 100;
  wxJSONWriter writer( wxJSONWRITER_RECOGNIZE_UNSIGNED );
  wxString     jsonText;
  writer.Write( v, jsonText );

  // the output is:
  {
    "counter" : +100
  }

Now the wxJSONReader class assigns the value to a unsigned integer data type. Note that this feature is not strict JSON and may be handled incorrectly by other JSON mplementations so, by default, this feature is disabled; you have to use a special wxJSONWriter's flag to get this. Also note that other JSON implementations may fail to read such integers: you should only use the feature if your applications only use the wxJSON library for reading JSON text.

Error reporting in the parser

When you read a JSON text using the wxJSONReader::Parse() function, the parser stores in the class's data members two arrays of strings: one for the errors and one for the warnings. Warnings are reported for the wxJSON extensions: if extensions are OFF, warnings are reported as errors. You get the error's and warning's arrays using the wxJSONReader memberfunctions.

The string that describes the error / warning contains the line and column number of the unrecognized text. The following is an example (line numbers are added in this documentation but does not apear in the real input text

The JSON input text (this line is ignored by the parser)
 1 {
 2    "book" :
 3    / seems a C++ comment (forgot slash)
 4    {
 5       "title"  : "The title", ,
 6       "author" : "Captain Hook"
 7       pages      : 300,
 8       "pages2" : abc300,
 9       "price"  : 30.30,
10       "price2" : 30.30abc,
11       "price3"  30,
12       "translations" :
13        [
14           "italian",,
15           "german",
16           "spanish",
17           "spanish2" : 
18        }
19   }
20

The array of errors and warnings is as follows:

Error: line 3, col 6 - Strange '/' (did you want to insert a comment?)
Error: line 5, col 33 - key or value is missing for JSON value
Error: line 7, col 14 - Value 'pages' cannot follow a value: ',' or ':' missing?
Error: line 7, col 20 - ':' not allowed where a 'name' string was already available
Error: line 7, col 25 - Value '300' cannot follow a value: ',' or ':' missing?
Error: line 8, col 26 - Value 'abc300' is incorrect (did you forget quotes?)
Error: line 8, col 26 - cannot store the value: 'value' is missing for JSON object type
Error: line 10, col 28 - Value '30.30abc' is incorrect (did you forget quotes?)
Error: line 10, col 28 - cannot store the value: 'value' is missing for JSON object type
Error: line 11, col 21 - Value '30' cannot follow a value: ',' or ':' missing?
Error: line 11, col 21 - cannot store the value: 'key' is missing for JSON object type
Error: line 14, col 23 - key or value is missing for JSON value
Error: line 17, col 24 - ':' can only used in object's values
Warning: line 18, col 10 - Trying to close an array using the '}' (close-object) char
Warning: line 21, col 1 - '}' missing at end of file

In this example the wxJSONReader was constructed with the wxJSONREADER_TOLERANT flag. If the wxJSONREADER_STRICT was used, the two warning messages at the end of the output would be errors instead of warnings.

wxJSON and the std::string

Some wxJSON's users wrote to me to have an implementation of wxJSON that works with std::string strings.

The issue when using wxJSON is that the parser stores strings in wxString objects which has to be converted to std::string. If the input is from a UTF-8 stream, the string goes through a double, sometimes unnecessary, conversion:

Also note that the internal encoding of std::string on some platforms is UTF-8 so you can understand that the final encoding is the same as the first one!

The hint of these users is: "who cares about internal encoding of strings? Just copy the UTF-8 buffer to the wxString object and let the user do the necessary conversion if needed".

Well, I think this is not a good idea. The purpose of having wxJSON is, of course, to let data be easily accessible from within wxWidgets. Because wxWidgets uses wxString intensively, you can undestand that wxJSON has to store string values in wxString and that they has to be stored in the native encoding in order to be easily accessed, processed and manipulated.

I cannot find a good solution for this issue and if some of you will have a good idea, let me know but remember the main purpose of wxJSON: it was written for wxWidgets and not for the stdlibc++ library.


Generated on Wed Dec 30 14:42:30 2009 for wxJSON by  doxygen 1.5.5