New in SVN
----------
* Updated the type system's behavior, in order to better support backwards
compatibility with code that was written before 64-bit integer support was
introduced. Here's how it works now:
* isInt, isInt64, isUInt, and isUInt64 return true if and only if the
value can be exactly represented as that type. In particular, a value
constructed with a double like 17.0 will now return true for all of
these methods.
* isDouble and isFloat now return true for all numeric values, since all
numeric values can be converted to a double or float without
truncation. Note however that the conversion may not be exact -- for
example, doubles cannot exactly represent all integers above 2^53 + 1.
* isBool, isNull, isString, isArray, and isObject now return true if and
only if the value is of that type.
* isConvertibleTo(fooValue) indicates that it is safe to call asFoo.
(For each type foo, isFoo always implies isConvertibleTo(fooValue).)
asFoo returns an approximate or exact representation as appropriate.
For example, a double value may be truncated when asInt is called.
* For backwards compatibility with old code, isConvertibleTo(intValue)
may return false even if type() == intValue. This is because the value
may have been constructed with a 64-bit integer larger than maxInt,
and calling asInt() would cause an exception. If you're writing new
code, use isInt64 to find out whether the value is exactly
representable using an Int64, or asDouble() combined with minInt64 and
maxInt64 to figure out whether it is approximately representable.
New in JsonCpp 0.6.0:
---------------------
* Compilation
- LD_LIBRARY_PATH and LIBRARY_PATH environment variables are now
propagated to the build environment as this is required for some
compiler installation.
- Added support for Microsoft Visual Studio 2008 (bug #2930462):
The platform "msvc90" has been added.
Notes: you need to setup the environment by running vcvars32.bat
(e.g. MSVC 2008 command prompt in start menu) before running scons.
- Added support for amalgamated source and header generation (a la sqlite).
Refer to README.txt section "Generating amalgamated source and header"
for detail.
* Value
- Removed experimental ValueAllocator, it caused static
initialization/destruction order issues (bug #2934500).
The DefaultValueAllocator has been inlined in code.
- Added support for 64 bits integer:
Types Json::Int64 and Json::UInt64 have been added. They are aliased
to 64 bits integers on system that support them (based on __int64 on
Microsoft Visual Studio platform, and long long on other platforms).
Types Json::LargestInt and Json::LargestUInt have been added. They are
aliased to the largest integer type supported:
either Json::Int/Json::UInt or Json::Int64/Json::UInt64 respectively.
Json::Value::asInt() and Json::Value::asUInt() still returns plain
"int" based types, but asserts if an attempt is made to retrieve
a 64 bits value that can not represented as the return type.
Json::Value::asInt64() and Json::Value::asUInt64() have been added
to obtain the 64 bits integer value.
Json::Value::asLargestInt() and Json::Value::asLargestUInt() returns
the integer as a LargestInt/LargestUInt respectively. Those functions
functions are typically used when implementing writer.
The reader attempts to read number as 64 bits integer, and fall back
to reading a double if the number is not in the range of 64 bits
integer.
Warning: Json::Value::asInt() and Json::Value::asUInt() now returns
long long. This changes break code that was passing the return value
to *printf() function.
Support for 64 bits integer can be disabled by defining the macro
JSON_NO_INT64 (uncomment it in json/config.h for example), though
it should have no impact on existing usage.
- The type Json::ArrayIndex is used for indexes of a JSON value array. It
is an unsigned int (typically 32 bits).
- Array index can be passed as int to operator[], allowing use of literal:
Json::Value array;
array.append( 1234 );
int value = array[0].asInt(); // did not compile previously
- Added float Json::Value::asFloat() to obtain a floating point value as a
float (avoid lost of precision warning caused by used of asDouble()
to initialize a float).
* Reader
- Renamed Reader::getFormatedErrorMessages() to getFormattedErrorMessages.
Bug #3023708 (Formatted has 2 't'). The old member function is deprecated
but still present for backward compatibility.
* Tests
- Added test to ensure that the escape sequence "\/" is corrected handled
by the parser.
* Bug fixes
- Bug #3139677: JSON [1 2 3] was incorrectly parsed as [1, 3]. Error is now
correctly detected.
- Bug #3139678: stack buffer overflow when parsing a double with a
length of 32 characters.
- Fixed Value::operator <= implementation (had the semantic of operator >=).
Found when adding unit tests for comparison operators.
- Value::compare() is now const and has an actual implementation with
unit tests.
- Bug #2407932: strpbrk() can fail for NULL pointer.
- Bug #3306345: Fixed minor typo in Path::resolve().
- Bug #3314841/#3306896: errors in amalgamate.py
- Fixed some Coverity warnings and line-endings.
* License
- See file LICENSE for details. Basically JsonCpp is now licensed under
MIT license, or public domain if desired and recognized in your jurisdiction.
Thanks to Stephan G. Beal [http://wanderinghorse.net/home/stephan/]) who
helped figuring out the solution to the public domain issue.