Unexpected precision loss on parsing numbers

[ivanstojic]

I'm seeing an unexpected result when parsing numbers from JSON, and how they parse to doubles. Sample source code:

#include "rapidjson/document.h"
#include "rapidjson/writer.h"
#include "rapidjson/stringbuffer.h"
#include <iostream>

using namespace rapidjson;

int main() {
    const char* json = "{\"thing\":-1.1030161663305387}";
    Document d;
    d.Parse(json);

    StringBuffer roundtripBuffer;
    Writer<StringBuffer> beforeWriter(roundtripBuffer);
    d.Accept(beforeWriter);

    Value& s = d["thing"];
    double thingVal = -1.1030161663305387;
    s.SetDouble(thingVal);

    StringBuffer nativeSetBuffer;
    Writer<StringBuffer> afterWriter(nativeSetBuffer);
    d.Accept(afterWriter);

    std::cout << json << std::endl;
    std::cout << roundtripBuffer.GetString() << std::endl;
    std::cout << nativeSetBuffer.GetString() << std::endl;
    return 0;
}

Output of this program is:

ivans@cram:~$ g++ simpledom.cpp && ./a.out
{"thing":-1.1030161663305387}
{"thing":-1.103016166330539}
{"thing":-1.1030161663305387}

The code takes a number (first line of output) and parses is from a JSON string, then takes that RapidJSON Document and reencodes it to string (second line). Then I reset the value of the property with a native float value, and use a writer to produce another JSON representation. That's the third line.

What I would have expected: all three lines to be the same.

[michal2612]

@ivanstojic I am facing same issue. Have you managed to solve this?