Floating point precision lost

[nickvkaam]

Input:

json data;
data["number"] = 11.4f;

output:

std::string dataString = data.dump();

result:

{
    "number": 11.3999996185303
}

[FabianoGK]

That is because you cannot represent 11.4 as a single-precision floating-point number. Take a look into ExploringBinary - Floating Point Converter.

Here is an example having the literal number on different precisions:

Input:

json data;
data["number-float"] = 11.4f;
data["number-double"] = 11.4;
data["number-long-double"] = 11.4l;

Output:

std::string dataString = data.dump();

Result:

{
    "number-double":11.4,
    "number-float":11.3999996185303,
    "number-long-double":11.4
}

[phoenixchinar]

You need to use decimal floating point to solve this issue. Normal float, double, and long double are based on a base of 2 and cannot represent all base 10 floating point numbers accurately. gcc has decimal32, decimal64 and decimal128 classes as well as underlying C types defined to address this.
In your compiler try this
if((0.1+0.2)==0.3)
std::cout<<"true"<<endl;
else
std::cout<<"false"<<endl;

You will need to change the underlying types, as the library serializes the underlying types accurately.

[nlohmann]

Can this ticket be closed?