.NET C# float.Parse returning different result to double.Parse [duplicate]

Question

Can some more experienced guys explain this strange error I found today? I was getting strange amounts when I was loading table data with my C# script.

As it turns out the problem is a different output from similar functions:

string amount_in_string = "1234567,15";
double amount_in_double = double.Parse(amount_in_string);
float amount_in_float = float.Parse(amount_in_string);

//amount_in_double = 1234567.15;
//amount_in_float = 1234567.13;

Why do I get such a different result when float and double are similar types (floating point). Can the precision make a difference with small amounts like these?

Answer 1

When “1234567.15” is converted to double, the result is the closest value representable in double, which is 1234567.1499999999068677425384521484375. Although you report in the question that the value is 1234567.15, the actual value is 1234567.1499999999068677425384521484375. “1234567.15” would be displayed when the value is displayed with a limited number of decimal digits.

When “1234567.15” is converted to float, the result is the closet value representable in float, which is 1234567.125. Although you report the value is 1234567.13, the actual value is 1234567.125. “1234567.13” may be displayed when the value is displayed with a limited number of decimal digits.

---

Observe that 1234567 exceeds 1,048,576, which is 2²⁰. The 32-bit floating-point format used for float uses 24 bits for the significand (fraction portion of the number). If the high bit of the significand represents 2²⁰, the low bit represents 2²⁰⁻²³ = 2⁻³ = ⅛. This is why you see “1234567.15” converted to a value rounded to the nearest eighth.

Answer 2

Floating point numbers are never exact, they are representations of numbers. An example commonly used is think of

1/3 + 1/3 = 2/3

...so the answer in floating point numbers, .33333 + .33333, is not 2/3rds exactly, it is .66666.

Long story short, the more precise fraction you take that can't be converted to an exact binary is going to always have a rounding number. The more precise the more likely it will have rounding errors.

Keep in mind if you do multiple different fractions, you can even have multiple different rounding errors that either make the number accidentally correct, or even further off.

Answer 3

A fiddle you can see the results (there is a culture problem here too)

https://dotnetfiddle.net/Lnv1q7

    string amount_in_string = "1234567,15"; // NOTE THE COMMA in original
    string amount_in_string = "1234567.15"; //my correct culture
    double amount_in_double = double.Parse(amount_in_string);
    float amount_in_float = float.Parse(amount_in_string);

    Console.WriteLine(amount_in_string);
    Console.WriteLine(amount_in_double);
    Console.WriteLine(amount_in_float);

Results (parsing in the incorrect culture!)

1234567,15
123456715
1.234567E+08

Results (parsing in the correct culture!)

1234567.15
1234567.15
1234567

Another one that demonstrates the loss of precision with float

float flt = 1F/3;
double dbl = 1D/3;
decimal dcm = 1M/3;
Console.WriteLine("float: {0} double: {1} decimal: {2}", flt, dbl, dcm);

Result

float: 0.3333333  
double: 0.333333333333333  
decimal: 0.3333333333333333333333333333

floats should only be used in cases were some loss of precision is not extremely valuable. This is because floats are 32bit where decimal is 128 bit floats are mainly used in pixel coordinates, and loss of precision doesn't matter because a the consumer translates locations to more precise coordinates any way.

in .NET floats "should" be actively avoided unless you don't care about loss of (some) precision. Which is probably never now a days (unless you writting games)

This is where the banking problem came from when 100th's of a penny/cent across transactions where lost, seemingly invisible per transaction but amounting to large amounts of "missing" money.

Use decimal