Arduino parsing Double is incorrect

Question

Im having some problems with parsing gps coordinates with the arduino ide.

To be honest I want to process the data without a library.

From my sensor I get this String: 4815.98450

The result of String.toDouble(): 4815.9843750

I also parsed it on my own, but i got the same result: (yes this code is not nice ;))

char buf[10];
 msg.toCharArray(buf, 19);
 double num = 0.0;
 int c;
 c = buf[0] - '0';
 double num0 = (c * 1000);
 c = buf[1] - '0';
 double num1 = (c * 100);
 c = buf[2] - '0';
 double num2 = (c * 10);
 c = buf[3] - '0';
 double num3 = (c * 1);
 c = buf[5] - '0';
 double num5 = (c * 0.1);
 c = buf[6] - '0';
 double num6 = (c * 0.01);
 c = buf[7] - '0';
 double num7 = (c * 0.001);
 c = buf[8] - '0';
 double num8 = (c * 0.0001);
 c = buf[9] - '0';
 double num9 = (c * 0.00001);
 num = num + num0;
 num = num + num1;
 num = num + num2;
 num = num + num3;
 num = num + num5;
 num = num + num6;
 num = num + num7;
 num = num + num8;
 num = num + num9;

Does somebody have a solution for my problems?

Thank you very much :=)

Answer 1

There is no easy solution. The Arduino you are using does not support double precision floating point. The type double is just an alias for float, which has 24-bit accuracy.

Your program is actually giving you the correct result, as the float closest to 4815.98450 is exactly 4815.984375.

• Benjamin_Ellmer: Thank you for your answer! You helped me a lot because now i know it is no bug. But now im searching for a workaround :)
• Edgar Bonet: @Benjamin_Ellmer: Maybe store the integer and fractional parts separately?

Answer 2

double (or float) is just an inappropriate data type.

Perhaps you take int(4815) and float(0.98450) as two numbers?

---

And BTW, I bet from your sensor you don't get a String object.

• Benjamin_Ellmer: Im reading the output from the TX of my sensor, so im pretty sure this is a string ;) i figured out that i can do my calculation without the first 2 digits of the number. there are some cases where it does not work but those cases are very unlikely

Answer 3

You can convert this to a long, do your arithmetic, and convert it back to an ASCII-Decimal number:

Remove the decimal point from the string. Convert the remaining string of digits to a long. Now you have the coordinate in binary, in units of the least significant decimal digit; 1 lsb == 0.00001, if you used 5 decimal places.

Manipulate your data however you need to using long integer arithmetic.

Divide the value by 10^ -(# of fraction digits you used). The quotient is the whole number part; the remainder is the fractional part. Print the quotient and a decimal point. Print the remainder padding with zeros to the left as necessary to result in your printing (# of fraction digits you used) total fraction digits.

The output conversion is even easier if you're willing to use sprintf():
Divide the value by 10^ -(# of fraction digits you used), as above. Let's assume 5 places as in your example:

char buffer[BIGENOUGH+1];
sprintf(buffer, "%ld.%05ld", value/100000, value%100000);

Note that I made two divisions to get the quotient and the remainder and assumed you could tolerate the performance hit. If you're expecting to crank out a lot of values and speed is a factor, you could look up or write a division routine that can return both the quotient and the remainder results from a single division.

Update: @edgar_bonet points out that the compiler is smart enough to notice this fragment uses both results of the numerically same division and avoids dividing twice after all.

• Edgar Bonet: Re the "routine that can return both the quotient and the remainder results from a single division": this is exactly what avr-gcc does when compiling the example you provided.