ESP8266 "ovf" when doing small double maths - arduino

Question

I'm using a Wemos D1 based on the ESP8266 wifi chip with the Arduino C framework to do some simple small math.
As far as I can gather, double precision is available so I'm using it - with a maximum of something like 1.8*10^103.
But I'm getting an ovf when I try to calculate a number around 5*10^8.
Any ideas please?
void setup(){
Serial.begin(9600);
while(!Serial);
Serial.println("Hit any key to start math");
double te = 6800;
double res = te * te;
Serial.println(res);
te = 68000;
res = te * te;
Serial.println(res);
te = 680000;
res = te * te;
Serial.println(res);
}
void loop(){
}
Prints
46240000.00
ovf
ovf

Answer 1

Arduino does supports 4-byte double data type. and your code is indeed produced a valid double value that is not overflow (ovf).
The problem that you see has to do with the way Arduino implement Serial.print() function which is almost like a half-cooked hack to support floating-point and is not as reliable as vprintf() that is available in avr-libc. You can see the source code here which print ovf for anything bigger than 4294967040.0 or smaller than -4294967040.0. I thought that ESP8266 Arduino Core has fixed this instead of inheriting the ugly implementation of Arduino Serial.print(), but apparently not.
Luckily the ESP8266 Arduino Core does have a Serial.printf() method that provide better floating point rendering. This code will shows that your number is indeed a valid number for double data type.
void setup(){
Serial.begin(9600);
while(!Serial);
double te = 68000;
double res = te * te;
Serial.println(res); //this will produce 'ovf'
Serial.printf("Result=%f\n", res); //this will produce correct 4624000000.000000
}
Please noted that Serial.printf() is ESP8266 specific implementation to support full floating point vprintf() in the standard library. It is not available for standard Arduino Boards.
For Arduino boards, there is a way to use sprintf() which inherits from vprintf() to print the correct floating point with some twist on linker options during code compilation process. I have a blog post Do you know Arduino - sprintf() and floating point talk about how to do that.