[rg3]

Note in C you can get the correct result if you use long doubles, which normally go to 80 bits[1]:

printf("%Lf\n", 9999999999999999.0L - 9999999999999998.0L);

In my x86_64 computer it breaks when you add enough digits. At this point it started outputting 0.0 as the difference:

printf("%Lf\n", 99999999999999999999.0L - 99999999999999999998.0L);

With 63 bits for the fraction part you more or less get around 19 decimal digits of precision, and the expression above uses 20 significant digits.

[1] https://en.wikipedia.org/wiki/Extended_precision#x86_extende...