On 4/21/20 9:43 AM, Chris Angelico wrote: > Right. Now try diagnosing weird bugs in floating-point calculations, > and you aren't sure if it's because sometimes you have 53 bits of > mantissa and sometimes only 52. > And sometimes only 1. The smallest positive floating point number has only 1 significant digit (leading 0's are NOT significant), so the second smallest floating point number is twice that number. The key is that once you hit the denormals, you no longer have a relative accuracy like in the normal numbers, but all the denormals have the same absolute accuracy.
-- Richard Damon -- https://mail.python.org/mailman/listinfo/python-list