re: Fun With Floating Point Arithmetic, Part Three

Norman Diamond — Tue, 18 Jan 2005 02:12:00 GMT

Sounds like this:
var x = 9223372036854777856
print(x)
would print 9223372036854777000
instead of 9223372036854778000

Did anyone ever complain about the rounding being wrong? Of course you pointed us to a standard which allows the rounding to be wrong, and efforts to provide nice "round trip" conversions do require rounding to be wrong in some cases, but anyone who doesn't understand floating point well enough will surely not understand these points, and we might expect them to complain.

I really dislike this part of both the standard and the behaviour. Any programmer who doesn't understand floating point well enough for this should be doing other kinds of programming instead.

The following looks like hyperbole but unfortunately it isn't. Imagine a programmer who doesn't understand floating point well enough and yet uses floating point in controlling a car. In fact imagine doing a printf() with a precision of 3 decimal places, reading it back in, and depending on the imagined pseudo-fact that the value will be the same as before it was printf()'ed. The result could be a crash. After my boss told me that the original programmer was not a programmer and did not understand floating point, I understood to take the original code as a rough model instead of a spec.

"Not Understanding Floating Point"

Dave — Tue, 18 Jan 2005 19:59:00 GMT

"...did not understand floating point..."

If floating point was obvious and easy to understand then there is no way Eric could milk it for several long blog entries. :)

re: Fun With Floating Point Arithmetic, Part Three

Randall Knapp — Wed, 16 Dec 2009 02:47:13 GMT

I quote:

# k³ 1, 10k-1 £ s < 10k, the number value for s ´ 10n-k is m

What? I think some kind of character encoding got messed up here. What is this actual equation, and would you repeat the other equations in the algorithm?