OpenEuphoria

Re: power overflow

• Posted by "Carl W." <euphoria at cyreksoft.yorks.com> Jun 05, 2002
• 424 views

a.tammer at hetnet.nl wrote:

> There's one way, to solve the problems on this issue, I guess.

> If you use 'Anomaly' as Carl White calls it, it's an FPvalue,
> and a strange one too, it's 0 AND ! 0
> I think Carl will allow me to give you it's exact specs.

I called it 'Anomaly' before I looked into what value it actually
represents, because it really is an anomalous number in the way that
Euphoria handles it. [Note that some other languages have the similar
problems]

A better name for 'Anomaly' would be something like 'float64_minimum' or
'tiniest_atom' since it's true value is power(2, -1074). Surprisingly, the
power() function actually generates it:

constant
  X       = power(2, -1073), -- Anomaly-like, but not the smallest possible.
  Anomaly = power(2, -1074), -- Smallest representable IEEE-64bit atom.
  Y       = power(2, -1075), -- Too small to represent.
  Zero    = 0

? X              -- prints 0
? X       = Zero -- prints 0, so it *looks*like* X = 0 and != 0

? Anomaly        -- prints 0
? Anomaly = Zero -- prints 0, so it *looks*like* Anomaly = 0 and != 0

? Y              -- prints 0
? Y       = Zero -- prints 1 - Too small so is identical to 0.

Also, if you try dividing by Anomaly (or 'X'), you'll get a 'Divide by Zero'
error even though Anomaly is really greater than Zero. This is down to the
fact that 1/power(2,-1074) = power(2,1074) which is too big to fit in an
atom or a float64.

> Specs 'Anomaly': FP64 63 0binary positions, last position = 1
> last= LSD [not the dope, dope-heads ]

I was going to suggest using float64_to_atom({0,0,0,0,0,0,0,1}) [or is that
float64_to_atom({1,0,0,0,0,0,0,0})?], but using power() doesn't require an
include and it's easier to remember which way round to put the numbers. ;)

Carl

--
[ Carl R White -=- aka -=- Cyrek the Illogical ]
[ () E-mail...:     cyrek{}cyreksoft.yorks.com ]
[ /\ URL......: http://www.cyreksoft.yorks.com ]