• Joshua Reusch (5/6) Feb 14 2012 there is the std.math.isNaN function which works correctly, but why can
• Bernard Helyer (3/9) Feb 14 2012 Use `float.nan is float.nan`; all other expressions with NaNs in
• =?UTF-8?B?QWxpIMOHZWhyZWxp?= (10/23) Feb 14 2012 In the general case, std.math.isNaN works:
• bearophile (6/9) Feb 14 2012 By design, the hardware that manages floating point numbers makes a NaN ...
• Joshua Reusch (2/11) Feb 14 2012
• =?utf-8?Q?Simen_Kj=C3=A6r=C3=A5s?= (4/10) Feb 14 2012 My favorite explanation: Would you expect this to pass?
• Stewart Gordon (10/15) Feb 14 2012 A NaN typically denotes some kind of invalid computation. If the result...
• Manfred Nowak (6/7) Feb 14 2012 ... it would be called not-a-boolean. Of course it may make sense to

Joshua Reusch <yoschi arkandos.de> writes:

Hello,

why does this assertion fail:

 assert(float.nan == float.nan);

there is the std.math.isNaN function which works correctly, but why can 
I not just use the comparison ?

Thanks, Joshua

"Bernard Helyer" <b.helyer gmail.com> writes:

On Tuesday, 14 February 2012 at 15:39:37 UTC, Joshua Reusch wrote:
 Hello,

 why does this assertion fail:

 assert(float.nan == float.nan);

 there is the std.math.isNaN function which works correctly, but 
 why can I not just use the comparison ?

 Thanks, Joshua

Use `float.nan is float.nan`; all other expressions with NaNs in 
them will be false (or result in NaN).

=?UTF-8?B?QWxpIMOHZWhyZWxp?= <acehreli yahoo.com> writes:

On 02/14/2012 07:48 AM, Bernard Helyer wrote:
 On Tuesday, 14 February 2012 at 15:39:37 UTC, Joshua Reusch wrote:
 Hello,

 why does this assertion fail:

 assert(float.nan == float.nan);

 there is the std.math.isNaN function which works correctly, but why
 can I not just use the comparison ?

 Thanks, Joshua

 Use `float.nan is float.nan`; all other expressions with NaNs in them
 will be false (or result in NaN).

In the general case, std.math.isNaN works:

import std.math;

void main()
{
     float f;
     assert(f is float.nan);   // fails
     assert(isNaN(f));         // passes
}

Ali

bearophile <bearophileHUGS lycos.com> writes:

Joshua Reusch:

 why does this assertion fail:
 
  > assert(float.nan == float.nan);

By design, the hardware that manages floating point numbers makes a NaN not
equal to everything else, including other NaNs:
http://en.wikipedia.org/wiki/NaN

In D2 "is" performs a bitwise comparison, but keep in mind there are many
different NaNs (I think double.nan and double.init are different in the bits
too).

Bye,
bearophile

Joshua Reusch <yoschi arkandos.de> writes:

Thank you for the explanation !

bearophile wrote:
 Joshua Reusch:

 why does this assertion fail:

   >  assert(float.nan == float.nan);

 By design, the hardware that manages floating point numbers makes a NaN not
equal to everything else, including other NaNs:
 http://en.wikipedia.org/wiki/NaN

 In D2 "is" performs a bitwise comparison, but keep in mind there are many
different NaNs (I think double.nan and double.init are different in the bits
too).

 Bye,
 bearophile

=?utf-8?Q?Simen_Kj=C3=A6r=C3=A5s?= <simen.kjaras gmail.com> writes:

On Tue, 14 Feb 2012 16:39:37 +0100, Joshua Reusch <yoschi arkandos.de>  
wrote:

 Hello,

 why does this assertion fail:

  > assert(float.nan == float.nan);

 there is the std.math.isNaN function which works correctly, but why can  
 I not just use the comparison ?

 Thanks, Joshua

My favorite explanation: Would you expect this to pass?

assert( sqrt(-1.0) == 0.0 / 0.0 );

Stewart Gordon <smjg_1998 yahoo.com> writes:

On 14/02/2012 15:39, Joshua Reusch wrote:
 Hello,

 why does this assertion fail:

 assert(float.nan == float.nan);

 there is the std.math.isNaN function which works correctly, but why can I not
just use the
 comparison ?

A NaN typically denotes some kind of invalid computation.  If the results of
two invalid 
computations were considered equal, it would probably cause problems.

Another way to think of it is to consider NaN as an unknown value.  Two unknown
values 
cannot be considered to be equal to each other.  OK, so whether they're equal
or not is 
actually unknown.  If there were such a thing as bool.nan, equality comparisons
in which 
at least one operand is NaN would probably evaluate to it.  Indeed, null in SQL
works this 
way.  But in the absence of this in D, the best design has turned out to be to
consider 
NaNs unequal.

Stewart.

Manfred Nowak <svv1999 hotmail.com> writes:

Stewart Gordon wrote:

  If there were such a thing as bool.nan

... it would be called not-a-boolean. Of course it may make sense to 
compute something using such poisoned values. But if such values make 
sense, D is not prepared to use them, especially there is no 
"if_then_else_otherwise" statement in D.

-manfred