## digitalmars.D - Check floats for .nan

- Acarion <asdfa asf .com> Oct 07 2007
- BCS <ao pathlink.com> Oct 07 2007
- Daniel Keep <daniel.keep.lists gmail.com> Oct 07 2007
- Nathan Reed <nathaniel.reed gmail.com> Oct 07 2007
- Sean Kelly <sean f4.ca> Oct 07 2007
- "Janice Caron" <caron800 googlemail.com> Oct 07 2007

I have this code: float number=getFloat(); assert(number!=float.nan); //this passes assert(to!(char[])(number)!="nan"); //this fails writefln(string.toString(number)); //this outputs: "nan" writefln( to!(char[])(to!(long)(number)) ); //this generates an error What did I do wrong?

Oct 07 2007

Reply to acarion,I have this code: float number=getFloat(); assert(number!=float.nan); //this passes

IIRC this will also pass: assert(float.nan != float.nan); this is because D used IEEE floating point semantics. Under IEEE a comparison with NaN will always be not equal If you want to test for nan use "testedValue !<>= 0"* or the isnan functionassert(to!(char[])(number)!="nan"); //this fails writefln(string.toString(number)); //this outputs: "nan" writefln( to!(char[])(to!(long)(number)) ); //this generates an error What did I do wrong?

* this is the not grater than, less than, or equal to operator.

Oct 07 2007

BCS wrote:Reply to acarion,I have this code: float number=getFloat(); assert(number!=float.nan); //this passes

IIRC this will also pass: assert(float.nan != float.nan); this is because D used IEEE floating point semantics. Under IEEE a comparison with NaN will always be not equal If you want to test for nan use "testedValue !<>= 0"* or the isnan function

You can also use (x!=x) to test to see if x is NaN, and (x==x) to make sure it is not NaN. -- Daniel

Oct 07 2007

Janice Caron wrote:On 10/7/07, BCS <ao pathlink.com> wrote:If you want to test for nan use "testedValue !<>= 0"

Are we really sure about that? If testedValue is complex (with non-zero real part) then it will not be less than 0, it will not be greater than zero, and it will not be equal to zero. But it will also not be NaN.

I believe you'd get a type error in that case (real and creal can't be directly compared, etc.) Haven't tried it though. Thanks, Nathan Reed

Oct 07 2007

BCS wrote:Reply to acarion,I have this code: float number=getFloat(); assert(number!=float.nan); //this passes

IIRC this will also pass: assert(float.nan != float.nan); this is because D used IEEE floating point semantics. Under IEEE a comparison with NaN will always be not equal If you want to test for nan use "testedValue !<>= 0"* or the isnan function

I tend to use: testedValue !<>= testedValue Though I've occasionally wondered whether the 'is' identity operator should work for checking against nan: testedValue is float.nan Sean

Oct 07 2007

On 10/7/07, BCS <ao pathlink.com> wrote:If you want to test for nan use "testedValue !<>= 0"

Are we really sure about that? If testedValue is complex (with non-zero real part) then it will not be less than 0, it will not be greater than zero, and it will not be equal to zero. But it will also not be NaN.

Oct 07 2007