comp.lang.idl-pvwave archive: archive » Re: Testing for NODATA presence in a dataset

Home » Public Forums » archive » Re: Testing for NODATA presence in a dataset

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

Re: Testing for NODATA presence in a dataset [message #33464 is a reply to message #33374]

Mon, 30 December 2002 06:52

tam
Messages: 48
Registered: February 2000

Member

David Fanning wrote:
> Tom McGlynn (tam@lheapop.gsfc.nasa.gov) writes and
> Bill Thompson confirms:
>
>
>> That doesn't distinguish NaN from the infinities.
>> The standard trick in any language for looking for NaN's is
>>
>> if x ne x then begin
>> print,'This is a NaN'
>> endif else ...
>
>
> Humm, well, consider this little test in IDL 5.5 or 5.6
> for Windows:
>
> IDL> a = [ 1.0, 2.0, !Values.F_NAN, 4.0, !Values.F_NAN ]
> IDL> print, a
> 1.00000 2.00000 NaN 4.00000 NaN
> IDL> print, a(1)
> 2.00000
>
> All well and good so far. Test the algorithm.
>
> IDL> if a(1) ne a(1) THEN print, 'NAN' ELSE print, 'Number'
> Number
>
> Perfect. Working fine. Now text NAN.
>
> IDL> print, a(2)
> NaN
> IDL> if a(2) ne a(2) THEN print, 'NAN' ELSE print, 'Number'
> Number
> % Program caused arithmetic error: Floating illegal operand
>
> Oh, oh. What's up with that? And a floating illegal operand to
> boot. :-(
>
> How about the array in general?
>
> IDL> print, array ne array
> 0 0 0 0 0
> % Program caused arithmetic error: Floating illegal operand
>
> Humm. I presume you guys have a reason for thinking
> like you do. Any insights?
>
> Cheers,
>
> David

Just to follow up on Bill's message.... I did warn in my first message
that interpreters had been known to screw up this comparison, but I
believe the behaviour you see above is clearly non-compliant with
the IEEE 754 floating point standard. I almost never run IDL
under Windows, but I'd call this a bug -- though I daresay RSI
will call it a feature.

Using IDL 5.2 under Linux I have:

IDL> a=sqrt(-1)
%Program caused arithmetic error. Floating illegal operand.
IDL> print, a
-NaN
IDL> print a ne a
1
IDL> z=[0,0,a,a,0]
IDL> print, z ne z
0 0 1 1 0

I believe this to be 'correct' behavior but it appears that
it is not universally implemented this way within IDL. Of
course IDL has been implemented on non-IEEE machines (e.g., VAX)
and so completely consistent behavior may be impossible.

Let me add my apologies though for misleading anyone looking for
how to actually do something, rather than how they should be able
to do it.

Regards,
Tom McGlynn

Report message to a moderator

[Message index]

		Re: Testing for NODATA presence in a dataset By: thompson on Thu, 26 December 2002 12:58
		Re: Testing for NODATA presence in a dataset By: David Fanning on Thu, 26 December 2002 08:25
		Re: Testing for NODATA presence in a dataset By: thompson on Tue, 24 December 2002 07:11
		Re: Testing for NODATA presence in a dataset By: Kenneth P. Bowman on Mon, 23 December 2002 06:42
		Re: Testing for NODATA presence in a dataset By: tam on Sun, 22 December 2002 18:33
		Re: Testing for NODATA presence in a dataset By: David Fanning on Sat, 21 December 2002 14:20
		Re: Testing for NODATA presence in a dataset By: tam on Mon, 30 December 2002 06:52

Previous Topic:	Can't get an cell select event
Next Topic:	3d sphere revisited

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Sat Oct 11 16:53:14 PDT 2025

Total time taken to generate the page: 0.23425 seconds