| Re: differences idl5.5 and idl5.6 [message #34073] |
Wed, 19 February 2003 10:17  |
R.G. Stockwell
Messages: 363
Registered: July 1999
|
Senior Member |
|
|
Kenneth Bowman wrote:
> In article <b30ble$dpt7$1@zam602.zam.kfa-juelich.de>,
> Reimar Bauer <R.Bauer@fz-juelich.de> wrote:
>
>
>> on aix the result is a bit more different,
>> IDL> a=1./0 & print,a,fix(a)
>> Inf -1
>>
>>
>> I would prefer as result NaN!
>
>
> The IEEE standard defines distinct special values for Inf (result of
> division by zero) and NaN (result of operations on Inf's and Nan's).
> Being able to distinguish them is probably a good thing. The FINITE
> function has keywords for this purpose. Without keywords, FINITE
> detects both.
>
> Unfortunately, perhaps, standard integer formats do not include
> "special" values to represent inifinties, Nan's, etc. You have to
> handle them yourself. Hence all those ASCII data sets filled with
> -999's!
>
> Conversion of IEEE Inf's and NaN's to integers may well be system
> dependent.
>
> Ken
With integers using two's complement, there is no representation
for these (nan/inf) values (every bit pattern is a valid integer).
The standard way (in C Java IDL etc, perhaps a IEEE standard) to handle
integer division is to flag an exception if the divisor is 0, and
the result is undefined.
It seems that IDL (sensibly) follows the rule of integer math results
in an integer (rather than automatically casting the result to a float, and the
appropriate nan/inf). It is up to the programer to decide if they want
a float division or an integer division.
Also, casting a floating point nan/inf to integer should also throw
an exception (i.e. there is no conversion of inf/nan to integer).
Cheers,
bob
|
|
|
|
| Re: differences idl5.5 and idl5.6 [message #34074 is a reply to message #34073] |
Wed, 19 February 2003 09:26  |
K. Bowman
Messages: 330
Registered: May 2000
|
Senior Member |
|
|
In article <b30ble$dpt7$1@zam602.zam.kfa-juelich.de>,
Reimar Bauer <R.Bauer@fz-juelich.de> wrote:
> on aix the result is a bit more different,
> IDL> a=1./0 & print,a,fix(a)
> Inf -1
>
>
> I would prefer as result NaN!
The IEEE standard defines distinct special values for Inf (result of
division by zero) and NaN (result of operations on Inf's and Nan's).
Being able to distinguish them is probably a good thing. The FINITE
function has keywords for this purpose. Without keywords, FINITE
detects both.
Unfortunately, perhaps, standard integer formats do not include
"special" values to represent inifinties, Nan's, etc. You have to
handle them yourself. Hence all those ASCII data sets filled with
-999's!
Conversion of IEEE Inf's and NaN's to integers may well be system
dependent.
Ken
|
|
|
|
| Re: differences idl5.5 and idl5.6 [message #34075 is a reply to message #34074] |
Wed, 19 February 2003 08:56 |
Paul Van Delst[1]
Messages: 1157
Registered: April 2002
|
Senior Member |
|
|
Reimar Bauer wrote:
>
> Dick Jackson wrote:
>> "Reimar Bauer" <R.Bauer@fz-juelich.de> wrote in message
>> news:3E539FE4.4000801@fz-juelich.de...
>>
>>> Hi,
>>>
>>> today I found a difference in handling infite numbers by the new
>>
>> version.
>>
>>>
>>> IDL 5.6
>>>
>>> IDL> a=1./0 &print,a,fix(a)
>>> Inf -32768
>>>
>>>
>>> IDL 5.5
>>>
>>> IDL> a=1./0 &print,a,fix(a)
>>> Inf 0
>>>
>>>
>>>
>>> So I can't say which is right both results are terrible.
>>>
>>> Now it is totally clear that this case must be tested.
>>
>>
>> I don't see any change in behaviour on Windows 2000 Pro; in 5.4, 5.5 and
>> 5.6, I get these identical results:
>>
>> IDL> a=1./0 &print,a,fix(a)
>> Inf 0
>> % Program caused arithmetic error: Floating divide by 0
>> % Program caused arithmetic error: Floating illegal operand
>>
>> Cheers,
>> --
>> -Dick
>>
>> Dick Jackson / dick@d-jackson.com
>> D-Jackson Software Consulting / http://www.d-jackson.com
>> Calgary, Alberta, Canada / +1-403-242-7398 / Fax: 241-7392
>>
>>
>
> Thanks Dick,
>
> then it seems to be a bug on linux and aix
>
> on aix the result is a bit more different,
> IDL> a=1./0 & print,a,fix(a)
> Inf -1
>
> I would prefer as result NaN!
Hmm. Wouldn't the integer form of infinity still be infinite? If you think the value of
some variable may be infinite, how come you don't test the value _before_ using it in an
intrinsic function (e.g. FIX()) ?
I.e.
IF ( FINITE( a ) ) THEN a = FIX( a ) ELSE a = !VALUES.F_NAN
And from the IDL docs, it would appear that there is no such thing as an integer infinity
or nan (in IDL at least).
paulv
--
Paul van Delst
CIMSS @ NOAA/NCEP/EMC
Ph: (301)763-8000 x7274
Fax:(301)763-8545
|
|
|
|
| Re: differences idl5.5 and idl5.6 [message #34077 is a reply to message #34075] |
Wed, 19 February 2003 08:33 |
R.Bauer
Messages: 1424
Registered: November 1998
|
Senior Member |
|
|
Dick Jackson wrote:
> "Reimar Bauer" <R.Bauer@fz-juelich.de> wrote in message
> news:3E539FE4.4000801@fz-juelich.de...
>
>> Hi,
>>
>> today I found a difference in handling infite numbers by the new
>
> version.
>
>>
>> IDL 5.6
>>
>> IDL> a=1./0 &print,a,fix(a)
>> Inf -32768
>>
>>
>> IDL 5.5
>>
>> IDL> a=1./0 &print,a,fix(a)
>> Inf 0
>>
>>
>>
>> So I can't say which is right both results are terrible.
>>
>> Now it is totally clear that this case must be tested.
>
>
> I don't see any change in behaviour on Windows 2000 Pro; in 5.4, 5.5 and
> 5.6, I get these identical results:
>
> IDL> a=1./0 &print,a,fix(a)
> Inf 0
> % Program caused arithmetic error: Floating divide by 0
> % Program caused arithmetic error: Floating illegal operand
>
> Cheers,
> --
> -Dick
>
> Dick Jackson / dick@d-jackson.com
> D-Jackson Software Consulting / http://www.d-jackson.com
> Calgary, Alberta, Canada / +1-403-242-7398 / Fax: 241-7392
>
>
Thanks Dick,
then it seems to be a bug on linux and aix
on aix the result is a bit more different,
IDL> a=1./0 & print,a,fix(a)
Inf -1
I would prefer as result NaN!
regrads
Reimar
--
Reimar Bauer
Institut fuer Stratosphaerische Chemie (ICG-I)
Forschungszentrum Juelich
email: R.Bauer@fz-juelich.de
------------------------------------------------------------ -------
a IDL library at ForschungsZentrum Juelich
http://www.fz-juelich.de/icg/icg-i/idl_icglib/idl_lib_intro. html
============================================================ =======
|
|
|
|
| Re: differences idl5.5 and idl5.6 [message #34079 is a reply to message #34077] |
Wed, 19 February 2003 07:54 |
Dick Jackson
Messages: 347
Registered: August 1998
|
Senior Member |
|
|
"Reimar Bauer" <R.Bauer@fz-juelich.de> wrote in message
news:3E539FE4.4000801@fz-juelich.de...
> Hi,
>
> today I found a difference in handling infite numbers by the new
version.
>
>
> IDL 5.6
>
> IDL> a=1./0 &print,a,fix(a)
> Inf -32768
>
>
> IDL 5.5
>
> IDL> a=1./0 &print,a,fix(a)
> Inf 0
>
>
>
> So I can't say which is right both results are terrible.
>
> Now it is totally clear that this case must be tested.
I don't see any change in behaviour on Windows 2000 Pro; in 5.4, 5.5 and
5.6, I get these identical results:
IDL> a=1./0 &print,a,fix(a)
Inf 0
% Program caused arithmetic error: Floating divide by 0
% Program caused arithmetic error: Floating illegal operand
Cheers,
--
-Dick
Dick Jackson / dick@d-jackson.com
D-Jackson Software Consulting / http://www.d-jackson.com
Calgary, Alberta, Canada / +1-403-242-7398 / Fax: 241-7392
|
|
|
|
| Re: differences idl5.5 and idl5.6 [message #34218 is a reply to message #34075] |
Thu, 20 February 2003 01:01 |
R.Bauer
Messages: 1424
Registered: November 1998
|
Senior Member |
|
|
Paul van Delst wrote:
> Reimar Bauer wrote:
>
>> Dick Jackson wrote:
>>
>>> "Reimar Bauer" <R.Bauer@fz-juelich.de> wrote in message
>>> news:3E539FE4.4000801@fz-juelich.de...
>>>
>>>
>>>> Hi,
>>>>
>>>> today I found a difference in handling infite numbers by the new
>>>
>>> version.
>>>
>>>
>>>> IDL 5.6
>>>>
>>>> IDL> a=1./0 &print,a,fix(a)
>>>> Inf -32768
>>>>
>>>>
>>>> IDL 5.5
>>>>
>>>> IDL> a=1./0 &print,a,fix(a)
>>>> Inf 0
>>>>
>>>>
>>>>
>>>> So I can't say which is right both results are terrible.
>>>>
>>>> Now it is totally clear that this case must be tested.
>>>
>>>
>>> I don't see any change in behaviour on Windows 2000 Pro; in 5.4, 5.5 and
>>> 5.6, I get these identical results:
>>>
>>> IDL> a=1./0 &print,a,fix(a)
>>> Inf 0
>>> % Program caused arithmetic error: Floating divide by 0
>>> % Program caused arithmetic error: Floating illegal operand
>>>
>>> Cheers,
>>> --
>>> -Dick
>>>
>>> Dick Jackson / dick@d-jackson.com
>>> D-Jackson Software Consulting / http://www.d-jackson.com
>>> Calgary, Alberta, Canada / +1-403-242-7398 / Fax: 241-7392
>>>
>>>
>>
>> Thanks Dick,
>>
>> then it seems to be a bug on linux and aix
>>
>> on aix the result is a bit more different,
>> IDL> a=1./0 & print,a,fix(a)
>> Inf -1
>>
>> I would prefer as result NaN!
>
>
> Hmm. Wouldn't the integer form of infinity still be infinite? If you think the value of
> some variable may be infinite, how come you don't test the value _before_ using it in an
> intrinsic function (e.g. FIX()) ?
Normally we check this but sometimes we missed it. The problem now is
that it is not easy to find because of the different results it gives on
different platforms.
The routine which was going wrong now was written in 1997. As it was
programmed no one thougt about that it makes sense to use this routine
with only one color.
And for the people which were using it with only one color it works
perfectly in the past.
For crossplatform programming it is better that failures are described
same on each of the idl platforms. (My feeling) In the past it was this way.
regards
Reimar
>
> I.e.
>
> IF ( FINITE( a ) ) THEN a = FIX( a ) ELSE a = !VALUES.F_NAN
>
> And from the IDL docs, it would appear that there is no such thing as an integer infinity
> or nan (in IDL at least).
>
> paulv
>
--
Reimar Bauer
Institut fuer Stratosphaerische Chemie (ICG-I)
Forschungszentrum Juelich
email: R.Bauer@fz-juelich.de
------------------------------------------------------------ -------
a IDL library at ForschungsZentrum Juelich
http://www.fz-juelich.de/icg/icg-i/idl_icglib/idl_lib_intro. html
============================================================ =======
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
