Re: 2.3.2 vs. 2.4 netCDF question

> Keywords: 199604222201.AA01657
> From: address@hidden (John Sheldon)
> Subject: Re: missing_values
> To: address@hidden (Steve Emmerson)
> Date: Mon, 22 Apr 1996 18:03:13 -0400 (EDT)

Hi John,

> I've got a user here who has generated two copies of a dataset, one
> using the 2.3.2 library and one using the 2.4.1 library, both on our
> C90.  In both cases, an ncdump shows the variable TEMP, for example, to
> be defined as:
> 
>         float TEMP(TIME, ZT, YT, XT) ;
>                 TEMP:long_name = "Temperature" ;
>                 TEMP:units = "deg C" ;
>                 TEMP:grid = "T-grid" ;
>                 TEMP:missing_value = 9.9999994e+34f ;
> 
> 
> For the 2.3.2 case, the data portion of the ncdump looks like:
> 
>  TEMP =
>   9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 
>     9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 
>     9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 
>     9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 
>     9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 
>     9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 
>     9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 
>     9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 
>     9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 
>     9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 , 24.18345 , 
>     24.23452 , 23.87136 , 23.70983 , 23.55063 , 23.42406 , 23.6777 , 
>     23.73085 , 23.54887 , 23.30024 , 22.97255 , 23.05056 , 23.11214 , 
>     23.03383 , 23.06242 , 23.12209 , 22.89102 , 22.86344 , 22.78358 , 
>     22.54668 , 21.9193 , 21.56955 , 21.27124 , 21.46646 , 21.50529 , 
>     21.64895 , 21.30128 , 20.80376 , 20.163 , 19.67662 , 19.38359 , 
>     19.10613 , 18.8098 , 18.43744 , 18.12563 , 17.90977 , 19.2415 , 
>     20.76276 , 22.01661 , 9.999999e+34 , 9.999999e+34 , 9.999999e+34 ,
> 
> 
> whereas for the 2.4.1 case it looks like:
> 
>  TEMP =
>   1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 
>     1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 
>     1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 
>     1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 
>     1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 1e+35 , 
>     24.18345 , 24.23452 , 23.87136 , 23.70983 , 23.55063 , 23.42406 , 
>     23.6777 , 23.73085 , 23.54887 , 23.30024 , 22.97255 , 23.05056 , 
>     23.11214 , 23.03383 , 23.06242 , 23.12209 , 22.89102 , 22.86344 , 
>     22.78358 , 22.54668 , 21.91931 , 21.56956 , 21.27124 , 21.46646 , 
>     21.50529 , 21.64896 , 21.30128 , 20.80376 , 20.163 , 19.67662 , 
>     19.38359 , 19.10613 , 18.8098 , 18.43745 , 18.12563 , 17.90977 , 
>     19.2415 , 20.76276 , 22.01661 , 1e+35 , 1e+35 , 1e+35 ,

This one's a mystery to me.  I can't duplicate the problem on the the
platforms I have access to.  When I use the default floating-point missing
value instead, everything seems to work fine on both 2.3.2 and 2.4.1.

Since ncdump is only printing the default 7 significant digits for floating 
point
constants, I can't tell from the above output whether the values of TEMP
printed as "9.999999e+34" or "1e+35" are actually equal to the 
TEMP:missing_value
attribute or not.  If you could use more precision for the ncdump output
with a "-d 9" or even "-d 10" option to ncdump and send back the same
extract of the output, it might be helpful.

> What happens is that the downstream plotting codes seem unable to
> detect that there are "missing_values" in the second (2.4.1) file.
> There seems to be a mismatch in how "1e+35" is stored in the attribute
> vs. how such values stored in the data portion.
> 
> Any ideas?  (Sorry for the sparse info earlier...rough day here...)
> If you want to see the files themselves, let me know.  (They are about
> 7.6MB apiece, but I might be able to ncextr a part if that's too big.)

If you could make the files (or smaller versions of them) available for FTP,
that might help.  The ideal would be if you could send us a small program
that created files that demonstrated the problem, so we could see if we
could duplicate it on any of our machines or on the Cray YMP we have access
to.

Thanks.

--Russ

______________________________________________________________________________

Russ Rew                                           UCAR Unidata Program
address@hidden                              http://www.unidata.ucar.edu