Problem with nc_get_varm_float function???

• To: NetCDF mailing list <netcdfgroup@xxxxxxxxxxxxxxxx>
• Subject: Problem with nc_get_varm_float function???
• From: Timothy Hume <tim.hume@xxxxxxxxx>
• Date: Wed, 25 Oct 2000 12:30:35 +1300 (NZDT)

Hi people,

I _may_ have encountered a problem with the nc_get_varm_float function in
the version 3.4 NetCDF C library. I would be interested to know if anyone
else has encountered a similar problem.

Basically, I have a NetCDF file containing a lot of 4 byte floating point
numbers. My initial investigations indicate the NetCDF file is not corrupt
... the data is exactly as it should be.

However, when I use the nc_get_varm_float function to read this data back,
the read back data is sometimes slightly wrong. It would seem that only a
few of the floating point numbers I read back get "corrupted" (4 or 5
floats out of 5 or 6 thousand). What seems to be the problem is that
occasionally bit number 19 (where the sign bit is bit 1, the exponent bits
are 2-9, and the mantissa bits 10-32) is wrong.

I am using NetCDF on a pentium II machine with redhat 6.2, however I can
exactly duplicate this problem on a PIII with redhat 6.0. Shortly I will
test the function on a Sun workstation. The problem can also be duplicated
using the beta version of the 3.5 library.

Maybe the problem has nothing to do with the NetCDF function ... I am
still in the "detective" stage, trying to identify the cause of
it. However, I would be interested to know if anyone else has encountered
similar problems.

Regards,

Tim Hume

-- 
------------------------------------------------------------------------
Tim Hume
Mesoscale Modeller
Meteorological Service of New Zealand
PO Box 722
Wellington
New Zealand
------------------------------------------------------------------------