Hi There,
I just recently recompiled my netCDF v3.5.1 Fortran API with the latest
PGI v5.2-1 f95 compiler. I'm now having problem with some Fortran90 code
I wrote that has been working up till now. I thought I'd ask this group
in case someone else has had a similar problem and has some sort of work
around.
I have a character array, each element of which is 512 characters in
length, that I want to write to a netCDF file,
INTEGER, PARAMETER :: PDSL = 512
CHARACTER( PDSL ), DIMENSION( : ), ALLOCATABLE :: Description
How I do this, after allocating the array to some length, e.g.
ALLOCATE( Description( n_Profiles ) )
is to first define the dimensions:
! -- The number of profiles
NF90_Status = NF90_DEF_DIM( NC_FileID, &
PROFILE_DIMNAME, &
n_Profiles, &
Profile_DimID )
! -- The string length
NF90_Status = NF90_DEF_DIM( NC_FileID, &
DESCRIPTION_DIMNAME, &
PDSL, &
PD_StrLen_DimID )
Then I would define the variable itself,
NF90_Status = NF90_DEF_VAR( NC_FileID, &
DESCRIPTION_VARNAME, &
NF90_CHAR, &
DimIDs = (/ PD_StrLen_DimID, Profile_DimID /), &
VarID = VarID )
In my old files, the description variable appears like so from ncdump:
netcdf UMBC.AtmProfile {
dimensions:
n_profiles = 48 ;
pdsl = 512 ;
variables:
char profile_description(n_profiles, pdsl) ;
profile_description:long_name = "Description of profile" ;
...etc...
where I've taken into account the difference in dimension ordering
between Fortran and C.
Now, when I run this code linking to the PGI v5.2-1 compiler library,
the NF90_DEF_VAR() function call fails with the error:
"NC_UNLIMITED in the wrong index"
When I run the code linking to the Lahey lf95 v6.2 compiled library, the
code generates no such error.
I run on linux systems running under Red Hat Enterprise WSE3.0
Does anyone there have any ideas about what could be causing this error
so I can try a work around with the PGI compiler?
cheers,
paulv
--
Paul van Delst
CIMSS @ NOAA/NCEP/EMC
