
cprnc README
------------

cprnc is a generic tool for analyzing a netcdf file or comparing
two netcdf files.


Quick Start Guide:
------------------

> setenv INC_NETCDF $local_netcdf_include_path
> setenv LIB_NETCDF $local_netcdf_lib_path
> gmake 
> ./cprnc file1          # for analyze mode
> ./cprnc file1 file2    # for compare mode

 Usage: cprnc [-m] [-v] [-ipr iprs[:ipre]] [-jpr jprs[:jpre]] [-kpr kprs[:kpre]] file1 [file2]
 -m: Compare each time sample. Default is false, i.e. match "time" 
     coordinate values before comparing
 -v: Verbose output
 -ipr iprs[:ipre] -jpr jprs[:jpre] -kpr kprs[:kpre]:
     Print variable values for the specified 1st index subrange. If present,
     -jpr jprs[:jpre] and -kpr kprs[:kpre] must also be specified for 2nd and
     3rd index subranges, respectively. For variables with only 1 or 2 non-time
     dimensions, -jpr and -kpr settings which do not apply will be ignored.


Users Guide:
------------

cprnc is a Fortran-90 application. It relies on netcdf version 3 or
later and uses the f77 netcdf interfaces.  It requires a netcdf include
file and a netcdf library.  The Makefile assumes those are in 
/usr/local/include and /usr/local/lib respectively.  Users can override 
these defaults by setting environment variables INC_NETCDF and LIB_NETCDF 
to the local values of the netcdf paths.  The Makefile has entries for 
IRIX64, AIX, Darwin, Cray X1E, OSF1, SUN, and Linux.  On Linux, entries 
exist for pgf90 and lf95 compilers. The Makefile is quite extensible for 
new architectures.  gmake should be used to build cprnc.  The Makefile
relies on a hard-wired Depends file.

cprnc generates an ascii output file via standard out.  It initially
summarizes some characteristics of the input file[s].  A compare file is
generally 132 characters wide and an analyze file is less than 80 
characters wide.

In analyze mode, the output for a field looks like

                      (   lon,   lat,  time, -----)
              259200  (   587,   134,     1) (   269,    59,     1)
 FX1           96369   8.273160400390625E+02  0.000000000000000E+00
            avg abs field values:   9.052845920820910E+01

and a guide to this information is printed at the top of the file

                      (  dim1,  dim2,  dim3,  dim4)
              ARRSIZ1 ( indx1, indx2, indx3) file 1
 FIELD        NVALID           MAX                   MIN


The first 10 characters of the field name are identified in the first
  dozen columns of the third line.  
The first line summarizes the names of the dimensions of the field
The second line summarizes the indices of the maximum and minimum value 
  of the field for the first three dimensions.  If the fourth dimension 
  exists, it's always assumed to be time.  Time is handled separately.
The third line summarizes the number of valid values in the array
  and the maximum and minimum value over those valid values.  Invalid
  values are values that are identified to be "fill" value.
The last line summarizes some overall statistics including the average 
  absolute value of the valid values of the field.

In comparison mode, the output (132 chars wide) for a field looks like

               96369  (   lon,   lat,  time, -----)
              259200  (   422,   198,     1) (   203,   186,     1)           (    47,   169,     1)         (   224,   171,     1)
 FIRA          96369   1.466549530029297E+02 -3.922052764892578E+01   1.4E+02 -3.037954139709473E+01 1.0E+00 -3.979958057403564E+00
               96369   1.321966247558594E+02 -1.603044700622559E+01            1.084177169799805E+02          3.982142448425293E+00
              259200  (   156,    31,     1) (   573,   178,     1)           (
          avg abs field values:    6.778244097051392E+01    rms diff: 1.4E+01   avg rel diff(npos):  4.6E-02
                                   5.960437961084186E+01                        avg decimal digits(ndif):  1.2 worst:  0.0

and a guide to this information is printed at the top of the file

              NDIFFS  (  dim1,  dim2,  dim3,  dim4)
              ARRSIZ1 ( indx1, indx2, indx3) file 1
 FIELD        NVALID1          MAX1                  MIN1            DIFFMAX  VALUES                RDIFMAX  VALUES
              NVALID2          MAX2                  MIN2
              ARRSIZ2 ( indx1, indx2, indx3) file 2

The information content is identical to the information in analyze
mode with the following additions.  Two additional lines are added
in the main body.  Lines 4 and 5 are identical to line 3 and 2
respectively but are associated with file 2 instead of file 1.
In addition, the right hand side of lines 2, 3, and 4 contain
information about the maximum difference, the location and values
of the maximum difference, the relative difference and the location
and values of the maximum relative difference.  The last two line
summarize some overall statistics including average absolute values
of the field on the two files, rms difference, average relative 
difference, average number of digits that match, and the worst
case for the number of digits that match.

At the end of the output file, a summary is presented that looks like

SUMMARY of cprnc:
  A total number of    119 fields were compared
          of which      83 has non-zero differences
          including      0 that had inconsistent size/shape
  A total number of      0 fields in file 1 were analyzed (non-compare mode)
  A total number of     10 fields in file 1 could not be analyzed at all
  A total number of      0 fields in file 2 were analyzed (non-compare mode)

This summarizes the number of fields that were compared, that differed
and the number of fields that could not be compared but were analyzed.


Developers Guide:
-----------------

The tool works as follows.

Fields can be analyzed if they "kosher".  That is, they
are real (float or double), have between 0 and 4 dimensions 
of length greater than 0 length.  If the field has a time 
dimension, it must be the last dimension.  Support for fields 
that are integers is planned. 

In general, "kosher" fields that appear on both files are 
compared.  If they are different shapes or sizes, no difference
statistics are computed and only a summary of the fields on
the files are presented.  If "kosher" fields only appear
on one file, those fields are analyzed.

The time dimension is treated uniquely.  In general, for files
that have a dimension named "time", the time axes are compared
and matching time values on the two files are compared one
timestep at a time.  Time values that don't match are skipped.
To override the matching behaviour, use cprnc -m.  In this mode, 
timestamps are compared in indexical space.  In analyze mode,
the fields are analyzed one timestamp at a time.  In general,
if there is a "time" axis, it will be the outer-most loop in
the output analysis.  In compare mode, fields with a time axis
and a timestamp that are not common between the two files are
ignored.

The time and variable loops as well as associated logic is
in driver.f90.  The statistics are computed in gatherstats.f90.

