=============================
 ncDataReader2 - User Manual
=============================
-------------
version 2.4.0
-------------

.. Note: this text is written in the "reStructuredText" format. You
   may read this raw text or the PDF or HTML version generated from it.
   (http://docutils.sourceforge.net/rst.html)
.. contents::
.. sectnum::

Overview
========

ncDataReader2 is a library of C functions to access data stored in netCDF files
using different interpolation and extrapolation methods. The aim of this library
is to provide access from simulation systems like Modelica_ to data sets like
weather data or measured time rows. As such systems usually require strictly
continuity of functions and their derivatives, smooth spline interpolation is
included.

netCDF_ is a very efficient binary file format for structured multidimensional
data. The netCDF library is freely available on all major platforms.
ncDataReader2 works both with netCDF versions 3.x and 4 (which is based on
HDF5_).

ncDataReader2 supports reading one dimensional data (like generated or measured
time rows of simple quantities), using periodic extrapolation if needed.
Interpolation methods currently supported are discrete steps, linear, akima
splines and smoothed steps.

Support for variables that depend on two dimensions (scattered points, lists of
x,y,z-pairs) is included but not very well tested. The 2D functions use the csa_
library for cubic spline interpolation by Pavel Sakov.

ncDataReader2 will build as a static or dynamic library on Linux, Windows and
MacOS X.

Author and License
==================

ncDataReader2 was developed by Joerg Raedler (joerg@j-raedler.de). The code is
released under the terms of the 'GNU Lesser General License'_. The code in the
files ``csa.c``, ``csa.h``, ``csa_config.h``, ``svd.c`` and ``svd.h`` was taken
from the csa_ library which has its own open source license.


Building and Installation
=========================

The build process uses cmake_ to configure the sources. To compile ncDataReader2
you will need netCDF, cmake and a compiler/development system for C/C++.
ncDatareader2 was tested with:

 - gcc and tcc on linux platforms
 - cygwin_, MinGW_ or Microsoft Visual Studio (including
   the free `Express Edition`_) on Windows platforms
 - XCode developer tools on MacOS X

Use ``cmake`` to configure the sources and build system, then build the library
and examples. On linux you use the command ``make . && make`` in the source
folder to do this.

The installation procedure is not yet automated, you should copy the relevant
files manually to the needed location. To compile your programs with
ncDataReader2 you need a library file and the header file(s). Library files are:

 - Linux:
     * ``libncDataReader2.a`` (static) or
     * ``libncDataReader2.so`` (dynamic)
 - Windows (Visual Studio):
     * ``ncDataReader2.dll`` and ``ncDataReader2.lib`` (dynamic) or
     * ``ncDataReader2.lib`` (static)
 - Windows (MinGW or cygwin):
     * ``libncDataReader2.a`` (cygwin or MinGW)
 - MacOS X:
     * ???

The header file is called ``ncDataReader2.h`` for the general API and
``ncDataReaderEA.h`` for the easy API.


Examples
========

You will find some examples in the folder ``examples``. You should run
``GenerateFile(.exe)`` first to create the netCDF file the other example
programs will need. ``GenerateBigFile(.exe)`` will create a large file that
is used by some of the examples.

Another very simple example::

    #include "ncDataReader2.h"
    #include <stdio.h>
    int main(void) {
        NcDataSet1D *t;
        NcVar1D     *y;
        t = ncDataSet1DNew("data.nc", "time", EpPeriodic, LtFull, 10);
        y = ncVar1DNew(t, "y", IpAkima, LtFull);
        printf("The value of y for time=42.0 is %g.\n", ncVar1DGet(y, 42.0));
    }

This will open the independent variable ``time`` and the dependent variable
``y`` in a file ``data.nc``, calculate the interpolated value of ``y``
for ``time=42.0``. The variable ``time`` will be used periodic, ``y`` will be
interpolated by the Akima method. All data will be fully loaded.


Concept
=======

A one dimensional data set (NcDataSet1D) is the representation of one
independent variable in a netCDF file. This data can be equally spaced, but it
doesn't need to. A one dimensional variable (NcVar1D) is the representation of a
dependent variable that has a dependency to exactly one NcDataSet1D. The value
of a NcVar1D at a certain point can be evaluated (usually interpolated). A
NcDataSet1D can be referenced by more than one NcVar1D.

Example: a file contains weather data as time rows (e.g. hourly values). One
variable (time) contains the time values at which other quantities were
measured. The other time rows (temperature, humidity, radiation) contain the
measured values. With ncDataReader2 we would reference 'time' as a NcDataSet1D.
'temperature', 'humidity' and 'radiation' are referenced as single NcVar1D's
which are connected to the this set. For every possible value of 'time' we can
now evaluate the quantities and get (possibly interpolated) values.

You can reference the same variables in a file multiple times with different
parameters as different NcDataSet1D or NcVar1D.


Interpolation
-------------

Discrete
~~~~~~~~
This is the simplest but fastest method. The value of a variable is the value of
the last data point where the value of the independent variable is smaller or
exactly equal to the demanded point. This will lead to steps at the interval
boundaries. Neither the function nor the derivatives are continuous.

Linear
~~~~~~
Linear interpolation between the points leads to a continuous function with
non-continuous derivatives and is very fast.

Akima
~~~~~
Akima interpolation is a cubic spline interpolation method. The calculation is
not fast, but the result is a very smooth function (continuous curve and
derivative). The continuity of the second derivative was abandoned to get only
local dependencies of the parameters. This is a big advantage in comparison to
normal cubic splines where all values of a data set have to be taken into
account to calculate a single value.

By using Akima interpolation with ncDataReader2 you can get a smooth
interpolation of very large variables by reading only a few values of the
required range from the file.

Sine Steps (SinSteps)
~~~~~~~~~~~~~~~~~~~~~
This is a variation of the discrete method where the steps are smoothed by
inserting parts of the sine function. In the middle of an interval the values
are still constant (step-like behavior) but at the interval boundaries a
smooth transition is ensured. The amount of smoothing can be configured by
defined by an influence radius around the points. The resulting curve is
continuous and has a continuous first derivative. Strictly speaking, this is not
an interpolation method since the data points are usually not met.

.. image:: Interpolation.png
   :width: 95%

Cosine Window (CosWin)
~~~~~~~~~~~~~~~~~~~~~~
This is method does also more an approximation than a real interpolation. It
will calculate the weighted average of all points and their linear interpolation
inside a window. The weighting function is cos(x) (scaled and shifted).

The result is continuous and has a continuous first derivative. For small
windows the curve will follow the linear interpolation with some smoothing
around the points. Large windows lead to more smoothing of the data set. The
window size should be much smaller than the data range of the abscissa.

.. image:: CosWin.png
   :width: 95%


Transformations
---------------
Every variable can be automatically shifted and scaled by setting an offset and
a scale factor to avoid later conversions. This is very handy if you need to
convert between different units. A NcDataSet1D can be used in periodic mode
where the data virtually continues after the end or before the start. This way
you may use a generated weather file with one year of data to simulate severals
years with continuous time values.

Loading Data
------------
As variables in netCDF files can be very large different methods of loading are
supported. Small variables (few values) can be loaded completely into memory
to get the fastest access. The other extreme is to load every single value only
on demand. This is significant slower but still fast because of the very
efficient netCDF file access. A third possibility is to load chunks of data on
demand.

Optimization
------------
To optimize the calculation different caches can be activated. All caches are
implemented as ring buffers with a specific capacity. With a capacity of x, the
last x items are cached and can be retrieved very fast. But large capacities
will lead to a large overhead and slow down the calculation. That's why the
cache sizes can can't be optimized globally but should carefully be adapted to
the current problem. An example program CacheTests(.exe) demonstrates the
effect of the caches.

Value Cache
~~~~~~~~~~~
The value cache stores the calculated values of a NcVar1D for a specific value
of the independent variable. DAE based simulation systems tend to call the same
functions with the same arguments very often. A value cache can speed up the
calculations in this case.

Lookup Cache
~~~~~~~~~~~~
When the value of a NcVar1D is requested the first action is to search the
corresponding interval of the NcDataSet1D with a nested search over the whole
data set. For large data sets this may need a lot of time. The lookup cache
stores the last used intervals and their boundaries. If the next requested
value is in the same interval as the last one(s), this may speed up the search.

Parameter Cache
~~~~~~~~~~~~~~~
Linear and Akima interpolation methods need to calculate the parameters of a
linear or a cubic function for one interval. These parameters can be stored
in a cache. If the next requested value is in the same interval as the last
one(s), this may speed up the calculation (in particular for the Akima method).


Scattered Points (2D)
---------------------
Variables that depend on two dimensions are defined as a list of 3D points that
can be scattered in 3D space. Those points are read from a two dimensional
variable in the netCDF file. At initialization time a spline surface is
constructed from the points. Some parameters (npmin, npmax, k and nppc) affect
the construction and the quality of the spline surface.

After this step the interpolated value of z for arbitrary values of x and y can
be calculated.


File annotations / netCDF Attributes
------------------------------------
All the parameters like interpolation and extrapolation methods,
transformations, cache sizes and other can be explicitly set or can be read
from the data file. netCDF files may contain generic attributes (global and and
variable specific). Some attributes with special names are honored by
ncDataReader2. Parameters set with explicit functions always have precedence
over annotations. The following attributes are supported for netCDF variables:

  ===============  =================  ===================================
       Name         Possible Values                Meaning
  ===============  =================  ===================================
  scale_factor     float value        scaling factor
  add_offset       float value        offset
  extrapolation    "default",         extrapolation method for indep.
                   "periodic"         variables
  interpolation    "discrete",        interpolation method for dep.
                   "linear",          variables
                   "akima",
                   "sinsteps"
  load_type        "auto",            data load type
                   "full",
                   "none",
                   "chunk"
  chunk_size       integer value      chunk size for data loading
  smoothing        float value        smoothing radius for SinSteps
  window_size      float value        window size for CosWin
  lookup_cache     integer value      capacity of lookup cache
  value_cache      integer value      capacity of value cache
  parameter_cache  integer value      capacity of parameter cache
  csa_npmin        integer value      npmin parameter for csa
  csa_npmax        integer value      npmax parameter for csa
  csa_k            integer value      k parameter for csa
  csa_nppc         integer value      nppc parameter for csa
  ===============  =================  ===================================


General API
===========

To use the general API you should include the header file ``ncDataReader2.h``.
Data sets and variables are represented by C-structs. You should not try to
initialize or destroy these struct objects yourself but to use the provided
functions instead.

NcDataSet1D
-----------
NcDataSet1D is a struct object which holds all information on a data set. A new
NcDataSet1D will be created with the following function::

  NcDataSet1D *ncDataSet1DNew(const char *fileName,
                              const char *varName,
                              Extrapolation extra,
                              LoadType loadType,
                              size_t lookupCacheSize);

``fileName`` is the name of the file, ``varName`` is the name of the independent
variable. ``extra`` is the method of extrapolation (the behavior when the
defined data range is left). The variable ``loadType`` defines the way the data
is loaded from file to memory. The parameter ``lookupCacheSize`` is the size of
the lookup cache for this data set. Use ``NC_LOOKUP_CACHE_AUTO`` to read this
value from the file annotation. If not set, no cache will be used.

Possible values for the extrapolation method (``Extrapolation``) are:

 * ``EpDefault`` - use a method corresponding to the interpolation method.
   This is the first or last value for discrete or sine steps and the linear
   cubic extrapolation using the parameters of the first/last interval for
   linear or Akima extrapolation.

 * ``EpPeriodic`` - adjust values for periodic use. The first and last values
   of the data set must mark the boundaries of the periodic range.
   Example: time-dependent values for one whole day should start with a value
   for 0:00 and end with a value for 24:00 to get a daily periodic data set.
   If the first and last values of a NcVar1D are not equal, they will be
   replaced with an average transition value.

 * ``EpConstant`` - use the border values when outside.

 * ``EpAuto`` - the extrapolation method will be read from the file annotation.
   If not set ``EpDefault`` will be used.

The load type (``LoadType``) can be one of the following:

 * ``LtFull`` - the full variable will be loaded to memory.

 * ``LtNone`` - every single value will be read from the file on demand.

 * ``LtAuto`` - use the file annotation. If not set, ``LtFull`` will be used for
   small variables and ``LtNone`` for large ones. The limit is defined as
   ``LARGE_DATASET`` in ``ncDataReader2.h``.

 * ``LtChunk`` - load chunks of data on demand. The size can be set with an
   option. For a NcDataSet1D this will usually be slower than ``LtNone``
   because the interval search needs the whole data range.

A NcDataSet1D should be freed with the following function when no more NcVar1D
is connected to it. This will release all used memory and close netCDF objects
like variables and files.::

  void ncDataSet1DFree(NcDataSet1D *dataSet);

Interval search for a value of the independent variable::

  size_t ncDataSet1DSearch(NcDataSet1D *dataSet, double *x);

Get the value for one interval::

  double ncDataSet1DGetItem(NcDataSet1D *dataSet, size_t i);

Set an option for a data set with this var-arg function::

  int ncDataSet1DSetOption(NcDataSet1D *dataSet, DataSetOption option, ...);

Possible options are:

 * ``OpDataSetScaling`` - set scaling and offset of the variable with the
   following two double arguments. This corresponds to the netCDF attributes
   ``scale_factor`` and ``add_offset``.

 * ``OpDataSetLookupCacheSize`` - change the capacity of the lookup cache to
   the value of the following integer value.

 * ``OpDataSetChunkSize`` - change the chunk size.


NcVar1D
-------
NcVar1D is a struct object which holds all information on a variable. For
existing data set objects new variables can be defined::

  NcVar1D *ncVar1DNew(NcDataSet1D *dataSet,
                      const char *varName,
                      Interpolation inter,
                      LoadType loadType);

``dataSet`` is a ``NcDataSet1D`` object, ``varName`` the name of the dependent
variable in the file. You may choose the interpolation method
(``Interpolation``) from the following values:

 * ``IpDiscrete`` - discrete steps

 * ``IpSinSteps`` - discrete steps with smoothing by a sine function. The
   smoothing radius can be defined by setting the ``smoothing`` option. If not
   set, a value of 0.0 will be used which will lead to the same result as
   ``IpDiscrete``.

 * ``IpLinear`` - piecewise linear Interpolation

 * ``IpAkima`` - piecewise cubic interpolation

 * ``IpCosWin`` - cosine window approximation. The window size can be defined
   by setting the ``window_size`` option. If not set, a value of 1.0 will be
   used.

 * ``IpAuto`` - determine the interpolation method from file annotations. If not
   set, ``IpAkima`` is used.

The possible values for ``LoadType`` are the same as for the ``NcDataSet1D``.

The calculation of values from a NcVar1D (which is the main purpose of this
library) is done with the function::

    double ncVar1DGet(NcVar1D *var, double x);

``var`` is the NcVar1D object and x the value of the independent variable at the
requested point.

To get the value of the variable (without any interpolation) in one interval you
may call::

    double ncVar1DGetItem(NcVar1D *var, size_t i);

A NcVar1D should be freed with the following function when it's not needed
anymore. This will release all used memory and close netCDF variable object::

  void ncVar1DFree(NcVar1D *var);

Set an option for a variable with this var-arg function::

  int ncVar1DSetOption(NcVar1D *var, VarOption option, ...);

Possible options are:

 * ``OpVarScaling`` - set scaling and offset with the following two double
   arguments. This corresponds to the netCDF attributes ``scale_factor`` and
   ``add_offset``.

 * ``OpVarSmoothing`` - set the following double value as the smoothing radius
   for the interpolation method ``IpSinSteps``. This value has to be smaller
   than the smallest interval length of the data set.

 * ``OpVarWindowSize`` - set the following double value as the window size
   for the interpolation method ``IpCosWin``. This value should be much smaller
   than the data range.

 * ``OpVarValueCacheSize`` - set the capacity of the value cache to the
   following integer value.

 * ``OpVarParameterCacheSize`` - set the capacity of the parameter cache to the
   following integer value. This is only useful for the interpolation methods
   ``IpLinear`` and ``IpAkima``.

 * ``OpVarChunkSize`` - set the chunk size to the following integer value when
   using ``LtChunk``.


NcScattered2D
-------------
NcScattered2D is a struct object which holds all information on a data set of
scattered points and its spline interpolation. A new NcScattered2D
object can be defined with::

  NcScattered2D *ncScattered2DNew(const char *fileName, const char *varName);

``fileName`` is the name of the netCDF file and ``varName`` the name of the
variable that contains the point coordinates. ``varName`` should be a two
dimensional variable (list of 3D points).

Before you can request interpolated values you have to initialize the data
(construct the spline surface) by calling::

  void ncScattered2DInit(NcScattered2D *data);

To get an interpolated value you may call::

  double ncScattered2DGet(NcScattered2D *data, double x, double y);

A NcScattered2D object should be freed after usage by calling::

  void ncScattered2DFree(NcScattered2D *data);

Several options can be set by calling this var-arg function::

  int ncScattered2DSetOption(NcScattered2D *data,
                             Scattered2DOption option,
                             ...);

This is call is valid only before ncScattered2DInit() was called! Possible
options are:

 * ``OpScattered2DScaling`` - set scaling and offset with the following two
   double arguments. This corresponds to the netCDF attributes ``scale_factor``
   and ``add_offset``. This call will scale and shift all three dimensions!

 * ``OpScattered2DScalingX``, ``OpScattered2DScalingY``,
   ``OpScattered2DScalingZ`` - set scaling and offset only in one dimension.

 * ``OpScattered2DPointsMin`` - set the npmin parameter for csa

 * ``OpScattered2DPointsMax`` - set the npmax parameter for csa

 * ``OpScattered2DPointsPerCell`` - set the nppc parameter for csa

 * ``OpScattered2DK`` - set the k parameter for csa


Error handling
--------------
netCDF functions may return errors. Errors are represented by an integer id and
a message string. The default error handler will print the message to ``stderr``
and exit the program, on Win32 systems it will open an error dialog. You may
replace this with your own handler function of the form::

  void myhandler(int id, char *message);

by calling the function::

  NcErrorHandler ncSetErrorHandler(NcErrorHandler newHandler);

This will set the function ``newHandler`` to be the new error handler and return
a pointer to the previous handler.


Access Statistics
-----------------
To tune the different optimization parameters some statistics can be dumped::

    void ncDataSet1DDumpStatistics(NcDataSet1D *dataSet, FILE *f);
    void ncVar1DDumpStatistics(NcVar1D *var, FILE *f);

This will write some statistics about the data set or the variable to a
file. ``f`` may be a writable file pointer or NULL for stdout.


Easy API (EA)
=============

The easy API was motivated by the fact that languages like Modelica cannot
handle C-structs, pointers and other language elements used in ncDataReader2.
They require simple functions that return values without large initializations
blocks and local data storage.

The EA is a wrapper around the general API of the library that hides most of its
details. To use the EA you have to include the header file ``ncDataReaderEA.h``.
The EA is based on hash tables that store data sets and variables after the first
use. The main function is::

  double ncEasyGet1D(const char *fileName, const char *varName, double x);

It will return the interpolated value of the variable ``varName`` in the netCDF
file ``fileName`` at the point ``x``. At the first call the needed NcVar1D and
NcDataSet1D objects are initialized. Following calls to this function will reuse
these objects. All parameters like extrapolation, interpolation, scaling, cache
sizes and others are read from file annotations or set to default values.

A strict requirement to get this initialization automatically done is to follow
a naming convention: the independent variable in the file must have the same
name as the dimension that is used both for the independent and the dependent
variable.

The functionality for 2D interpolation is also exposed via the EA::

  double ncEasyGetScattered2D(const char *fileName, const char *varName,
                              double x, double y);

This will return the value of ``z`` for the position defined by ``x`` and ``y``
of a spline surface. This surface represents the list of scattered points
defined by the variable ``varName`` in the netCDF file ``fileName``. All
parameters for the surface will be read from file annotations or set to default
values. At the first call to this function the data is read and the surface is
constructed, following calls will reuse the objects.

If you want to clean all stored objects of the EA, you may call::

  void ncEasyFree();

Access statistics for all open data sets and variables can be dumped with
the function::

    int ncEasyDumpStatistics(const char *fileName);

There are some more functions that return attributes of the netCDF file or of
variables::

  double ncEasyGetAttributeDouble(const char *fileName, const char *varName,
                                  const char *attName);
  int ncEasyGetAttributeLong(const char *fileName, const char *varName,
                             const char *attName);
  const char *ncEasyGetAttributeString(const char *fileName, const char *varName,
                                       const char *attName);

These functions may be used to read additional data like location coordinates
for weather files. Special values will be returned on errors (like non-existent
attributes), defined as ``NC_DOUBLE_NOVAL``, ``NC_LONG_NOVAL`` and
``NC_STRING_NOVAL``. If ``varName`` is an empty string (``""``) the global
attribute is returned.


Tools
=====

ncdr2Dump1D
-----------
This is a command line client that includes most of the functions for 1D
variables. It will dump interpolated values for a variable in textual form.
CSV format is the default, but you may choose a gnuplot compatible output or
provide your own template (for printf()). The usage is as following::

    Usage: ncdr2Dump1D [parameter] filename
        filename    name of the netCDF file or DAP-URI
    mandatory parameters:
        -v string   name of the variable
        -a string   name of the abscissa / data set
    optional parameters:
        -o string   name of output file (default: stdout)
        -s float    start of data range
        -e float    end of data range
        -n int      number of points
        -i char     interpolation:
                    [a]kima, [l]inear, [d]iscrete, [s]insteps, [c]oswin
        -x char     extrapolation:
                    [d]efault, [p]eriodic, [c]onstant
        -l char     load type:
                    [f]ull, [n]one, [c]hunks (see -h)
        -w float    window size for coswin interpolation
        -m float    smoothing radius for sinsteps interpolation (default is 0)
        -k int      size of lookup cache
        -p int      size of parameter cache
        -c int      size of chunks for chunk loading
        -t string   template string for output (used with printf())
        -g          use gnuplot-compatible output (default is CSV)
        -d          dump timing information and access statistics to stderr
        -h          print this help and exit

ncdr2ImportCSV1D
----------------
This is a command line client that converts CSV data to netCDF files that are
compatible with ncDataReader2. The special attributes that ncDataReader2
respects can be added easily. The usage is as following::

    Usage: ncdr2ImportCSV1D [parameter] filename
       filename    name of the CSV input file
    optional parameters:
       -o string   name of output file
       -t string   comment to include in file
       -h          print this help and exit
    the following is stored as attributes of the variables:
       -i char     interpolation:
                   [a]kima, [l]inear, [d]iscrete, [s]insteps, [c]oswin
       -x char     extrapolation:
                   [d]efault, [p]eriodic, [c]onstant
       -l char     load type:
                   [f]ull, [n]one, [c]hunks (see -h)
       -w float    window size for coswin interpolation
       -m float    smoothing radius for sinsteps interpolation
       -k int      size of lookup cache
       -p int      size of parameter cache
       -v int      size of value cache
       -c int      size of chunks for chunk loading

The input file is a simple CSV file with the following structure:

 * fields delimited by comma ','
 * every row has the same number of fields
 * optional header row with the variable names
 * all other fields contain just numbers
 * the decimal separator for numbers is a point '.'
 * every column contains one variable
 * first column is used as the abscissa for all variables


Modelica Interface
==================
The Modelica package contains function wrappers for the Easy API as well as some examples.


Tips and Tricks
===============

1. A template for ``ncdr2Dump1D`` can contain special characters like TAB or NEWLINE.
   In C these characters are inserted using '\\t' or '\\n'. To use these characters
   on the Linux-shell ``bash`` you can use the following syntax::

        ncdr2Dump1D -v foo -a bar -t $'%g\t%g\n' file.nc

2. CSV files generated with software in some languages (like German) will use a
   comma as decimal separator. To use such files with ncdr2ImportCSV1D you can do
   the following:

   1. export to CSV with field separator set to semicolon ';'
   2. replace every comma ',' with a point '.'
   3. replace every semicolon ';' with a comma ','

   The steps 2 and 3 can be done in a text editor or with the following command
   line (UNIX only)::

        tr ',;' '.,' <data.csv >data_corrected.csv


Changes
=======

2.4.0
 * generally worked on an improved Modelica compatibility
 * changed directory structure to improve the Modelica compatibility
 * changed Modelica files to use Modelica 3.2.1
 * removed the hard exit from the default error handler
 * added pre-built binaries for linux32, linux64, win32 and win64
 * reimplemented StringHashTable based on uthash
 * fixed memory leak for easy API (ncEasyFree was never called)

2.3.1
 * added command line client ncdr2ImportCSV1D to import CSV data
 * changed all examples to work with optional file names
 * changed Modelica files to use Modelica 3.2

2.3.0
 * added CosWin approximation
 * added constant extrapolation
 * GenerateBigFile is much faster now
 * fixed a bug with Akima and default extrapolation near the right border that
   existed for a long time
 * added functions to dump statistics
 * added error dialog for Win32, useful with Dymola
 * added command line client ncdr2Dump1D to dump interpolated values
 * built with netCDF 4 on Windows including DAP supported



.. Hyperlinks

.. _netCDF: http://www.unidata.ucar.edu/software/netcdf/
.. _HDF5: http://www.hdfgroup.org/HDF5/
.. _csa: http://code.google.com/p/csa-c/
.. _cmake: http://www.cmake.org/
.. _cygwin: http://www.cygwin.com/
.. _MinGW: http://www.mingw.org/
.. _`Express Edition`: http://www.microsoft.com/express/
.. _Modelica: http://www.modelica.org/
.. _'GNU Lesser General License': http://www.gnu.org/licenses/lgpl.html
