$URL: https://repos.deltares.nl/repos/openda/public/trunk/core/native/TODO $
$Revision: 761 $, $Date: 2008-12-19 12:52:42 +0100 (Fri, 19 Dec 2008) $

COSTA is a work in progress. You are welcome to contribute. If you wish to
do so please let us know at info@costapse.org.

Be sure that you become familiar with the basic concepts of COSTA and its
design ideas first. For this you may use the Getting Started document
(see `doc/index.html') and browse through the Developers Guide.

Here a list of ideas for extension and refinement of the COSTA system.


------------------------------------------------------------------------ 

 * the Getting Started document should be extended. 
    - It should contain a complete hello-world application, for which the
      code could be provided in the examples-directory as well.
    - It should provide an example that illustrates the use of COSTA trees.
    - The programming conventions should be described, such as use of
      handles, build-up of component- and subroutine-names, etc.
    - The interfacing with OpenDA should be elaborated upon.
    - The description of the model-builders should be extended.
    - The description of parallel computing should be made consistent with
      the newest ideas on this.
    - A complete step-by-step guide should be added for the migration of an
      existing elementary model to a COSTA-application.

------------------------------------------------------------------------ 

 * in the Getting Started document the terminology for the use of object
   orientation in COSTA is slightly modified. Most data types are now
   called "classes" instead of "components". This change in terminology
   should be implemented throughout the COSTA system (code and
   documentation).

------------------------------------------------------------------------ 

 * a few elementary "methods" should be provided such that the migration
   of existing model code can be done step-by-step
    - the no-asm method should be added, that performs a deterministic
      model simulation without data-assimilation;
    - the no-asm method should be able to produce "observations" for
      performing a twin experiment;
    - the repeat-method should be added, that performs each time step of
      the deterministic model twice, and that tests whether the model step
      is repeatable.

------------------------------------------------------------------------ 

 * the *_tool-directories under old_stuff/methods contain rough
   implementations of data assimilation methods that extend on the existing
   methods available in src/methods. These extensions should be integrated
   into the central implementations in src/methods.

------------------------------------------------------------------------ 

 * the ideas with respect to object orientation in COSTA might be worked
   out further, among others with respect to different implementations of
   a base class. For instance the ideas for inheritance of a distributed
   vector or a sparse vector from the generic vector class might be worked
   out in detail.

------------------------------------------------------------------------ 

 * the time-class might be extended with various time-scales and
   representations, time-zones, etc.

------------------------------------------------------------------------ 

 * a simple hello-world application should be provided in the examples
   directory, illustrating the basics of Costa and the Costa string class.

------------------------------------------------------------------------ 

cta_datatypes.h:
 * remove usage of CTA_STRLEN_NAME and CTA_STRLEN_TAG
 * think of something smart for CTA_F77_STDOUT

cta_file:
 * Implementation of open function
 * support for both C as FORTRAN files

cta_functions:
 * The interface that is set should be a copy
 * returning the name of the function should be a COSTA string object

cta_model:
 * The contruction of a state is very complicated and not known outside
   the model. For that reason it is possible to have a state being created
   for you for one of the 'get'-methods. We should think of someting to
   add some extra configuration like e.g. what vector representation and
   state-vector class to use.
 * The covariance matrix of the noise parameters is now represented as an
   array of state vectors. Altough very usefull for rrsqrt, it is not really
   elegant. Maybe it must be represented differently in the future.
  
cta_obsdescr:
 * In the interface of CTA_ObsDescr_Get_Properties is argument datatype. 
   It shouldn't be there because it is part of a vector.