- Several comments on the design of the fitting part:
- Derived typed are defined to stay general, in particular no
array dimension is explicitly specified. This implied the use of
POINTERS.
- Everything can not be passed in the subroutine calling
sequences. Indeed, ALL minimization (and thus fitting) algorithms
ask for a function to be minimized which have a pre-defined calling
sequence. This calling sequence varies with the used minimization
methods and libraries. However, those calling sequence inevitably miss
some important parameters (as an example, they never pass function
parameters that the user want to stay fixed). Properties of the fitted
function (and also sometimes grid of measured points) must be passed by
another mechanism. In some F77 parts of GILDAS (e.g. UV_FIT), this
problem is solved by INCLUDE files and COMMON. Here, the F90 “use
association” mechanism is used (i.e. the variable to be shared is
define in a module that is called in all the needed subroutines).
- There are some duplication between the fitted function
calling sequence and the “association module”. This avoids
particular cases for every different minimization methods and
libraries.
- The plotting part is done through SIC procedures. This is very
flexible but this implies the definition of SIC structures associated to
the high levels TELCAL derived types (e.g. POINT_CROSS).
- The TELCAL demonstration could be the basis of a TELCAL test suit.
