[GRASS-dev] Update of r.surf.idw to FP

[Glynn Clements]

Markus Neteler wrote:

> > >> I have tried to update r.surf.idw to floating point:
> > >>
> > >> but I am not quite sure if this is efficient like this.
> > >> Currently it doesn't compile due to some "extern" magic
> > >> which I don't quite understand.
> > >> I have been inspired by r.example.
> > >>     
> > >
> > > The problem is that you have renamed some (but not all) occurrences of
> > > "cell" to "outrast", although the variable is still called "cell".
> > >
> > >   
> > (given that r.surf.idw should stay for more time)
> > 
> > I have updated the patch (find attached) to the current CVS version,
> > still crashing.
> > Advice/fix most welcome...
> 
> Still struggling with the FP patch (find attached).

1. A function cannot reference local variables of another function. 
The outrast variable needs to either be a global variable (like cell)
or passed as a parameter.

As it stands, the outrast declared in main() and the outrast
referenced by interpolate() and make_neighbors_list() are two
different variables (the first is local, the second global). Only the
former is actually getting initialised, hence the segfault.

2. Changing the type of the cell argument to row_lists() from CELL *
to unsigned char * stops the compiler complaining, but the compiler is
right to complain. If cell can refer to any of CELL, FCELL or DCELL,
any code which dereferences CELL needs three cases.

But those values are stored in the value field in the MELEMENT
structure, which is an int. So this field qwould need to be either a
double or a union of CELL, FCELL, and DCELL, and any code which
references that field would also need three cases.

Personally, I would suggest simply replacing CELL with DCELL
throughout, rather than trying to write three separate cases for
everything. If you want to be able to generate CELL output maps (it
doesn't necessarily follow that you want CELL out just because you got
CELL in), just open the output map as the appropriate type.

Most of the time, having separate CELL/FCELL/DCELL cases isn't worth
the effort. Sometimes it's worth providing special treatment for
integers, but not when you're perfoming interpolation (which naturally
produces a FP result even if the inputs are integers).

-- 
Glynn Clements <glynn at gclements.plus.com>