John Bray wrote:
>
> As a user, the whole pointer system in F90 has been nasty for years. The
> different requirements for dynamic allocation, aliases and genuine pointers
> were combined in F90 in a way that still confuses me. Virtually all our
> code just needs dynamic allocation inside derived data types, and all the
> POINTER logic we were forced to introduce was unnecessary complication.
>
Particularly the undefined pointer status at initialization! At
least F95 has null(). Numerical Recipes' approach of using a saved
flag works OK, but makes implementing destructors very difficult.
> I look forward to F2000 when I can change all my POINTER attributes to the
> ALLOCATABLE ones I originally wanted, and restrict my POINTER use to the
> small sections of the code where the alias properties are useful.
I do too. A question for the F2K savants: in F2K derived types, will
the compiler replace the "allocatable" with "pointer" and use the
knowledge that "this pointer can't be aliased, so optimize away"? Or
are allocatable components going to be treated differently from
pointers altogether?
Alvaro Fernandez
Rice University
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|