Re: ForTran-Ada + flamette + question

[broman@noscvax.UUCP (Vincent P. Broman)]

<>

richw@ada-uts justifies the lack of procedure pointers in ada by
asserting that they can be simulated satisfactorily with an 'eval'
procedure, that dispatches on a parameter of an enumerated type by
invoking the procedure that is logically associated with the
parameter's value.

This is insufficient. The canonical example is a numerical quadrature
package which integrates user-supplied functions.  The problem is that
the dispatcher that repeatedly invokes the integrand must know the NAMES
of all potential integrands in order to invoke them.  These names are
known only at user-compile-time, not library-compile-time.
The quadrature routine should not see integrand names, it should only
specify the parameter and result types of the eventual integrands.

A generic quadrature package mispleases because separate copies of
it must be instantiated for each function to be integrated.

I have used function/procedure pointers in Algol-family languages
fairly often.  A big attraction (to me :-) of Scheme and its cousins is
the power afforded by making procedures first-class objects.  If
language size is the real concern here, I would gladly give up in
return such dubious features of ada as: default values of parameters,
the raise statement, assignments to objects of unconstrained size,
etc.


Vincent Broman, code 632,  Naval Ocean Systems Center, San Diego, CA 92152, USA
Phone: +1 619 225 2365        {seismo,caip,ihnp4,ucbvax}!\\\\\\\
Arpa: broman@bugs.nosc.mil Uucp: {floyd,moss,bang,hp-sdd,sdcsvax}!noscvax!broman