From J.L.Schonfelder@liverpool.ac.uk Tue Mar 28 14:28:55 1995 Received: from mailhub.liverpool.ac.uk (mail.liv.ac.uk) by dkuug.dk with SMTP id AA23280 (5.65c8/IDA-1.4.4j for ); Tue, 28 Mar 1995 15:29:27 +0200 Received: from pop.liv.ac.uk by mail.liv.ac.uk with SMTP (PP); Tue, 28 Mar 1995 14:29:04 +0100 Date: Tue, 28 Mar 1995 14:28:55 GMT From: Lawrie Schonfelder Subject: Re: Pointers to procedures To: "John.Reid" Cc: comp-fortran-90@mailbase.ac.uk, SC22WG5@dkuug.dk Message-Id: Priority: Normal Mime-Version: 1.0 Content-Type: TEXT/PLAIN; CHARSET=US-ASCII X-Charset: ASCII X-Char-Esc: 29 On Tue, 28 Mar 1995 17:56:39 +1000 John.Reid wrote: > I gave a talk on Fortran 90 here at ANU, which has provoked a finite-element > expert to ask why we do not have pointers to procedures in Fortran 90. As usual I think the call for pointers to procedures is really a request for procedure variables and I think if we are going to have mutilevel indirection such as pointers I think we should have the single level provided by variables as well or first. Procedures as currently defined are essentially constants. The symbolic name always refers to the same procedure (generics and dummy procedures aside). Dummy Procedures have limited variable capacity in that the name may refer to different actual procedures on different calls but locally the name is essentially constant (like an intent(in) variable). For many applications where a pointer to a procedure is needed a procedure variable would do. This requires a procedure declaration where the interface is specified including a name for the variable but no body. The interface does both of these but the name is currently that of an external proc or a dummy proc. I think there is more syntax and semantics required here. The semantics I think we want is to establish the notion of "type" and "value" for a procedure. The type is the characteristics established by an interface definition. The value is the body of the procedure which is executed when the procedure is invoked. Normal existing syntax establishes a fixed symbolic name for a procedure type and value, a constant procedure. We new syntax to establish a name that has a specific procedure type but which may have different associated bodies established by assignment during the execution of the program. We might then need pointers and targets of procedure types (but fairly rarely I would have thought) when we wanted this extra level of aliasing. This would make procedures into regular named entities. John's proposal so far merely creates components which are pointers to procedures which I think is too irregular. There are also a number of questions that need thought in introducing procedure variables. Are the dummy argument/keyword names declared with the type and therefore a property of the variable or are they taken from the actual value procedure? How are generic procedures/overload sets handled? How does all this interact with argument passing; getting a seamless upgrade on the current procedure argument functionality will not be trivial? Useful functionality but not easy language design. The analogue from C with simple pointers to an address which happens to be the start address of a proc is not applicable to Fortran. -- Dr.J.L.Schonfelder Director, Computing Services Dept. The University of Liverpool, UK e-mail J.L.Schonfelder@liv.ac.uk phone: +44(51)794-3716 fax: +44(51)794-3759