Re: Procedure types and dynamic binding

[rjh@cs.purdue.EDU (Bob Hathaway)]

>    We are essentially getting into the realm of programs which modify
>    themselves, and I tend to think that this power requires extraordinary
>    regulation due to the ease with which such a program could modify itself
>    to the point that nobody can figure out how it got there in less than
>    omega (Ackermann's function) time, much less how to fix it.
>
>    I've suggested earlier that programs should theoretically be able to
>    edit source files, cause a recompilation, dump their local knowledge
>    into a transition file, invoke the successor on the local knowledge,
>    and finally commit suicide, but that this should be done with extreme
>    caution. 
>
>    Is there any reason why we need procedural variables in addition to this?

This method is not unusual and can be accomplished by providing an operating 
system dependent package with the necessary operations.  Your suggestion 
would move this operation into the higher-order domain and could
be useful in AI and program generation applications.  Retargetable compilers 
are a case in point.

And, speaking of letting proposals stand on their own merits, could we please
put the garbage collection issue to rest.  Each party could state a final
summary of their position, and leave time for consideration before Ada 9X.

                                          Thanks,
                                              Bob Hathaway
                                              rjh@purdue.edu