 | Robert A Duff wrote:
> Bernd Paysan writes:
[...]
> > I'm pretty sure that function pointers do not exist at all in the
original
> > Pascal...
>
> True, but the original Pascal allows one to pass a procedure or
function
> as a parameter, which allows some of the power supported by both
GCC-C
> and Ada 2005. It's not called a pointer in Pascal, but of course the
> implementation involves pointers in some way. What you can't do in
> Pascal is store a procedure (or pointer to procedure) in some data
> structure, and call it later. That is, when you pass a procedure as
a
> parameter, the only things the callee can do is call it and pass it
> along to another procedure -- you can't assign it into a variable,
> or return it to a more global (longer lived) place.
And this is an important restriction (for implementation efficiency),
because it guarantees that the lifetime of a closure is _never_ outside
the lifetime of the stack frame for the function call environment it
references. It would be OK to also permit assigning them to local
variables. The important part is that no closures escape upward in the
call chain. And this lets you keep everything on the stack.
[...]
> >... He did what he did, with his Lisp
> > background the result is easy to understand.
>
> Well, if he were mimicing Lisp, he would have to allow nested
functions
> to outlive their container, which is a whole 'nother thing!
Closures which can outlive the stack frame in which they were created
require garbage collection / reference counting. They have to be
allocated on the heap. And if you really need that many closures
floating around, perhaps C is not the language of choice here. When
you're programming in a really functional style, a LISP/Scheme/ML/etc..
compiler is going to be better at dealing with those control flow
patterns.
--Eric
|
|
