🚀 go-pugleaf

In article <id.KTGS.0N5@ferranti.com> peter@ferranti.com (Peter da Silva) writes:

[Elided]

| Basically, imagine integrating setjmp/longjmp with the language to a greater
| extent. ...

[More elided]

| But putting it directly in the language, you allow for more optimizations.
| By replacing goto, you clean up a messy part of the language without losing
| any general capability (switches with good optimization and this context
| can replace every use of goto).

[Stll more elided]

| Yes, you can build subroutine libraries that do this, but they're inherently
| non-portable. This sort of thing should be part of the language, and is much
| more important to me than expanding the type structure with object-oriented
| stuff.

What you describe was generalised by Peter Landin in '65, who called
it the J operator. It has been adopted, in particular, by languages
like Scheme and ML which use it to define powerful control structures.
(The FP/Lisp community calls it ``call with current continuation'' or
call/cc).  For examples of the usages that you outlined, see the
Scheme stream-based IO or the Cornell work on Concurrent ML.

Cheers,
Dipankar Gupta
Hewlett-Packard Laboratories
Bristol UK

PS: I don't think that ML has continuations as a part of the Standard,
but the New Jersey implementation provides call/cc as a first-class
language element.

References:
[1] Landin, P. A correspondence between Algol 60 and Church's lambda
notation: Part 1 [C. ACM 8(2) pp89-101, Feb. 1965]

[2] Clinger, W and Rees, J (eds). Revised^4 Report on the Algorithmic
Language Scheme.
--
--