Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!usc!apple!snorkelwacker.mit.edu!bloom-beacon!eru!hagbard!sunic!mcsun!ukc!edcastle!cs.ed.ac.uk!cs.edinburgh.ac.uk!nick
From: nick@cs.edinburgh.ac.uk (Nick Rothwell)
Newsgroups: comp.lang.functional
Subject: Re: Implementing Error Values
Keywords: pointers, papers, paraphernalia, error values, implementation
Message-ID: <2442@skye.cs.ed.ac.uk>
Date: 23 Nov 90 16:53:49 GMT
References: <86438@lll-winken.LLNL.GOV> <88@syrian.cs.utexas.edu> <2437@skye.cs.ed.ac.uk>
Sender: nnews@cs.ed.ac.uk
Reply-To: nick@lfcs.ed.ac.uk
Organization: Wavetables 'R' Us
Lines: 47

In article <2437@skye.cs.ed.ac.uk>, mikef@cs.ed.ac.uk (Mike Fourman) writes:
> >   This is pretty straightforward in Miranda.
> >
> >result ::= Val num | ERROR
> >
> [stuff deleted]
> 
> However, this solution won't allow you to propagate errors through 
> arbitrary function calls (because of the type system). In standard 
> ML there is an exception mechanism that allows errors to be raised
> and handled, and lets them carry values if you wish.

Way back when I implemented a functional language for my Ph.D., I had
a polymorphic type system (ML-like, essentially), but I had error values
as well. These carried the same types as ordinary values, but were
distinguished from them for the purposes of the primitive operations
(arithmetic and so on) and decomposition for pattern matching. Thus
(assuming modern ML syntax):

	- val x = 1/0;
	  x = <error: DIV> : int

	- val y = x + 1;
	  y = <error: ADD, DIV> : int

	- val z =
	    let val x :: y = nil in x end;
	  z = <error: MATCH> : 'a

	- fun isError <error: _> = true
	    | isError _ = false;
	  isError = fn : 'a -> bool

	- (isError z, isError 3);
	  (true, false) : bool * bool

The only problem I could see with the scheme (and didn't have time to
investigate) was that normally complete pattern matches (such as nil vs.
hd :: tl) were no longer so, and it wasn't clear that termination of
sensible functions could be guaranteed (a length-of-list function called
with an error value would probably loop indefinitely, for example...)

-- 
Nick Rothwell,	Laboratory for Foundations of Computer Science, Edinburgh.
		nick@lfcs.ed.ac.uk    <Atlantic Ocean>!mcsun!ukc!lfcs!nick
~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~ ~~
 "You ain't seen nothing yet. I can take this floor out too, no trouble."