Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!ucsd!ucbvax!CS.WASHINGTON.EDU!wgg
From: wgg@CS.WASHINGTON.EDU
Newsgroups: comp.lang.icon
Subject: RE:  reversible assignment
Message-ID: <9005020359.AA04237@june.cs.washington.edu>
Date: 2 May 90 03:59:24 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Distribution: inet
Organization: The Internet
Lines: 47

Richard Goerwitz's answer is correct:  data backtracking is possible only
if the control backtracks to the expression involved. 

In the expressions

	z <- 1
	z = 10

each has an implicit semi-colon at the end:

	z <- 1;
	z = 10;

The first semicolon delimits the assignment from the comparison, and prevents
backtracking back into the assignment from the comparison if it fails.  
Thus the assignment expression is ``bounded'' by the semicolon, and cannot
be resumed once it has yielded a result. 

On the other hand, in the expression

	(z <- i) & (z = 10)

The assignment itself is not bounded, and it is possible to backtrack from the
comparison into the assignment, if the comparison fails. 

In most cases the semantics of traditional-appearing control structures in
Icon is to bound an expression so that it produces only one result.  This
prevents ``surprises'', and also avoids the overhead of often unneeded
backtracking.  Hence the control structures if-then and while-do bound their
control expressions (but not their bodies!).   Of course, it is easy to
phrase ``backtracking'' versions of these control structures:

	basic				backtracking
-------------------------------------------------------
if X then Y else Z			(X & Y) | Z

while X do Y				every X do Y

X;Y					X & Y

return X				suspend X

One could easily argue that I've chosen the wrong analogues.  (Suppose 
that the analogue for while-do resumes X only if Y fails, otherwise X
just starts over.  Consider, too, its behavior when Y contains a break.)

					Bill Griswold