Re: Bug in do loop conditions?

Warwick Harvey wrote:
> 
> There appears to be a bug with regard to counting conditions in do loops.  I
> would have assumed that if I had `for(I, 1, 3)' as a condition for a do
> loop, the loop would not have iterated more than 3 times.  This is not the
> case, as shown in the example below.  The same thing happens using `count'
> instead of `for', and using a `fromto' instead of a `foreach'.


The loop terminates iff all the termination conditions are true.

The loop continues if all the termination conditions are false.

Otherwise, depending on the exact condition
	- it continues
	- it fails
but it never terminates with success.

In your example, the only other implementable behaviour is
failure after 3 instead of failure after 5 iterations.


Even though we could "fix" the for-condition (at a cost),
in general we cannot get rid of the fail/continue ambiguity
because we allow the general fromto-condition. Look at a
single fromto and the decision of whether to continue or
terminate (at the very start of the iteration):

				continue	terminate
fromto([a],[_|T],T,[])		y		n
fromto( [],[_|T],T,[])		n		y
fromto(  _,[_|T],T,[])		y		y
fromto([a], _   ,T,[])		y		n
fromto( [], _   ,T,[])	*	y		y
fromto(  _, _   ,T,[])		y		y

There are sensible fromtos that are ambiguous.

In your example, after 3 iterations, the for-condition
(corresponding to the fromto marked with *) makes a y-y
decision, the foreach makes a y-n decision, and the
result is continuation.

The other possibility would have been a n-y and a y-n
which would have led to immediate failure.


So is it worth changing the behaviour of for and count,
given that you can still get the bahaviour with fromto?

-- Joachim