I've never previously heard of a "loop and a half", so did some digging.
Eric S Roberts, you picked an oft-referenced person, Frank :-)
Here's an example of the discussion (not from Roberts, but referencing him):
"The problem with using goto isn't the keyword itself-rather, it's the use of
goto in inappropriate places. The goto statement can be a useful tool in
structuring program flow and can be used to write more expressive code than
that resulting from other branching and iteration mechanisms. One such example
isthe "loop-and-a-half" problem, as coined by Dijkstra. Here is the traditional
flow of the loop-and-a-half problem in pseudocode: -
loop
read in a value
if value == sentinel then exit
process the value
end loop
The exit from the loop is accomplished only by the execution of the exit
statement in the middle of the loop. This loop/exit/end loop cycle, however,
can be quite disturbing to some people who, acting on the notion that all uses
of goto are evil, would write this code as follows: -
read in a value
while value != sentinel
process the value
read in a value
end while
Unfortunately, as Eric S. Roberts of Stanford University points out, this
second approach has two major drawbacks. First, it requires the duplication of
the statement(s) required to read in a value. Any time you duplicate code, this
results in an obvious maintenance problem in that any change to one statement
must be made to the other. The second problem is more subtle and probably more
damning. The key to writing solid code that's easy to understand, and therefore
maintain, is to write code that reads in a natural manner. In any noncode
description of what this code is attempting to do, one would describe the
solution as follows: First, I need to read in a value. If that value is a
sentinel, I stop. If not, I process that value and continue to the next value.
Therefore, it's the code omitting the exit statement that's actually
counterintuitive because that approach reverses the natural way of thinking
about the problem. Now, let's look at some situations in which a goto statement
can result in the best way to structure control flow."
Breezing over the "evil" (straw man technique, that was) we get to two major
(alleged by Roberts) drawbacks: "it requires the duplication of the
statement(s) required to read in a value". No, it doesn't. Unless you are only
interested in "snippets" rather than programs/systems.
Secondly, and that author was saving the best, more damning, for second, "The
key to writing solid code that's easy to understand, and therefore maintain, is
to write code that reads in a natural manner." Oh. Ah, they develop:
"In any noncode description of what this code is attempting to do, one would
describe the solution as follows: First, I need to read in a value. If that
value is a sentinel, I stop. If not, I process that value and continue to the
next value."
And continues "Therefore, it's the code omitting the exit statement that's
actually counterintuitive because that approach reverses the natural way of
thinking about the problem."
Mmmm... snippets.
I'm a Mail Opening Operative. I arrive at my desk, there's a pile of mail, I
open a letter, bin the envelope (it is against policy, but I save any
nice-looking stamps for my cousin, who I think should collect stamps), see who
it is addressed to (or if it is "advertising", which Mr Johnson deals with -
I'm not allowed to throw any actual mail away) and then put it in the correct
pigeon-hole. Then I pick up the next letter, and repeat the process. When there
is no more mail, I count each bundle, put elastic bands around each bundle,
and pass them to the delivery boy (if he's bothered to turn up on time, he's
always with the girls in the punch room). I fill out a sheet of statistics,
stamp it (Mr Johnson is terribly upset if I forget that). Lately, it was my own
idea, and Mr Johnson was *very* pleased at my initiative, I have some
pre-written pieces of paper to make it extremely clear who each packet is for.
I'm sure *that boy* doesn't even appreciate it. Then it is on to internal mail,
until the second post arrives in the early afternoon...
Now, one day, I turn up - *AND THERE ARE NO LETTERS WAITING*. I'm all of a
flummox, and dial Mr Johnson immediately. He never likes being disturbed this
early, but this was really *IMPORTANT*.
<shivers briefly and artistically> That was character-acting, that was. Or
character writing.
Now, which of the above two "snippets" is reflective of Mail Opening? The one
where you actually start the work of the loop, and exit the loop in a
completely normal way, or the one where you do a bit of "priming", and realise
there is nothing to process, so Mr Johnson needs to know?
Snippets.
I found this (different author, also refers to Roberts), as well, I suppose a
variant of a loop-and-a-half to process "an array, which shows how it would be
written with a GOTO:
i = low_value
loop:
if i > high_value
goto endloop
statements
i = i + increment
goto loop
endloop:
Really? Let's think. I put something in "i", and then enter the loop and
immediately... test "i"? Didn't I just put it there?
i = low_value
loop:
statements
i = i + increment
if i <= high_value
goto loop
Is that so difficult? Now, it may be the case that when entering the loop,
high_value isn't really so high, and happens to be smaller than low_value (in
which case they are poor names):
if low_value > high_value
goto stuff_the_loop
i = low_value
loop:
statements
i = i + increment
if i <= high_value
goto loop
stuff_the_loop:
That's the "patch as you go along" approach to the use of GOTO ("I only use it
when it makes the code easier" usually accompanies it).
if low_value <= high_value
execute procedure loop_it
else
do something else for the unexpected
end-if
procedure loop_it
i = low_value
loop:
statements
i = i + increment
if i <= high_value
goto loop
end procedure
(In this language, of course, procedures can only be entered by "execute",
there is no "fall through").
Now, note that "i" is busted, and if "i" is valid outside the procedure (like
in COBOL) you may not want it exactly like that.
procedure loop_it
i = low_value
loop:
statements
if i < high_value
i = i + increment
goto loop
end-if
end procedure
I'm not suggesting the use of GO TOs, I'm just sayin' that when we did use
them, they didn't have to be used how the critics of GO TOs thought (imagined,
fantasised) we used them, because they, apparently, only taught and dealt with
snippets. Snippets.
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
