[Color Computer] [Coco] Bug in FOR NEXT ?

[Arthur Flexser]

Now that I've had a chance to digest what William wrote, it makes good sense as
an explanation of the bug.  I don't have a CoCo anymore to test his example to
see if it generates an ?NF error, but I'm almost certain it would.  I tried it
under MSDOS Qbasic and it ran fine, though.  Perhaps Microsoft was sufficiently
embarassed by the Byte article to use a better algorithm for Qbasic's FOR/NEXT
processing.  It is not obvious to me what the best fix would be, though.  Best I
could come up with is this:  In processing a FOR statement, reuse the FOR frame
for the variable only if the variable is uppermost on the stack.  One would also
probably want to search the stack for a prior occurrence of the variable and
purge its frame, to prevent the memory leak William refers to that would
otherwise occur if an uncompleted loop is accessed repeatedly.  The bug is
sufficiently obscure, and the fix complicated enough, that I can sympathize with
Microsoft continuing to use its bugged algorithm though the development of Basic
for a bunch of different machines.  Problem is, though, in the relatively
unlikely event you encounter the bug accidentally (as happened to me on one
occasion), you go nuts trying to figure out why you're getting an ?NF error.

Art

> 
> On Thu, 1 Jun 2006, William Astle wrote:
> 
> > Arthur Flexser wrote:
> > > The bug occurs when you exit an initial FOR/NEXT loop without completing it
> > > (which is perfectly legal to do) and then reuse the same loop variable name in a
> > > later loop.  It is critical that the NEXT statements contain the variable name
> > > (i.e., NEXT Z, not just NEXT) for the bug to occur.  I think maybe the variable
> > > that was previously used in an inner loop must be used later in an outer loop,
> > > or possibly vice versa.  As I said, my memory is pretty vague on this, even
> > > though I did at one time trace the source of the bug in the ROM code.
> > > 
> > > Anyone want to fiddle around with coming up with some code that produces the
> > > spurious ?NF error?
> > 
> > I think I know how it occurs, based on my memory of reading the rom
> > code. It's a combination of two things, the first being the thing I
> > described in my other email (which rambled on about gosubs and
> > for/next). The second is what the code does when a for loop is started.
> > If I recall correctly, it does the following:
> > 
> > 1. look for a FOR frame on the stack using the same variable
> > 2a. if a frame was found, update the parameters and reuse the stack frame
> > 2b. if a frame was not found, create a new one with the correct
> > parameters and push it onto the stack
> > 3. go on with the execution of the program
> > 
> > Now, suppose you do the following:
> > 
> > 10 FOR X = 1 TO 10
> > 20 IF X = 4 GOTO 40
> > 30 NEXT X
> > 40 FOR Y = 1 TO 10
> > 50 FOR X = 1 TO 10
> > 60 NEXT X
> > 70 NEXT Y
> > 
> > I suspect that you would get ?NF in line 70.
> > 
> > The reason is due to the operation of NEXT which is, IIRC, approximately
> > as follows:
> > 
> > 1. search the stack for a FOR frame
> > 2. if no frame found, ?NF error
> > 3. if frame found, is it the requested variable? (always true for no
> > variable specified) If not, go back to 1
> > 4. stack is now reset to point to the FOR frame that was found
> > 5. apply the STEP to the variable and test the ending condition, etc
> > 6. if loop is done, pop the FOR frame from the stack and resume at the
> > next statement after NEXT else resume execution after the FOR statement
> > 
> > So, as an artifact of reusing the FOR frame from the previous FOR X, the
> > order of the FOR frames on the stack is backwards so that when the NEXTX
> > executes, it removes the FOR frame for FOR Y from the stack.
> > 
> > My guess is that this was a (possibly) misguided attempt to avoid a
> > memory leak when an index variable is used repeatedly without the loop
> > ever running to completion.
> > 
> > -- 
> > William Astle
> > finger lost at l-w.net for further information
> > 
> > Geek Code V3.12: GCS/M/S d- s+:+ !a C++ UL++++$ P++ L+++ !E W++ !N w---
> > !D !M PS PE V-- Y+ PGP t+@ 5++ X !R tv+@ b+++@ !DI D? G e++ h+ y?
> > 
> > -- 
> > Coco mailing list
> > Coco at maltedmedia.com
> > http://five.pairlist.net/mailman/listinfo/coco
> > 
> 
> 
> -- 
> Coco mailing list
> Coco at maltedmedia.com
> http://five.pairlist.net/mailman/listinfo/coco
>