#76 CLIPS crashes when global variable modified

[xyando]

Hello,

The following commands makes CLIPS to crash.

$ ./clips -f globalbind.bat
CLIPS (6.4.2 5/7/24)
CLIPS> (defglobal ?foo = (create$ 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0))
CLIPS> (defglobal ?bar = (create$ 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0))
CLIPS>
(deffunction test (?foo ?bar)
(loop-for-count 10 do
(bind ?foo ?foo)
(bind ?bar ?bar)
)
(printout t "FOO " ?foo crlf)
(printout t "BAR " ?bar crlf)
)
CLIPS>
(test ?foo ?bar)

[PRNTUTIL3]
*** CLIPS SYSTEM ERROR ***
ID = EVALUATN7
CLIPS data structures are in an inconsistent or corrupted state.
This error may have occurred from errors in user defined code.

---

The patch attached seems to fix this problem.
Thanks.

[xyando]

The posted comment is erroneous because the asterisks are missing.
Please refer the attached file.

[xyando]

The revised patch is attached because the elder patch fails CLIPS test_suite.

[Gary Riley]

A fix has been checked into the svn repository. The issue occurs when setting the current value of a defglobal to its current value. This can cause the value to be prematurely released. The bug can occur outside the body of a deffunction, so the proposed patch won't fix the issue for all cases.

         CLIPS (6.4.1 4/8/23)
CLIPS> (defglobal ?*foo*  = (create$  0.0))
CLIPS> 
(loop-for-count 1 do
   (bind ?*foo* ?*foo*))
[ARGACCES2] Function 'loop-for-count' expected argument #3 to be of type integer, float, symbol, string, instance name, instance address, fact address, external address, or multifield.
FALSE
CLIPS> 

[xyando]

Thanks Gary,
but I think an additional modification is necessary(see attached patch)
because the following program results an endless looping.

(defglobal ?*x* = (create$ 1 2 3))
(deffunction test()
  (while TRUE do
    (bind ?*x* (rest$ ?*x*))
    (printout t ?*x* crlf)
    (if (= (length$ ?*x*) 0) then
      (break)
    )
  )
)

[Gary Riley]

Nice catch. I checked in an update.

[xyando]

Thanks Gary,
The endless looping is resolved. But the globalbind.bat fails again.

I think the equality checking once you have added is still necessary
(but after NormalizeMultifield(), but before displaying watch info)
to avoid premature release.

[Gary Riley]

I checked in a fix. I forgot to test the original issue. There are multiple ways to generate a bug with this issue and just modifying QSetDefglobalValue won't fix all of them. Here's an example which globalbind-3.patch will still generate an error:

(defglobal ?*foo* = (create$ 0.0))

(deffunction test (?foo)
   (bind ?*foo* ?foo)
   (println ?foo)
   (bind ?*foo* (create$ 3.0))
   (println ?foo))
(test ?*foo*)

What I did to resolve the issue is modify the routines which return the value of the global variable to return a copy of the multifield value instead of the value stored directly in the global. This preserves the multifield until garbage collection kicks in at the appropriate time.

[xyando]

It all works fine. Thanks.