Incorrect assignment with declared types inside a loop

[Patrick Poitras]

We have this function that is finding the min and max data points inside a structure.

I've stripped it out to a minimal example on sbcl 2.3.2. The same behavior is observed on a non-minimal example.

==== The example====

(defun minmax-function ()
  (let ((min 13d0)
        (max -13d0))
    (LOOP for iter from 0 below 1
          for value of-type double-float = 3d0
          do (LET ((G641 VALUE))
               (WHEN (OR (NOT min)
                         (< G641 min))
                 (SETF min G641)))

             (let ((G642 VALUE))
               (when (or (not max)
                         (> G642 max))
                 (setf max G642))))
    (values min max)))

We would expect min and max to both be 3. Instead, we get min = 3 and max = 13.
If we compile with debug, we get 3 and 3.

I have also verified this example in SBCL 2.2.10, and it does not pose this issue.

I've experimented a bit to find some variants and see if I could cut down the minimal example.

== Variants that are broken on both debug and regular compilation ==

(defun minmax-function ()
  (let ((min 13d0)
        (max -13d0))
    (LOOP for iter from 0 below 1
          for value = 3d0
          do (when (or (not min)
                       (< value min))
               (setf min value))

             (when (or (not max)
                       (> value max))
                 (setf max value)))
    (values min max)))

== Variants that work on both debug and regular compilation ==

(defun minmax-function ()
  (let ((max -13d0))
    (LOOP for iter from 0 below 1
          for value of-type double-float = 3d0
          do
             (let ((G642 VALUE))
               (when (or (not max)
                         (> G642 max))
                 (setf max G642))))
    (values max)))

== Other things ==

Linux <SESSION NAME> 5.15.0-60-generic #66-Ubuntu SMP Fri Jan 20 14:29:49 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux

[xizang (xizang)]

Before modifying the variable, print it first, and the result is correct again, why is this?

(defun minmax-function ()
  (let ((min 13d0)
        (max -13d0))
    (LOOP for iter from 0 below 1
          for value of-type double-float = 3d0
          do (LET ((G641 VALUE))
               (WHEN (OR (NOT min)
                         (< G641 min))
  (print min)
                (SETF min G641)))

             (let ((G642 VALUE))
               (when (or (not max)
                         (> G642 max))
   (print max)
                 (setf max G642))))
    (values min max)))

>
13.0
-13.0
3.0
3.0