Custom Query (1030 matches)

While running our code in production, we have been catching this "interesting" segfault, seemingly in the middle of garbage collection.

We strongly suspect our code was trying to print an XML message at that time using libxml2, which may or may not suggest some interaction between GC and FFI.

Xref: ITA bug 122988

(This is assuming ccl::conditional-store should work as a typical compare-and-swap, even though it's not exported.)

The RUN function below causes an assertion failure within 2000 iterations on my Linux-x86 machine. An svref place has the same issue. I did not see a problem on Darwin.

Removing the assertion (but keeping the conditional-store) results in a hang. Replacing that conditional-store (the second one) with setf also produces a hang.

(defstruct stack
  (head nil)
  (lock nil))

(defun call-with-spin-lock (stack fn)
  (loop :until (ccl::conditional-store
                (stack-lock stack) nil t))
  (unwind-protect
       (funcall fn)
    (assert (ccl::conditional-store
             (stack-lock stack) t nil))))

(defmacro with-spin-lock ((stack) &body body)
  `(call-with-spin-lock ,stack (lambda () ,@body)))

(defun push-stack (value stack)
  (with-spin-lock (stack)
    (push value (stack-head stack))))

(defun pop-stack (stack)
  (with-spin-lock (stack)
    (pop (stack-head stack))))

(defun test (thread-count)
  (let ((stack (make-stack)))
    (loop :repeat thread-count
          :do (ccl:process-run-function "test" #'push-stack t stack))
    (loop :repeat thread-count
          :do (loop :until (pop-stack stack)))))

(defun run ()
  (loop
     (test 1)
     (format t ".")))

The currently last COND clause in ACODE-OPTIMIZE-ASH tries to inline some calls to ASH based on information about the arguments' types; it erroneously assumes that if the types are subtypes of FIXNUM, they must be represented as NUMERIC-CTYPEs in CCL. Types specified as (e.g.) (MEMBER 23 5) are subtypes of FIXNUM that aren't represented as NUMERIC-CTYPEs.

This clause in ACODE-OPTIMIZE-ASH shouldn't make this assumption.

see ​http://clozure.com/pipermail/openmcl-devel/2012-November/013919.html