Re: [Gcl-devel] Random tester status

[Paul F. Dietz]

Camm Maguire wrote:

> Great Paul!  Thanks *very* much for your thorough testing work.  If
> possible, I'd really like to get the success rates even or higher than
> the best you've seen.  If it becomes burdensome and more CPU is
> needed, please let me know and also how you invoke the tests.

At this point when the failure rate drops this low, it's time to add
more things to the random generator.

I ran it through another 80,000 or so iterations, with no failures,
so I did that, adding FUNCALL and APPLY forms to FLET/LABELS functions.
This brought out this failure (added to misc.lsp):

;;; gcl (9 Nov 2003) bug
;;; Error in FUNCALL [or a callee]: Caught fatal error [memory may be damaged]

(deftest misc.160
  (funcall
   (compile nil
            '(lambda (c)
               (declare (notinline + funcall))
               (+ (labels ((%f1 () -14)) (funcall #'%f1))
                  (flet ((%f2 () (floor c))) (funcall #'%f2)))))
   0)
  -14)

Sooner or later I want to completely rewrite the random tester to be more object
oriented, and so the pruner acts on an intermediate representation rather
than on the lisp forms themselves.  But I may wait on this until the compliance
test proper is more complete.

I'll write up some instructions for running the generator in the near future,
and add a makefile entry.

        Paul