On Sat, Apr 19, 2008 at 12:52 PM, Isidor Zeuner <sbcl@...> wrote:
> when using the random number generator initialized from a predictable
> source, the random numbers will stay predictable. As will the object
> ID numbers of created objects when executing exactly the same forms
> in the same order from the same core image. But when saving core
> images using #'save-lisp-and-die, the core images always differ even
> with only predictable computations done. Where does the differing
> information come from? Would it be difficult to create a version of
> #'save-lisp-and-die that creates predictable core images?
If your inputs are fully deterministic (same user running on the same
host, in the same directory), and there are no hidden sources of
indeterminism (like allocation patterns changing slightly when
multiple threads are running, depending on the order the OS scheduler
happens to schedule them), the only things I can think of that should
change are time-related.
If that is indeed the only source of indeterminism for you
(predictable inputs, only a single thread running), and having
bit-identical cores is important (why?), then it should not be too
hard to force the timestamps to a known values.