[Openmcl-devel] Recursive tracing/advising?

[Gary Byers]

When a (globally named) function FOO is defined with DEFUN, the compiler's
allowed to assume that functional references to that function name
refer to the function being defined (unless they're lexically shadowed);
it can therefore skip the implicit SYMBOL-FUNCTION ("call whatever is
in the function cell of FOO") on a self-call (a call to FOO from within
FOO.)  This saves an instruction or two on those calls, but (since
TRACE works by changing what SYMBOL-FUNCTION returns) those inlined
self-calls can't be traced.

The compiler can't do this (can't even assume that something defined
by DEFUN won't be redefined later) if the function name is declared
NOTINLINE at the point of the self call:

(defun fact (n)         ; other traditional definition
   (declare (NOTINILINE FACT)) ; and the compiler can't ignore NOTINLINE
   (if (zerop n) 1 (* n (fact (1- n)))))

There's a way to say "as a matter of policy, I'd rather have the
ability to trace functions which call themselves and are defined with
DEFUN and don't care about saving a few cycles per self-call"; I'm
travelling at the moment and don't have a good enough net connection
to explain that, but if you're interested I'll try to explain it when
I get back.



On Tue, 24 Jul 2007, Jared C. Davis wrote:

> Hi,
>
> Is there a way to get trace to show me recursive calls in addition to
> the outermost call?  For example, if I try to trace fact as below, it
> only prints the call of (FACT 10 1):
>
>    ? (defun fact (x acc)
>        (if (= x 0)
>            acc
>          (fact (- x 1) (* x acc))))
>    FACT
>
>    ? (trace fact)
>    NIL
>
>    ? (fact 10 1)
>    0> Calling (FACT 10 1)
>    <0 FACT returned 3628800
>
> I'd like to be able to see the call of (FACT 10 1), (FACT 9 10), (FACT
> 8 90), etc., all the way down.
>
> Right now the only way I know how to do this is to change the
> definition of fact so it calls some new function, say fact-note, upon
> each iteration with the relevant arguments.  I can then trace
> fact-note instead of fact, to see all the calls.  But I don't like
> this, because it makes my definition more complicated to support
> tracing, and the function I'm actually interested in tracing is
> already very complicated.
>
> I've tried using both trace and CCL:advise with :when :after, but that
> seems to give me the same trouble.
>
> Thanks!
>   Jared
>
> -- 
> Jared C. Davis <jared at cs.utexas.edu>
> 3600 Greystone Drive #604
> Austin, TX 78731
> http://www.cs.utexas.edu/users/jared/
> _______________________________________________
> Openmcl-devel mailing list
> Openmcl-devel at clozure.com
> http://clozure.com/mailman/listinfo/openmcl-devel
>
>