[slime-devel] Package profiling in SBCL

[Constantine Vetoshev]

I noticed PROFILE-PACKAGE isn't implemented for SBCL. SBCL's native
SB-PROFILE:PROFILE does support package-level profiling given a string
argument. It has an interface somewhat like the one SLIME expects: it
unprofiles and reprofiles already-profiled functions, although it does
not support CALLERS-P or METHODS; I suppose that's why it isn't in the
SBCL backend now. Even with this limitation, I find these patches
useful:

Index: slime.el
===================================================================
RCS file: /project/slime/cvsroot/slime/slime.el,v
retrieving revision 1.308
diff -r1.308 slime.el
4148,4149c4148,4151
<                      (y-or-n-p "Record the most common callers? ")
<                      (y-or-n-p "Profile methods? ")))
---
>                      (unless (string-match "sbcl" inferior-lisp-program)
>                        (y-or-n-p "Record the most common callers? "))
>                      (unless (string-match "sbcl" inferior-lisp-program)
>                        (y-or-n-p "Profile methods? "))))
Index: swank-sbcl.lisp
===================================================================
RCS file: /project/slime/cvsroot/slime/swank-sbcl.lisp,v
retrieving revision 1.85
diff -r1.85 swank-sbcl.lisp
534a535,541
> ;;; FIXME: This does not meaningfully support CALLERS-P or METHODS.
> (defimplementation profile-package (package callers-p methods)
>   (when package
>     (let ((package-name (string-upcase package)))
>       (eval `(sb-profile:profile ,package-name))))
>   (format nil "Profiling all functions in package ~A." package))
> 

Out of curiosity, are there any plans to use the included
metering.lisp facility for all Lisp implementations, and not just
CLISP? The current profiling hooks in SLIME expect a CMUCL-like
interface, and supporting, e.g., the native profiler in Allegro CL
would require either changing the front end (functions in slime.el
need to ask different questions and provide different report options)
or resorting to low-level hackery to make the other Lisps' profilers
look like CMUCL's.

-- 
Regards,
Constantine Vetoshev