[ clisp-Feature Requests-1808029 ] better trace (with call counters)

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Feature Requests item #1808029, was opened at 2007-10-05 07:42
Message generated for change (Comment added) made by sds
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=351355&aid=1808029&group_id=1355

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Extend ANSI CL
Group: None
Status: Open
Resolution: None
Priority: 5
Private: No
Submitted By: Basile STARYNKEVITCH (bstarynk)
Assigned to: Nobody/Anonymous (nobody)
Summary: better trace (with call counters)

Initial Comment:
The details of the (trace ...) function behavior is left unspecified by the CL standard.

I suggest that each traced function would emit, in addition to what is actually printed today, a unique call counter.

As a concrete example, currently tracing the famous fact function gives

[5]> (fact 3)
1. Trace: (FACT '3)
2. Trace: (FACT '2)
3. Trace: (FACT '1)
4. Trace: (FACT '0)
4. Trace: FACT ==> 1
3. Trace: FACT ==> 1
2. Trace: FACT ==> 2
1. Trace: FACT ==> 6

I am suggesting that it becomes

[5]> (fact 3)
1. Trace #1: (FACT '3)
2. Trace #2: (FACT '2)
3. Trace #3: (FACT '1)
4. Trace #4: (FACT '0)
4. Trace #4: FACT ==> 1
3. Trace #3: FACT ==> 1
2. Trace #2: FACT ==> 2
1. Trace #1: FACT ==> 6

The rationale is that in real debugging sessions, finding quickly (e.g. with emacs Ctrl-R) the beginning trace matching the ending trace would be much easier.

----------------------------------------------------------------------

>Comment By: Sam Steingold (sds)
Date: 2007-10-08 16:38

Message:
Logged In: YES 
user_id=5735
Originator: NO

please try the appended patch.
File Added: trace-1808029-call-count.diff

----------------------------------------------------------------------

Comment By: Basile STARYNKEVITCH (bstarynk)
Date: 2007-10-08 13:20

Message:
Logged In: YES 
user_id=505600
Originator: YES

SDS suggestion of using :pre and :post to trace does not seems to work
easily, since I want to print the call counter (i.e. its value on function
entry) at return (when the function has been recursively called many times
so the counter has increased). In other words, the trace macro has to
expand each traced function to something like
   (let ( (local-trace-counter (incf global-trace-counter)) )
     ;;; the entering trace message using local-trace-counter
     ;;; the function to be traced
     ;;; the exiting trace message using local-trace-counter
  )

----------------------------------------------------------------------

Comment By: Sam Steingold (sds)
Date: 2007-10-08 12:55

Message:
Logged In: YES 
user_id=5735
Originator: NO

you already can accomplish something like that using the trace arguments
:pre & :post, see
http://clisp.cons.org/impnotes/environment-dict.html#trace

----------------------------------------------------------------------

Comment By: Basile STARYNKEVITCH (bstarynk)
Date: 2007-10-08 12:38

Message:
Logged In: YES 
user_id=505600
Originator: YES

Sorry for putting a simplisitc example. I really meant a call counter, so
that in the same process, if later on, after having traced (fact 3) like
above, (fact 2) is called again, it would give
(fact 2)
1. Trace #5: (FACT '2)
2. Trace #6: (FACT '1)
3. Trace #7: (FACT '0)
3. Trace #7: FACT ==> 1
2. Trace #6: FACT ==> 2
1. Trace #5: FACT ==> 6

I'm actually coding a tiny Lisp to C compiler which I sort of bootstrap
with Clisp, so that the above feature would be helpful. See my GCC Summit
2007 paper (on my home page http://starynkevitch.net/Basile/ )

Thanks for reading -- Basile Starynkevitch

----------------------------------------------------------------------

Comment By: Sam Steingold (sds)
Date: 2007-10-08 10:21

Message:
Logged In: YES 
user_id=5735
Originator: NO

well, strictly speaking the number in the first column is the stack depth,
not call counter.

----------------------------------------------------------------------

Comment By: szergling (szergling)
Date: 2007-10-06 04:52

Message:
Logged In: YES 
user_id=1630354
Originator: NO

There's numbers along the first column already. For what you want, won't
(i)search-forward-regexp work? You may for example search for C-M-s ^4

I think some Lisps, or maybe even clisp2-41? (can't check at the moment)
indent the trace output according to stack depth, that could work...

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=351355&aid=1808029&group_id=1355