Re: [Chicken-users] Debuggery

[felix]

Peter Keller wrote:
> On Wed, Mar 17, 2004 at 09:12:52PM +0100, felix wrote:
>
> > Agreed. The problem is that there is often very little information available
> > at runtime, partly because I wanted to keep too much overhead out of 
> > compiled
> > code and partly because I simply didn't show much forethought.
> > I'm open to suggestions, of course.
>
>
> I've spoken quite a long time with Felix about fixing Chicken to allow
> debugging of the native binaries at runtime by redoing some of the
> code translation a little bit to create explicit stack frames that
> contain debugging information in them. But I unfortunately don't have
> much time to invest in it currently (and for the next several months
> at least). If someone would like to help me finish the design for it
> so it can be presented to felix for his final judgement on the method,
> I would be very appreciative.
>
> I don't have much formally written up, but I can dig up my conversations
> with Felix and put something together if someone is seriously willing
> to take a look at it.

[Oops, I just sent my reply to Michele, mentioning your ideas]

Hi, Peter!

A full native code debugger would be HEAVEN. No doubt about it.
But changing the compiler and runtime-system so completely is currently
not an option, I'm afraid. The very fact that we can use standard C
call-frames makes the Cheney-on-the-MTA approach so elegant (and I
believe efficient as well).

Perhaps an extra argument per call that contains the debug-info?
Or a global variable? Currently we have C_trace(), which could perhaps
be extended to add extra information to the global trace-ring...

Portability is an issue as well. To seamlessly debug Scheme and C, one
probably has to write something on top of gdb, or even write a low-level
debugger from scratch (shudder!). I have absolutely *no* idea how such
stuff is done, and (naturally) it wouldn't work on Windoze (which is not
a big issue, but unfortunate).


cheers,
felix