On Jun 3, 2009, at 3:39 PM, Michele Simionato wrote:
It only uses one ikarus-specific feature (call/cf
which gives you an eq?-comparable pointer onto the stack). It
also has a simple interface between the compiler/interpreter
and the debugger, e.g., rewrite all
(proc args ...)
=>
(debug-call '<source-info> proc args ...)
and you get debuggable/traceable code.
But how does it work with TCO? In debug mode TCO
is disabled or not?
Of course not. If you disable that, most programs won't run at all.
(and that's why it should never be called "optimization")
The real details of how it works is in the code, since I didn't write
any of the details on paper (which I should, once I finish the paper
at hand).
In short, the debugger is like the tracer, it doesn't care (or need
to know off-hand) if a call is really a tail call or not. It can
figure that out dynamically simply by observing that:
"one call is in tail position with respect to another
if both calls have the same continuation"
So, the debugger uses call/cf (short for call-with-continuation-frame)
which is like the "raw" equivalent of call/cc. The debugger doesn't
need to do anything with the actual frame other than comparing it with
the last captured frame: if they're eq?, we're in tail call, if not,
then it's a nontail call. The debugger then treats the two cases
accordingly (details omitted).
[you can ask yourself why I choose to do this dynamically instead of
determining tail/nontail calls statically at the source level. It's
a good puzzler since that's how I thought it should work initially]
Aziz,,,
