Re: more srfi-12 rationale?

[Richard Kelsey <kelsey@xxxxxxxxxxxxxxxxxxx>]

   Date: Sun, 17 Oct 1999 09:54:20 -0400
   From: Dave Mason <dmason@xxxxxxxxxxxxxxx>

   I'm not worried about explicit aborts in my program - I can do as you
   say.  My concern is built-in low-level aborts.  Like (/ x 0).  If
   these dynamically create ``condition'' values, it will be expensive.

What need is there for (/ x 0) to be cheap?  And consing up a
simple exception object shouldn't take a much time or a lot of
memory.

There are two different usages envisaged for exceptions, based on
two different models:

 Machine-style exceptions:
   - used to extend the underlying Scheme implementation
   - handlers are installed rarely and have global extent
   - exceptions are frequent
   - restarts are common

 Programming-language-style exceptions:
   - used to return abnormal results across module boundaries
   - handlers are installed frequently and have dynamic extent
   - exceptions are rare
   - restarts are uncommon

SRFI-12, with easy restarts and dynamic extent, is a little of
each.  It would be great if we could use the same mechanism for
both, but the different requirements make it hard to satisfy
everyone.

   Without restarts, WITH-EXCEPTION-HANDLER doesn't have to be exposed in
   the programming interface, and I see implementation approaches that
   could make HANDLE-EXCEPTIONS much lower cost.  With restarts, you
   essentially need the dynamic-wind, call/cc pair to do
   HANDLE-EXCEPTIONS.

HANDLE-EXCEPTIONS needs the call/cc because the handler
it creates is called with HANDLE-EXCEPTIONS's continuation.
That continuation needs to be saved somehow.  Both
HANDLE-EXCEPTIONS and WITH-EXCEPTION-HANDLER need DYNAMIC-WIND
because that is the only portable way to give handlers dynamic
extent.

Both HANDLE-EXCEPTIONS and WITH-EXCEPTION-HANDLER can be done
cheaply if you are allowed to fiddle with the underlying
implementation.  For that matter, if continuations are already
in the heap, as in SMLNJ, and you have a reasonable implementation
of DYNAMIC-WIND, the implementation given in SRFI-12 will be fairly
zippy.
                                -Richard Kelsey