🚀 go-pugleaf

re: You can do an optimization, but don't you have to be able to
    de-optimize if someone later defines an applicable method on
    SLOT-VALUE-USING-CLASS?

The beauty of Lisp is in "lazy definition" -- that is, you can always
patch things up on the fly by redefining (or, defining for the first
time) even after some optimization has been put into place [perhaps the 
EXPORTations of the package system is one of the very few places where 
this idea is out of place; DEFSTRUCT redefinitions after compilation are 
also questionable.]  For example, "cacheing" sorts of optimizations
merely have to hook into the definer, to clear the cache at definition
time.  In the spirit of Lisp, then, I would say that the answer to your 
question should be "No, you should have to back out of the optimization 
at that time."


It ought to be possible to retract the accessor optimizations whenever
an applicable SLOT-VALUE-USING-CLASS method is added, without an overly
complicated implementation.  I know that this doesn't work in Lucid's 4.0
release, but hey not much of the formal MOP was promised for that release
anyway.  Stay tuned.



-- JonL --