[e-lang] Non-transparent memoization: is this safe?

[Kevin Reid]

Context:

E-on-CL includes a memoizer which takes a DeepFrozen object and  
returns a DeepFrozen result which memoizes all calls in which the  
arguments are DeepFrozen and the result is DeepPassByCopy.

http://www.eros-os.org/pipermail/e-lang/2006-May/011207.html shows  
how to extend the memoizer to results that are DeepFrozen and not  
different after each call.


Question:

Is it OK to consider safe a memoizer which instead memoizes results  
if they are just DeepFrozen?

This would be non-transparent in that it behaves differently from the  
original reference, but it would not be unsafe, in that it is  
equivalent to a DeepFrozen object which has references to the results  
of every call that will ever be made to it.

In other words, this would allow selective lazy evaluation.

I believe <import> already provides a limited form of this, in that  
it caches any result-of-an-emaker that is DeepFrozen. The memoizer  
would extend this by allowing the specification of arbitrary  
functions to be memoized, rather than the static set of all emakers.


Reasons why adding this would be useful:

   - It allows additional memoization for efficiency purposes,  
because it is simple to make an object DeepFrozen and complex to make  
it Selfless; consider quasi-parsers:
     memoize(
       def foo__quasiParser implements DeepFrozen {
         to valueMaker(template) :any {
           return def vm implements DeepFrozen { to substitute ... }
         }
       }
     )
     I believe E-on-CL would benefit greatly from this.

   - It makes <import> less magical, by allowing unprivileged  
construction of more of its behavior.

   - It may help fix an unrelated problem with <import> which I will  
post later.

-- 
Kevin Reid                            <http://homepage.mac.com/kpreid/>