Re: Another approach to agent architecture [ was Re: agents .. ]

Nathaniel Borenstein writes:
 > 
 > This strikes me as the $64M question.  Language translators are a
 > difficult thing to get right.  I have no idea, a priori, whether it is
 > easier to translate Python and Scheme into Tcl, or Python and Tcl into
 > Scheme, or Scheme and Tcl into Python.  I don't see any path to a quick
 > answer, and that concerns me.  -- Nathaniel

I think Java is evidence that the safe virtual machine approach is
viable. (scheme48 is more evidence, though they haven't done as much
engineering on the "safety" features). Language translators are a
tricky art, to be sure, but compilers are old hat.

I hope we see (maybe even write) Java bytecode compilers for Python,
scheme, Modula-3, and SmallTalk very soon. (I'd like to see designs
for dealing with continuations in the Java VM... I believe it's
possible, but I wonder if it's feasible). The possibility for code
reuse is staggering.

It's not clear that the library interfaces will be reusable across
these languages, but in my opinion, it looks promising.

I'm keen on the Java solution to multiple inheritance and binding time
of methods. Single inheritance, call by signature, and interfaces
looks very powerful. I wonder if the Modula-3 and ILU folks would be
interesting in supporting this model.

Dan

Received on Friday, 7 April 1995 16:49:02 UTC