Anderson@Cmu-Cs-C@sri-unix (12/01/82)
From: David B Anderson <Anderson@Cmu-Cs-C> Look for the paper "Smalltalk-80 on the Intel 432: A Feasibility Study," by Almes, Borning, and Messinger of the University of Washington in the forthcoming book "Smalltalk-80: Implementation Considerations," edited by Glenn Krasner at PARC. They conclude that by re-microprogramming the 432 an interesting (and perhaps reasonable) implementation is possible - but they do point out some problems: (1) object storage overhead resulting from associating Smalltalk objects with 432 objects, (2) insufficient microcode space, and (3) garbage collector overload. David (Anderson @ Cmu-CS-C)
txr.usc-cse@Udel-Relay@sri-unix (12/02/82)
I have heard of Glen's book, interesting to hear some more details. It seems to me that none of what was said (in the reported feasability study) changes anything I said about the 432. The 432, as it is, is not a good architecture for doing Smalltalk. Even if it were re-microcoded (which basically re-defines the entire machine architecture) it would still have problems due to the way that the 432 realizes "objects" as compared to Smalltalk objects. We're talking performance here, but eventually performance has to be an issue -- after all, all these things are Turing computable and so can be simulated (at some performance level) by any machine. From what you told me, the 432's view of what an object is is so deeply ingrained that even re-microprogramming it has substantial problems. The 432: Data Encapsulation? You bet. Object oriented? No. Tim
txr.usc-cse@Udel-Relay@sri-unix (12/02/82)
(I apologize if some of you get this twice, the original To: address was unknown to the mail system. If Anderson@CMU-CS-C isn't getting this message (i.e., isn't on info-micro, etc.) you might send him a copy, as well as let him know that he may be putting out bogus return addresses.) I have heard of Glen's book, interesting to hear some more details. It seems to me that none of what was said (in the reported feasability study) changes anything I said about the 432. The 432, as it is, is not a good architecture for doing Smalltalk. Even if it were re-microcoded (which basically re-defines the entire machine architecture) it would still have problems due to the way that the 432 realizes "objects" as compared to Smalltalk objects. We're talking performance here, but eventually performance has to be an issue -- after all, all these things are Turing computable and so can be simulated (at some performance level) by any machine. From what you told me, the 432's view of what an object is is so deeply ingrained that even re-microprogramming it has substantial problems. The 432: Data Encapsulation? You bet. Object oriented? No. Tim