Optimizing Squeak

Hans-Martin Mosner hm.mosner at cityweb.de
Thu Feb 25 20:36:00 UTC 1999

Alan Lovejoy wrote:

> Exactly.  It should have been obvious years ago that the day would eventually
> come when the code size expansion of Smalltalk compiled directly to native code
> would asymptotically approach irrelevancy.  That day will be here soon, if it has
> not already arrived.

I have to disagree vehemently. One of the really nice points about Squeak is that it fits in
so little memory with very reasonable performance. Until recently, my Mac Performa had 16M,
which was barely enough for VisualWorks but plenty for Squeak. It now has 64M, but my
PowerBook still has 32M, so an all-native-code Smalltalk would probably swap considerably,
making it unusable for me on that machine. And dare I mention the various Squeak-on-a-PDA
approaches? Try to tell those guys about 128M of main memory just to start up!
So please, let the Squeak VM come in at least 2 (compatible) flavors:
1. pure bytecode interpreter with very small footprint and acceptable performace
2. threaded or native code JIT for better performance, requiring reasonable additional
and probably
3. pure bytecode interpreter with reduced primitive set and less inlining for the really
memory-challenged PDAs


More information about the Squeak-dev mailing list