Andreas Timmermann wrote:
> 
> > Note that SOAR never actually ran any applications, just tiny tests. By
> > the end of the project it had become obvious that dynamic translation
> > (Jitter) of bytecodes was more practical than native compilation. This
> > lesson was applied to VisualWorks, Mushroom and Self.
> 
> Could you elaborate on this?  The only reason why bytecodes would be more
> practical seems to be that the code is recompilable if you don't have the
> source any more.
Perhaps the most important reason for liking an abstract instruction set
is that the analysis tools - scanning for variable access, debugging,
all that stuff - need only be written once. Then the dynamic translator
needs only to produce code that accurately implements the abstract
instruction stream at which it is pointed and some support to allow the
debugger to map back from the native code state to the abstract code
state. Sounds dead easy when you write it quickly, doesn't it!

tim