Re: small thoughts on performance engineering

[Roger Gregory <roger>]

You're on track so far.  The other thing to remember, is that
the translation process often brings up other problems that
need to be addressed.  I can see certain types of problems that
are uniquely C++, the Smalltalk unique problems being dealt with 
rapidly, the C++ problems become more important, and more expensive 
to correct.  For instance, inlining of code is something that
has to be measured in C++ and the correction may involve some changes
to the translator.  I expect a factor >2 perhaps >5, because of
the way we use the language.

I expect that most of us have a fairly good feeling for the mapping
between C++ & Smalltalk performance.  As long as we keep checking and
correcting our intuition with gprof we are probably going to do quite well.

Also note that the major way we expect to speed things up is with casheing,
this should work identically in C++ & Smalltalk.  This is the biggest win.