> > The Strongtalk compiler does not do any special optimization for
> > timesRepeat:.  It is able to inline it without a special
> > case, using the general inlining heuristics.  So it should be
> > just as fast with timesRepeat2:.  The benchmark is a perfectly
> > fair one (for a microbenchmark) for Strongtalk- it is all
> > the *other* implementations that are "cheating" :-).
>
> I know and I agree. But Viktor was implicitly claiming that "Dolphin is
> a pure interpreter too" and apparently much faster than Squeak. So my
> point was about that this microbenchmark is no good for comparing "the
> performance" of Squeak and Dolphin since it's measuring quite different
> things. (I only brought VW into the arena since I had it handy and
> thought it'd be worthwhile to show that little compiler tricks can make
> a huge difference when it comes to microbenchmarks).

I never said it is benchmark simulating "real application", I know it is
not. It is simply short benchmark, nothing more.

Yeat another ST: Smalltalk MT 4.0 (fully compiled ST), time 150 ms, 333% of
Strongtalk.

Viktor