PhiHo,

>     One more thing, SqueakFFIPlugin class >> moduleName 
> returns 'FFIPlugin', should it be 'SqueakFFIPlugin' ? Anyway,
> I changed it and still got 'callout mechanism not available' when
> tried 'Win32Window coloredEllipses'.

As I said before if you change the module name you'll have to change
#primitiveCallOutToFFI to reflect the change. Oh, and you'll have to make
sure that the Squeak VM finds the "right" FFI plugin if you have more than
one on your disk.

>     Btw, my Squeak.exe is now 187KB, with 26 external plugins 
> (a few are still missing inaction, because of the linking problem
> of the MinGW dlltools). The total disk space required is 984 KB.

That's a bit to large. It should be about 600k - did you strip the DLLs?!

>     I didn't get the 10% gain in speed as you did for Acorn VM:

And I would've been very surprised if you had. What makes you think that
there should be any speedup at all?!

>     It's good news to me. At least it's not slower ;-)

But it's not a good benchmark because you're *not* testing any plugin stuff.
You're simply testing if your compiler does the same optimization that my
compiler does (which it should). A much better test would be to see if
there's a difference in calling (for instance) BitBlt; the large integer
primitives; the balloon engine or any of the stuff that actually resides in
an external plugin - #tinyBenchmark does not use any of these.

Cheers,
  - Andreas