>I just caught this on the Strongtalk mailing list, and thought that 
>it may be of
>interest to some of you. See  http://www.strongtalk.org  for more info.

This could be the most important thing to happen in the Smalltalk 
community in years.  The Strongtalk VM was faster than any other when 
it was written, and I believe it is still comparable to the 
VisualWorks VM (it would be fun to test).

One could ignore the type system and simply port all of Squeak into 
Strongtalk (of course there are parts of Strongtalk that are better 
and should not be lost ;-).  Then, or in the process, if one did a 
tasteful job of supporting the types optionally (ie a browser switch 
to show them or not), it would be the first opportunity to have the 
best of both worlds in Smalltalk -- or anywhere for that matter.

The Strongtalk VM is organized as a high-performance interpreter (2-3 
times Squeak speed, I believe), and an inlining JIT that achieves 
roughly 6x Squeak speed.  Gilad reports the following on his Intel 
Mac:

	Squeak 3.8    345,712,356 bytecodes/sec;  7,855,215 sends/sec
	Strongtalk  1,805,996,472 bytecodes/sec; 48,075,256 sends/sec

My mind reels at these numbers.  Moreover Robert Griesemer had a 
design for an even better JIT and, if this became an active project, 
I bet he would help out.

Strongtalk is set up to support native windows, and it probably makes 
sense to keep it that way, but this would be a parting of the ways 
from Squeak's run-anywhere agility. It would be nice to introduce a 
layer in the UI with a separate bitblt-only implementation to retain 
extreme portability.

The VM is not simple -- it is a large body of C++ code.  However it 
was written by smart people and is well-organized (I haven't looked 
through it carefully).  It probably has some bugs, and it may take 
some archaeology to get it all to compile with the latest tools.

That said, I think there would be a tremendous reward for doing the 
work.  The ironman engineering of Strongtalk seems a perfect match 
for Squeak's cheerful insouciance.

	- Dan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20060912/4d385dd5/attachment.htm