[squeak-dev] Re: [Pharo-project] Xtreme Pharo hacker needed :)

Marcus Denker marcus.denker at inria.fr
Mon Jun 28 17:10:23 UTC 2010

On Jun 28, 2010, at 1:46 AM, Eliot Miranda wrote:

> Note: The Cog VM crashes on TestObjectsAsMethods.
> Yes.  I never tried to get this to work.  How important is this to people?

For Research: very. Lots of the experiments we did at SCG used this at least at some
stage of implementing the prototype:

	-> For the AOSTa experiments I looked at it originally 
	     (see http://www.slideshare.net/MarcusDenker/diplom-vortrag-slides)
	-> Persephone and therefore Reflectivity: http://scg.unibe.ch/research/reflectivity
	-> ChangeBoxes used it: http://scg.unibe.ch/scgbib?query=Denk07c&display=abstract
	-> I think ClassBoxes, too
	-> The most commonly used implementation of MethodWrappers these days uses it
             (and is *much* simpler than the method-compilation-stub-generating one)

It's a quite nice way to hook into "method execution" without having to resort to compile a stub method.
If you need an "in-image" JIT of some sort, it provides a real nice way of doing this purely from the 
image side (no VM change needed):


I think Andreas originally suggested ObjectsAsMethods years ago and I then made sure that it was 
integrated in the VM, as the AOStA experiments showed how powerful this can be for experiments. 

AOStA had to use a patched VM, then later persephone could run on a standard VM, which simplified things a lot...

Yes, stub-methods can do anything (and with care can be faster), and having a good code-generation framework 
simplified actually doing this a lot.... e.g. one can use bytesurgeon and do it on the level of bytecode.
(the paper uses MethodWrappers as an example: http://scg.unibe.ch/cgi-bin/scgbib.cgi/abstract=yes?Denk06a)
Or just the RB AST (with the code generator), maybe helped by Helvetia's quoting... 
(more on Helevetia's quoting implementation: http://scg.unibe.ch/research/helvetia/languageboxes)

Nut nevertheless: a MOP for method execution is  very valuable not just from an implementation standpoint, even more
from a concepual / thinking point of view. Good MOPs provide a better way of *thinking*, they enable exploration.

I always rated this as one of those changes that, while being simple, opened up a lot of space for experiments.

And it showed to me the value of a platform where a change like this is integrated and not ignored.


Marcus Denker  -- http://www.marcusdenker.de
INRIA Lille -- Nord Europe. Team RMoD.

More information about the Squeak-dev mailing list