new image format

Jecel Assumpcao Jr jecel at merlintec.com
Thu Oct 20 20:51:39 UTC 2005


Andreas Raab wrote on Thu, 20 Oct 2005 13:21:15 -0700
> Most people who looked at these changes seemed to dislike the fact that 
> the model exposed stack segments directly to the user. There seemed to 
> be general consensus that a VM might do such an optimization under the 
> hoods but that it shouldn't expose it to the user to keep the conceptual 
> model (liked frames) simple.

Ah, right! It does seem a bit silly since Self and Smalltalk V (among
others) did a good job of completely hiding the native stack they used.
In fact, Anthony also hides it in that there is no way to deal with the
stack of a process that is running so it can use any format it wants.
When a process is suspended (but not waiting) the stack gets converted
to the format he specified. He might as well have specified the
traditional format.

I guess he wanted to expose the overlapping frames format to the users
in order to make them deal with his new bytecodes instead of the blue
book ones. But no, that doesn't really make sense since though his
bytecodes allow access across more than one frame no real code can
actually use this, so showing the users separate frames would work just
as well.

Well, as Tim pointed out he isn't currently around to continue this work
so it might be better to use his simpler patches instead. Which is a
pity.

-- Jecel



More information about the Squeak-dev mailing list