Subject: Re: publishing
From: Todd D Millstein (todd@cs.washington.edu)
Date: Sun Feb 04 2001 - 11:14:48 PST
Don't update yet, if you haven't already. There seems to be some problem
between my stuff and Sorin's optimizations that I neglected to test (and I
didn't know about the make-whirlwind-profile script until I'd already
committed). If you've updated already, everything should be fine without
optimizations. I'll let you know when it's safe to update. Sorry about
this.
Todd
On Sat, 3 Feb 2001, Todd D Millstein wrote:
>
> I'm publishing my system. Let me know if there are any problems. Here
> are the details:
>
> *) Added support for "initial" declarations. These are declarations
> whose expressions are restricted to essentially "syntactic values,"
> akin to the restrictions imposed by letrec in Scheme. Such
> declarations are executed at the beginning of a scope's execution,
> regardless of where they appear textually in the scope. Variable
> declarations with the keyword "initially," "once" declarations, rep
> declarations, method declarations, and "extend" (inheritance)
> declarations are all initial.
>
> Initial declarations help us kill two birds with one stone:
>
> 1) They function as a letrec, thereby safely allowing recursion and
> mutual recursion. By allowing the declarations to be textually
> separate (unlike in a letrec), we facilitate mutual recursion across
> files, in the global scope.
>
> 2) By restriction the method and extend declarations, they rein in the
> method and class hierarchies, essentially ensuring that the hierarchies
> are statically analyzable.
>
> See notes/ir.txt for a more complete description.
>
> Aside from the whirlwind/ directory, make sure to update the
> following in the vortex/ directory:
>
> Java/front-end/wil-src/ -- C code to translate from Java to wil
> Java/bin/shell/JV-em/ -- the JV-em script
> Java/src/example/wil/ -- wil code for towers
> Java/wil-stdbib -- wil code for java stdlib
> Java/wil-stdlib/native -- wil code for java native stdlib
>
>
> *) Teased out from codegen an analysis that simply tracks the current
> scope as it traverses the given graph. Codegen is now a client of
> this scope analysis. [I had originally written another client as part
> of constructing the gf/class hierarchy, but I'm now doing that
> differently. Hopefully the scope analysis will be useful for other
> clients in the future.]
>
>
>
This archive was generated by hypermail 2b25 : Sun Feb 04 2001 - 11:14:52 PST