[ANN] Chuck type inferencer

Lex Spoon lex at cc.gatech.edu
Tue Jun 1 19:28:41 UTC 2004


Colin Putney <cputney at wiresong.ca> wrote:
> This is something I've been working on addressing for the next version 
> of OmniBrowser. Essentially, the next version of OB will operate on an 
> AST that begins with Package at the top and extends down through Class 
> and Method to sub-method parse nodes like MessageSend and Literal.

Yep, I would love to have something around like this and not maintain my
own.  Already I have stopped using  my custom parser in favor of RB's,
and I could probably have dumped my parse nodes though that is more
trouble than its worth so far.  I still am having to deal with keeping
an object model of the code in the image (or *not* in the image,
depending on what you are analyzing).  A lot of this object model could
surely be offloaded into a tool like OmniBrowser...  but of course, it
doesn't actually exist yet.  :)




> The thing that leaps immediately to mind 
> is inter-package dependencies. I bet it would be feasible to deduce the 
> minimal set of packages required to run, say, the unit tests for a 
> package. (And the minimal code from the "as yet unpackaged" base image 
> too!)

Yeah, that's an interesting application of a dead code remover.  Dead
code removal is an interesting application, if nothing else because
Agesen did it with his inferencer.  :)  Unfortunately it is not trivial
with Chuck being demand-driven; it takes some thought to make it work,
and I am not completely sure whether it will be effective.  It will
certainly improve things on the current image-shrinking scripts, but it
might not be good enough to give you good dependency information between
packages.

-Lex



More information about the Squeak-dev mailing list