[gclist] region inferencing

P. T. Withington ptw@harlequin.com
Thu, 6 Nov 1997 12:13:34 -0500


1-line summary:  region inference == flow analysis at compile time for data
liveness so the compiler can do correct, explicit deallocation.

[quoted from the gc-list archives]

>From: haahr@netcom.com (Paul Haahr)
>Date: Sat, 17 Aug 1996 09:38:38 -0700
>Subject: Re: [gclist] Compiler lifetime analysis

[...]

>Mads Tofte has done some work on what I would call the extreme end of
>this approach:  associating every object, at allocation time, with a
>stack frame, and deallocating it when that frame exits.  Obviously, the
>designated frame can be much earlier in the stack than the one in which
>the allocation occurs.
>
>As I remember, the work was done with closed-world analysis of ML
>programs, called region analysis, where type signatures are extended
>with annotations describing lifetimes of allocated objects.
>
>For details, see Tofte's publications, which generally can be found from
>http://www.diku.dk/users/tofte/publ/publ.html.  A couple of the more
>relevant sounding titles are:
>
>  A Theory of Stack Allocation in Polymorphically Typed Languages
>
>    http://www.diku.dk/users/tofte/publ/93.15.dvi
>
>  From region inference to von Neumann Machines via region
>  representation inference
>
>    http://www.diku.dk/users/tofte/publ/popl96.dvi

---
P. Tucker Withington, Harlequin Inc., One Cambridge Center, Cambridge MA 02142
Phone: +1 617 374 2557       Fax: +1 617 252 6505      "Honk if you love GC's"
The Memory Management Reference:  <URL:http://www.harlequin.com/mm/reference/>