[Haskell-cafe] GRIN and Urban Boquist's thesis

[Lemmih]

On Thu, Feb 24, 2011 at 9:36 PM, David Waern <david.waern at gmail.com> wrote:
> 2011/2/24 Lemmih <lemmih at gmail.com>:
>> They will also be in the lhc repository once I restore it on code.haskell.org.
>
> Lemmih,
>
> while you're here, what's the status of LHC? It's an interesting
> project but we haven't heard much from you lately.

Yeah, I've been out of commission for way too long due to college
preparation and stuff. There's good news, however. For the last few
weeks I've been working nearly full-time on adding garbage collector
support to the backend. It is my hope that the work is interesting
enough to be published somewhere.

Feel free to skip this paragraph if you're not interested in garbage
collection. Presently, nearly all Haskell compilers use some form of a
shadow stack to implement garbage collection. This has two major
drawbacks:
 * Managing the shadow stack adds overhead (roughly measured to around
%62 in the case of UHC).
 * Forcing objects down on the shadow stack inhibits many optimizations.
Instead of using a shadow shack, Boquist used a method that was
specialized to the GRIN object model and it didn't have any of these
drawbacks. I plan on generalizing said method so it works with any
object model. Hopefully I'll be able to show that you can keep track
of roots without incurring any overhead on the mutator.

-- 
Cheers,
  Lemmih