I will second  that the whole shadow stack is a bad approach from the start
, it needs a redesign. Eg apps can emit GC code snippets  , while not too
difficult to do it’s a architectural change which requires  more from
application and hence some politics.. The strength of MMTK is its simple for
the apps but it just doesn’t cut it when compared to competitive products (
Mono had a similar issue using Boehm compared to the CLR whose GC blows it
away) 

 

Ben

 

From: [email protected] [mailto:[email protected]] On
Behalf Of Jonathan S. Shapiro
Sent: Monday, April 04, 2011 2:05 PM
To: Discussions about the BitC language
Subject: Re: [bitc-dev] Request for advice on choice of a GSoC project

 

Jérémie:

Here are my thoughts.

Two important considerations in the GSoC evaluation are "impact" and
"realism". Impact means "how many people will benefit, and how important is
the effort?" Realism means "is the work proposed realistic?"

In my opinion, work on Jikes RVM is low impact, because the Jikes RVM is not
widely used. I'm actually a fan of Jikes, but it hasn't exactly taken over
the world.

Work on MMTk is even less useful than work on Jikes RVM, unless it benefits
LLVM.

Work on Mono or LLVM will have very broad impact, and GC is an area that
both are actively working on.

So I would suggest that you focus your attention on Mono or LLVM. The
challenge for Mono is more manageable: they understand the problem, and they
need to work on a precise GC implementation. The challenge for LLVM is
larger: their approach to dealing with GC is complete crap, and a viable GC
plan involves significant rework to the LLVM infrastructure - or at least,
this was true the last time that I looked. Also, the LLVM team does not seem
to take GC seriously, so there is lower likelihood of acceptance.

My suggestion: any improvement to mono will have very broad benefit to the
community, so that is a very good place to put your effort. Another good
place would be the Dalvik VM's GC. I know that GC in Dalvik is something on
Daniel Borenstein's "to do" list...

shap

_______________________________________________
bitc-dev mailing list
[email protected]
http://www.coyotos.org/mailman/listinfo/bitc-dev

Reply via email to