Re: I'm new and wanting to help!

[Joel Dice <dicej at mailsnare dot net>]

On Thu, 3 Aug 2006, Tom Tromey wrote:

> Joel> Would it be possible to use debug information to determine the stack
> Joel> layout, similar to how DWARF is used for unwinding?
>
> Yes.  Actually there is a paper from 10 years ago or so where the
> authors did just this, in GCC, for Modula-3 as I recall.  As I
> remember the paper, it was a lot of work.  And, the patches were never
> integrated.

Well, it's nice to know it's possible, anyway. :)

> I'm not a GC expert, so I don't know how well this would really
> compare to what we've got now.  And, maybe a mostly-copying GC would
> serve just as well.

I'm no expert either, but my understanding is that you pick copying GC 
when you want to maximize cache locality and minimize fragmentation.  In 
order to get an idea of how such a strategy compares with conservative GC, 
it may be worth benchmarking something like PLT Scheme, which supports 
either, I believe.

> Joel> Also, how might java.util.IdentityHashMap or the hash-synchronization
> Joel> option be implemented when using a copying GC, given that object
> Joel> addresses may change at any time?
>
> I'm not sure how other VMs do this.  I know earlier ones kept the
> "system" hash code in the object header, but I'm not sure whether this
> is current practice.

I've just assumed that other VMs don't use copying collectors.  One option 
would be to have support for IdentityHashMaps built in to the GC, so that 
each instance is rehashed after GC.

 - Joel