Re: proposal: VMStackBrowser

[Steven Augart]

Archie Cobbs wrote:
> Steven Augart wrote:
>
> > I propose adding a new class that may be optionally implemented by the
> > VM, gnu.classpath.VMStackBrowser.  [....]

> > Looks good. Presumably the idea here is to optimize this in such a
> way that you don't have to crawl the entire stack in the constructor
> (as is done by the default implementation). 

Yes, exactly.  Thank you for explaining it so concisely; I'll incorporate
that phrasing into the class's docs.

> In that case, to make such
> implementations easier, we should specify that the VMStackBrowser
> object is only valid until the instantiating method returns, 
That is a Very Good Point.  Absolutely.
> and that
> the instantiating method is required to call release() before returning
> (e.g., using try { ... } finally { x.release(); } or whatever).

Yes.  In both of these cases, my (Jikes-RVM-Centered) assumptions are
showing.  I swiped the methods from Julian Dolby's work on Jikes RVM's
VM_StackBrowser.  We turn off garbage collection before we initialize
a VM_StackBrowser's internal frame pointer, and then turn on GC
again when we're done.
> Then simple implementations that store the current position as a
> frame pointer (or whatever) are possible and don't need to worry
> about the stack changing out from under them.
Well-expressed, again.  It's great to have folks on here who read my mind
so well :).

Thanks also to Jeroen for his feedback.  I'll submit a formal patch to
classpath-patches sometime this week, and it will include converting
more places in Classpath to use the new VMStackBrowser.

--Steve Augart