On Sat, Sep 26, 2009 at 3:57 PM, James Abley <[email protected]> wrote:
> I saw your post on the JVM Language group as well. I believe Jython have
> implemented something like this to support generators in python [1].
> Hopefully someone from there might be able to share some practical
> experience / pain too.
Yes, I will talk to them. I believe for coroutines they had a simpler
time because of how limited coroutines are in Python. But I think
there was also the whole "greenlet" thing, as well as generators...and
perhaps there's some sharing possible.
> Further speculation (since I had to look up co-routines in Scott [2], I may
> well be just waffling here):
>
> Isn't there some continuation work going on in DaVinci? As a first step,
> possibly use that if available to deliver coroutines. Doesn't the compiler
> do something like that currently if invokedynamic is available? I guess that
> might be a route to delivering a first-class Enumerator#next and fibers
> implementation, falling back to being crippled otherwise. Not ideal, but
> what can you do?
There's definitely some continuation work in Da Vinci, but so far it's
pretty early days. I have high hopes that it can become the coroutine
API I really need, and given time I'd love to contribute and improve
it, as well as help hash out the dirty details and the API. So we'll
hopefully implement our thread-based coroutines in such a way that we
can swap in JVM support later on.
- Charlie
---------------------------------------------------------------------
To unsubscribe from this list, please visit:
http://xircles.codehaus.org/manage_email
