[Nestedvm] Ponderings...

[Charles Oliver Nutter]

On Sat, Feb 27, 2010 at 10:48 AM, Thorbjoern Ravn Andersen
<ravn at runjva.com> wrote:
>> On a separate email, I have been discussing ways to make NestedVM more
>> usable. So far we have considered the following possibilities:
>>
>>
>
> I've not seen this particular mail in the archives.  Was this off the
> nestedvm list?

Yes, I was emailing David Crawshaw and Wayne Meissner directly.

>> * adding a function-calling API based on Wayne Meissner's jaffl
>> library. He already has a start on this at his github project under
>> github.com/wmeissner
>>
>
> I am unfamiliar with jaffl, and it appears that all related content has been
> removed from the kenai website.
>
> Is this doing the same as JNA or is it a completely different approach?

Yes, it's basically the same sort of thing as JNA, but an improved
design with better performance characteristics. You can find the base
jaffl project and the jaffl-nestedvm project on Wayne's github account
at http://github.com/wmeissner

>> * providing a prepared VitrualBox VM with the toolchain preinstalled.
>> This may be available very soon.
>>
>
> This sounds like very a good idea.  Personally I use vmware, but I believe
> there has arisen a standardized distribution format for this.  Perhaps such
> a thing?

I'm not sure, but I know it's not going to be a VMWare image, since I
don't think VMWare images run on anything else. Currently the plan is
just straight-up VirtualBox.

>> * building the toolchain for MIPS directly, so it can be run on
>> NestedVM directly. This will take some time and hackery for sure.
>>
>>
>
> Self-hosting?  I don't think so.  It will require the complete unix
> environment to be up and running, and I also believe I've read that there is
> an issue with how large a program can be translated and run inside a JVM.
>  I'm not certain though.

I think Wayne mentioned that using NIO instead of a real array could
increase the size of the memory image, which might make this possible.
But yes, the self-hosting goal is a pretty hard one, since you'd need
most of the toolchain to be self-hosted as well.

>> The improvements you suggest all sound very logical. Perhaps we need
>> to find someone acting as "lead" of NestedVM to spin up a real bug
>> tracker and project site so we can start making a real go at this? The
>> success of libraries like sqlite-jdbc show me that this thing has
>> legs.
>>
>
> Most progress is driven by a concrete need, and I would think that the lead
> for this endaveour should have one.
>
> But the idea is sound, any takers?

I'm hoping to take a wack at using jaffl-nestedvm to bind sqlite3 to a
JRuby extension, which might help shake out any early issues. You
might be able to help by looking over the jaffl-nestedvm support and
trying to use it for something else.

Do you have a concrete need driving your interest?

- Charlie