[Vm-dev] How difficult to make a .NET VM?

[Aran Lunzer]

Hi

[Apologies if this has been addressed before or elsewhere; my Google
attempts didn't turn up anything that seemed relevant.]

Has anyone tried packaging the Squeak VM as a .NET class?  (...if that's a
sensible way to ask the question)

For the past couple of years I've been using and hacking the Squeak/.NET
bridge, created by John Pierce and Ben Schroeder, mainly so I can create
and control IE-like Web browser views within the Squeak desktop (see, for
example,
http://km.meme.hokudai.ac.jp/people/aran/papers/UIST06-LunzerHornbaek.pdf).

While this has been successful to the degree needed to be a platform for
HCI research, the setup is still - frankly - an untidy and fragile hack.

The main challenge has been how to combine the .NET views with the Squeak
desktop.  Originally we made them children of the Squeak window, but this
gave us a nasty, intermittent thread-based deadlock.  We worked around
this by keeping the .NET views independent but making them hover in front
of the Squeak window in the Windows Z order; some tweaks to the handling
of focus-change messages in the Squeak VM allow us to ensure - most of the
time - that the combination shares the desktop nicely with other Windows
applications.

However, as I say, it's pretty fragile.  So we're wondering whether
there's a clean way to make the .NET views be children of the Squeak
window, only without running into the COM vs. .NET threading issue that
hit us before.  In particular, unless we can crack this issue it seems
unlikely that we can get the Squeak+.NET combination to behave acceptably
when embedded within a Web browser.

I suspect that if we could package the Squeak desktop itself as a .NET
Form we could then make the other views be proper children of that
desktop, and the window management problems would all go away.  I also
imagine that we could then go the extra step of packaging this .NET-based
VM as a browser plugin.  If it all works, the result would be a pretty
useful (though Windows-specific) Squeak+.NET environment.

So: can anyone shed some light on how complex this is likely to be?  Has
anyone already tried it?

...or would we be better off just looking for another way to get Web pages
rendered?  A Squeak-native approach such as Scamper seems doomed to lag
the state of the art by years; has anyone looked into packaging a library
such as Gecko for use within Squeak?


Thanks for any thoughts -

Aran
--
Aran Lunzer
Hokkaido University, Japan
http://km.meme.hokudai.ac.jp/people/aran/