[Pipet Devel] Re: which handles distribution (was: vsh?)

Jean-Marc Valin jean-marc.valin at hermes.usherb.ca
Fri Mar 17 15:03:35 EST 2000

> > Yes surtainly does. I've proposed this before, but the overflow people
> > want to keep the processing speed
> > and dont want any wrapping.

I have nothing against wrapping in itself, nor do I think speed matters anymore
when you have the parts distributed across a network. What bothers me the most
is that your program would have to be written in "two different languages": one
GMS part, and one Overflow part, while I think it would be feasable to put the
GMS code (not rewrite it) into Overflow. Now, maybe the best thing would be to
do both: Adapt Overflow so that it can be used as a GMS node, but also add
network stuff to Overflow. The kind of task you want to perform will then tell
you which to use.

> This whole 'speed' issue has me confused.  If the small Overflow objects
> remain 'networked' on a sigle host (none are sent through the Internet), how
> can GMS affect their speed???
> And if small Overflow objects are 'networked' across the Internet, IT'S THE
> everone else?

True, my concern with speed was only that Overflow nodes could not be put in GMS
individually, but that's probably obvious to everybody. Sorry, I guess I have
not expresed myself correctly on speed. 

> You know, it's looking more to me as though both Jarl's and Jean-Marc's ideas
> for connecting/merging GMS and Overflow are not incompatible.  It seems
> they're two sides of the same coin.  Overflow will need GMS for
> Overflow-to-Overflow (across the Internet) communication, and GMS will need
> Overflow for application construction (and I think wrapping).
> The point of contention here is this: WHICH COMMUNICATES *DIRECTLY* WITH THE
> GUI???

I'm afraid we might need two GUIs (though they might be bonobized to look like
one), because GMS and Overflow will probably have different requirements. This
is what I'd find annoying in wrapping Overflow in GMS. I might be wrong on that

About Overflow integration in GMS... I'd like to point out that, except for the
GUI, Overflow is more a library than a program. It is easy to integrate into any
program. I'd like to know, is GMS build a bit the same?

OK, so this summerized my thoughts:
1) I don't mind if Overflow is integrated in GMS (mainly it's not a lot of work)
and I can help those who want to do it (It would probably be as simple as
writing a couple new nodes). 
2) I would like to try the opposite so for most simple "networked applications",
just Overflow would be fine.

As a last, about communicating objects between different parts, Overflow already
has an Object hierarchy (GTK-like, but cleaner) that supports
serialization/de-serialization, so most of the stuff would already be done. Does
GMS has something like that (in which case, we'd need to have compatibility
between the two)?

I hope it clarifies a bit what I think.


Jean-Marc Valin
Universite de Sherbrooke - Genie Electrique
valj01 at gel.usherb.ca

More information about the Pipet-Devel mailing list