Thanks for the suggestions, Marc-André. I'll definitely look into them. Since I obviously still have a great deal of learning to do, I think it's probably best that I go dark for a while and study the code, documentation, and existing discussion archives at the qemu & spice level. I have no interest in becoming a hostile fork project and little interest in becoming a downstream ugly 'mod' hack, so I'm going to study up and hopefully come up with something that has good benefit, minimal code impact and is supportable by others so that it might get adopted with minimal fuss. Anyway, I think you're right, this isn't looking like the weekends & evenings 'cut & paste' project I was hoping for and looking realistically at my obligations, priorities & schedule unfortunately it can't be much more than that right now.
If anyone has suggestions for how to best hook the existing model with little or no impact, I'd like to hear about it. My short-term goal is to create a separate playground module with minimal linkage where I can experiment. For what it's worth, currently I'm thinking that using existing spice/vnc infrastructure for boot, etc until a hand-off to either a driver 'dma-bypass' mode or user process can take over the screen-scraping acceleration function. Then probably use a timeout mechanism for fallback to the existing methods, this should be fairly robust. (?)
