> It is slick to include the display buffer in the checkpoint. We did this
on
> the Luna port.
Slick, but it has hideous paging implications. The first time a newline is hit in the first terminal window the entire display must be marked copy on write.
This trick was feasible on a one megapixel monochrome display. It requires too much memory on a modern >megapixel 32-bit color card.
In EROS, the display buffer is outside the checkpoint contract. A decent
graphics system already needs to know how to issue a "redraw" request in any
case. Also, user code is better able to deal with things like "the display
card got swapped, now what the hell do I do?"
Observation: I may be wrong about this being a good idea, but it's really
easy to add the display buffer to the checkpoint area later. It's really
hard to go the other way around and edit the gui/app protocol.