z-machine on pdp8 (or paging Woodelf)

On 4/4/07, David Griffith <dgriffi at cs.csubak.edu> wrote: This is more targeted towards the VM model for a native Zmachine, but the philosophy holds, I think... The model that Infocom used was to not swap the interpreter ever. The interpreters written in assembler for the 6502 were a few K-bytes. The one that Mike Riley wrote for the 1802 is also a few K-bytes. By "few", I mean more than 5, less than 10. The pure (unchanging) game file pages could be overwritten at any time because it was easier to re-fetch them from known places on the disk than to track where they might end up if written dynamically. The "impure" game file storage (variables, stack...) were read in from the floppy *once* per session, then written out as the saved game. They were never swapped out during game play (the performance hit would be *terrible*), though there's nothing in the architecture that forbids it. The model was to never write to the game floppy, only to read from it. Games were saved by request, all at once, by recording all of the changed game pages to a floppy (a C-64 floppy could hold, IIRC, 4 v3-game saves per 170K, but there's no guarantee they were using the disk efficiently). I would think that emulating a processor that's running an interpreter would make it exceedingly difficult to differentiate memory types used but the application. This would turn the -8 emulator code into, essentially, a huge exercise in disk swapping with the side effect of interpreting PDP-11 code, I'd think. I would think that if you were going to try to emulate an -11 on an -8, you'd want the simplest possible -11 program running under that environment - Frotz isn't that simple. Megan Gentry has a nicely working Zmachine written in MACRO-11 for RT-11. Frotz, owing to its origins in C, is going to have huge amounts of library overhead. -ethan