z-machine on pdp8 (or paging Woodelf)

On 4/4/07, woodelf <bfranchuk at jetnet.ab.ca> wrote: When you say "game data", do you mean the impure area? If so, I'd agree for v5 games, but it's substantially less for v3 games - i.e., the ones that shipped for the Apple II and the TRS-80 and the C-64, etc. The larger of the home 8-bitters could run some v5 games, but I think 32K was the cutoff - machines that had 32K or less were never asked to run v5 games. Two years ago, I assisted Mike Riley, the author of ElfOS for the 1802, with a new from-scratch Z-machine. The basic platform was 1802 CPU, 32K of RAM, an IDE interface, and a few K of ROM for some OS toolbox routines. The SpareTimeGizmos Elf2000 conforms to that hardware requirements list, as does a machine built by Mike Riley himself. We were able to get the RAM-resident portion of the OS, the interpreter, and the impure area for a v3 game into that 32K with about, IIRC, 1.5K left over to load pure (read-only) game file data into. This lead to a lot of disk activity since there weren't enough pages to keep the parser resident, but part of that difficulty is that the page size of 512 bytes is double the page size of, say, a C-64 floppy, so when you only need one string or one subroutine from the far end of the game, you end up flushing 512 bytes to grab the enw stuff, and if it was a block with parser code, re-fetching that part of the parser, back and forth. It hasn't been effectively possible to write games in Inform for v3 for several revisions of the Inform library, but I think the compiler can still emit v3 game files. It limits what one can run, but does not mean that there are absolutely no files to use (or that it's imposible to generate test game files to exercise parts of a z-machine under test). The output from ScottFree is v3 compatible, and doesn't use the libraries at all - the parser is so simple that ScottFree just tacks it onto the reparsed Scott Adams egine statements when it builds a game. Having looked at the possibility of a 12-bit Z-Machine in the past, I think it's tough, largely because the Z-Machine architecture is optimized for an 8-bit (or 16-bit) host. Keeping track of what your current virtual address is going to require a 24-bit pointer on the PDP-8, even though only 18 bits or so of the pointer is ever going to get populated. The integer math is all 16-bits, etc. Not a huge amount of work on a 6502 or a Z-80, but plenty of extra humping for a 12-bitter. I'm not saying that makes it impossible, but it does make it more difficult. I'd love to see a 12-bit Z-machine interpreter, but I just haven't had the time to work on it over the years. Instead, I've studied game internals and written a few games from scratch in Inform. Still... it'd be fun to play Zork I on the PDP-8, just for the staggering geek potential. -ethan