Where should the type information be: in tags and descriptors 458

+--------------- The 32:16 was a lovely piece of kit - even had a fairly usable menuing system over Unix, and a word processor that looked a lot like Wang's dedicated WP systems +---------------

Yes, well... "For:Word" was coded from scratch with the primary design goal of being as close to a perfect clone of the Wang-OIS word processor as possible. ;-}

+--------------- (fantastic keyboard, I remember). +---------------

The touch (stroke, force feedback, etc.) was indeed very nice. A lot of engineering went into that. But the above requirement of "perfect Wang compatibility" perverted the layout of the keyboard considerably. In fact, we practically had a mutiny in Engineering when Marketing wanted to leave the "Unix keys" off the keyboard entirely! The "Unix keys" included left- and right-curly bracket, backslash, backquote, vertical bar ("pipe") and tilde -- imagine a Unix machine without those on the keyboard!! Fortunately, the crisis was averted by a "compromise": The "Unix keys" could stay, but they were banished to a completely separate gulag column to the left of the main keyboard. (Ugh.)

+--------------- By today's standards, it'd be horrifically slow... +---------------

Indeed: The 68k clock was only 5.5 MHz, and the I-O bus took three clocks for memory or DMA cycles and a fixed *nine* cycles for PIOs.

The blame for that is all mine -- I wanted to avoid variable wait states for PIOs so we could guarantee small latencies for fast DMA devices with small FIFOs (such as a certain Ethernet design), so I made them *all* fairly slow. Using larger FIFOs in the DMA devices and letting the PIO timing "float" from 3-20 cycles or so, as needed, would have been a better choice, since far too much I-O ended up being done with those fixed, slow PIOs.

And it was *small*: 1 MiB maximum main DRAM, and the kernel took up ~128 KiB of that.

Worse, Marketing's paranoia about software piracy led to the multi-user kernel which was an extra-cost option and all of the value-added apps being *encrypted*! And when someone pointed out that all you had to do was load up the apps, the run a large process to push the others out into swap space, then power-off the system and go read the swap disk elsewhere, Marketing forced Engineering to encrypt swap space, too!! This resulted in serious performance degradation, e.g., a trivial "cc -o foo foo.c" compilation using the internal-only unencrypted developmental version of the compiler and linker took 8 seconds; the version shipped to customers took over 12 seconds for the same thing!

+--------------- ...but they were mightily impressive at the time. Another one of those Unix "nearly success" stories - didn't Ford buy a load of them at one point? +---------------

Yup. And IIRC, they were also sold in quanbreasty to Reynolds & Reynolds for use as POS terminals in car dealerships.

-Rob

----- San Mateo, CA 94403 (650)572-2607