Re: Almost bought a QL
Was the M68k really 32 bit internally? I never used it in anger (I went 6502 - ARM) but my recollection is that it was essentially 16 bits with some instructions capable of operating on pairs of registers, somewhat like the Z80. Would you call the Z80 a 16 bit chip?
The reduced-width data bus was common back then. Every pin added cost, not only in traces on the motherboard but the CPU's packaging - DIL packages become very cumbersome when you try to make them with enough legs to support 32 bit addresses and 32 bit data as well as all the control signals, power etc.
In the early 1990s I worked with Intel's MCS-96 family, a "16 bit" family, at least one variant of which had an 8-bit memory bus multiplexed with one half of the 16 memory address lines. Retrieving a 16 bit value from memory involved four steps - latching the address, an 8-bit read, a second latch, a second read. All to save perhaps six or seven pins on the package - although you saved 8 data lines, you had to have an additional line (or two?) to signal whether it was address or data on the multiplexed lines. Oh, and you also had to fit an external data latch.
Bearing in mind that I was a student on placement, and had to self-teach pretty much everything on this project (did all the digital hardware as well as 90% of the software in assembler), I remember one "lightbulb" moment very well. I couldn't work out why my code wasn't saving values properly to EEPROM. It took a couple of days of pouring through the code and probing signal lines before I realised that the EEPROM had a 1ms write cycle time. The RAM had no problem with 16-bit accesses, but the EEPROM couldn't keep up with the double-write required. Solved it by writing as descrete 8 bit writes with a few NOPs in between.
I really pity the poor person who had to take that code on when my placement year was over...