And?

"However, that device may be incapable of holding the data on its input port for more than a millisecond."

This isn't a problem in this case, really. It's far more important that the signal isn't missed (and with interrupts there's a much larger time window for that than polling, at the very least twice as long) than timely servicing.

However I didn't think of interrupt chaining. It's certainly relevant for, say, the vertical retrace interrupt, because IRQ2 can be shared.

That said, with a decent processor, latency is not an issue. Running in NTVDM, the problem is getting pre-empted by Windows. If you poll, you've lost huge gaps. If you use interrupts, you've got a much better chance of hearing about something, eventually.


One has to keep in mind that what worked best on the XT isn't necessarily what will work best on a much faster computer in a multitasking environment.