rolling a PCB rev

Depends how good the first layout was. Part of my job at Cambridge Instruments was "modifications" which was dealing with the problem the final test technicians ran into getting a board calibrated and working right.

I remember a particular board that was crawling with comparators which were very prone to self-oscillation. The prototype might not have been, but as the manufacturers improved their processing and shrank the sizes of their devices life for final test got progressively more difficult.

We eventually laid out the board again to minimise cross-talk and put the decoupling capacitors where they would do most good. The new layout wasn't perfect, but the new boards didn't self-oscillate nearly as often, and the tweaks that killed the oscillations mostly worked first time. Final test wasn't entirely happy, but they were lot lot happier than they had been.

It wouldn't have been difficult to add extra components to make the board even more stable (where there was room for them) but I don't think that we did. Keeping the parts list the same was definitely the preferred option

I believe 1b+ came out after all the others, and is after 1b but before

how do you track documentation and sofware changes, or match them to each board rev? sounds like nightmare.

A pcb can go through a lot of layout iterations before we ever etch a board. The numbered iterations are (like B2, B3...) are not revs. B is a rev, and that gets etched.

Firmware is associated with actual, buildable board revs.

Actually, our automated test sets verify that the loaded code and cal table and hardware rev are compatible.