Caterpillar

Yesterday, I talked about stepping backwards in a debugger. There is an inherent overhead to this. Not only do you need to generate the instructions to do this on the fly, you also need to store them. And computers can execute a lot of instructions in a short time, so such an "undo instruction" buffer might get quite long.

I'm not too worried about the performance overhead of logging the undo instructions. For now, performance is not a priority anyway. And I have ideas on how to alleviate, and in many cases completely eliminate, the overhead. Maybe I'm being too optimistic, but either way, that's a topic for another day.

I'm currently more worried about the length of the undo buffer. It would need to be limited, and on memory-restricted platforms (consider microcontrollers), it might have to be quite short. This restricts the usefulness of this feature, but I have ideas on how to get around that too.