Maybe name this self.function instead?

I think there was a reason I used self.expression. But I don't recall why (it was a while ago, while I was making the stepping-graphs for key instructions. It could just have easily been a mistaken assumption). I'm happy to change it to self.function and if we run into issues (perhaps with printing?) we can reassess, rather than have unnecessarily cryptic code.

How do the arguments to this work? I would assume that this function would be willing to accept a on_line argument, and I'd also assume that in the absence of any line argument ("on" or "from/to") this command is essentially meaningless, so it should be necessary to pass at least one.

Do you feel strongly that on_line is better than from_line when to_line is omitted? My preference is to keep it to just from_line as it still makes sense to me and keeps things simpler, but I don't mind changing it if you'd prefer.

I agree that we need at least one line arg - done.

Nit, I don't know how I feel about this particular adjectivalisation, maybe we could avoid it by changing it to "Cannot eval DexContinue commands" or similar, or maybe you can just ignore this comment as being a matter of taste!

I agree with you, but I copied this from other existing commands, so I'm inclined to keep it for uniformity.

I'd suggest creating two separate words for it. In debugger parlance, I think "hit_count=n" means after n hits, maybe we could replace the alternative form with "repeat_count"?

DexFinishTest works in the hit_count sense, but other commands work in the repeat_count sense. Could we land this as-is and separately fix this inconsistency across all the commands?