See discussion in T1279, particularly https://secure.phabricator.com/T1279#95

Some modern discussion in T4481. I actually think this behavior is desirable as-is: state S was acceptable, and then we went to state S' and back to S, but I don't see the result as different from the original, acceptable state. Particularly, I would expect planning changes by accident and then immediately requesting review to cancel one another out, rather than require me to go bug people. Are there specific situations where this workflow is bad / undesirable? e.g., is there a reason that we're often actually ending up in state S'', not back in S?

Or, rather, there are two cases here:

1. You update the revision.
2. You "Request Review" the revision, without updating.

I think in (1), we should invalid the old accept (we've changed states), but in (2) we should not (we're back in the same state we were in before).