My actual scheme is similar in some ways to what you describe. More complex, because it involves many different types of structures.

Also, my approach is more explicit, I don't "restart a parse" in a general way, as I contemplate each possible change explicitly. This is prohibitive to do by hand for a full grammar, but I'm not doing full C/C++/C# analysis, so it's doable. It also makes the dynamic parser fully incremental.