Update: I first had a naive approach and stored the full content of the text on each stroke. I updated the code with some small algorithm to reduce data and avoid storing the full text when a character was added or removed.