Interesting. My students used Language Server Protocol data to make syntax-aware diffs. Very promising project. Unfortunately, everyone moved on. I am currently looking for ways to revitalize it.

https://github.com/shishyando/tokenized-myers-diff