How diffsitter works, How performant it is and What cases are exemplatory good and bad

Difftastic has a complete section dedicated what methods it uses, but has the performance not written in the repo but in the inspired projects:
https://github.com/Wilfred/difftastic#how-it-works
Could you elaborate what algorithm you use for the diff? A* and dijkstra for graph difference are the most prominent ones, where A* is much more performant on multiple changes (in a function) but has overall worse results.

Performance of examples (ideally with benchmarks), Known problems and Non-Goals would complete the first impression.

I am asking due for an evaluation of [neovim what to use](https://github.com/neovim/neovim/issues/15064), but I am not affiliated or core member (only neovim user) and would like to see things moving forward.