There could be noticeable perf improvements if we were to stop using the reduce spread pattern in the diffing methods.

Some benchmarks from changing the diff method to use basic for loops:

original x 377,694 ops/sec ±0.62% (86 runs sampled)
new x 788,699 ops/sec ±0.55% (91 runs sampled)

Reference: https://www.richsnapp.com/blog/2019/06-09-reduce-spread-anti-pattern

I have most of the work done in a fork. Would you be open to a PR?