This is really surprising! I've never considered the possibility that using an equality test to skip a write that would be a no-op could break a dependency and thus lead to higher perf overall if the "equal" outcome occurs often enough. This might be applicable in many situations where you "edit" some data in-place, but most of the time there are few or no changes.
3 comments