Another toplevel comment claims it is relevant for the use case where you stuff the whole corpus into a single stream. When you want to add new data, you don't

Another toplevel comment claims it is relevant for the use case where you stuff the whole corpus into a single stream. When you want to add new data, you don't want to start over compressing everything. https://news.ycombinator.com/item?id=27441474 https://news.ycombinator.com/item?id=27441474