helios175
commented
4 years ago Hi, thanks!
This started as a small effort inside Square to provide a way to build our recycler views. To be honest, I didn't know of the ListAdapter at that point. When I introduced diffing, the question about how to do "async" came up, and coroutines were the answer. So yes, ListAdapter wouldn't have worked in that setting.
Nevertheless there's a side-effect which I think is positive: we are in control on how the update works. We use the same diffing algoritm (DiffUtil) but we decide what to do with the results. If you check the Recycler and Update classes you'll see how a second update invoked while the first updated is being diffed, works just fine. When a diffing (non synchronous list of "notifiy" updates) ends the "original data" gets compared to know if this diff is still up-to-date or was superseded by a different update.
Also there's the Update.addChunk feature, which is handled differently. Maybe it can still be done with ListAdapter (I didn't check) but having complete control doesn't hurt.
Otherwise –control and coroutines– there's no special reason it didn't use ListAdapter.
PS: in Square we are working with declarative UI, and in that context, a recycler component, would receive "added chunks" or new data pages, as a completely new data source. We still want to avoid a big diff in such cases, so we will need to come out with a way to provide new chunks of data to a declarative-recycler-component. Some mechanism for it to know that "oh, this is a different data source, but it's the old one plus some extra data at the end". So there might be improvements in the update area of the Recycler, where maybe the "notifications lambda generation" gets moved out from Update.
emmano
Great library. I was just wondering why
RecylcerView.AdapterandDiffUtilwere used instead ofListAdapterandDiffUtil.ItemCallback? Was it to leverage coroutines for diffing operations?