Functional Reactive Programming in Practice

Talk given at StrangeLoop 2014

1. FRP In Practice: Taking a look at Reactive[UI/Cocoa]
2. Why?
3. We can no longer write synchronous software — Bob Dylan, quoting Abraham Lincoln
4. GitHub Desktop
5. Reactive Extensions: Async Data Flow FRP
6. Rx Primitives Creating Observables: Observable.Return(42) Transforming Observables: something.Select(x => x * 5) Changing Contexts: something.ObserveOn(RxApp.Ma inThreadScheduler)
7. ReactiveUI: Rx + Your Favorite UI Toolkit
8. ReactiveCocoa: ObjC + Your Favorite UI Toolkit whose name literally ends in "Kit"
10. ReactiveUI Primitives WhenAny: Observe properties of objects ToProperty: creates read-only derived properties from Observables Command: an abstraction to invoke and marshal an asynchronous method Observable Lists: mutable lists whose changes can be Observed
11. WhenAny this.WhenAny(x => x.Sha, x => x.Value) .Select(sha => sha == null || sha == unknownSha ? "(n/a)" : sha.Length > 7 ? sha.Substring(0, 7) : sha) .ToProperty(this, x => x.ShortSha, out shortSha);
12. Commands LoadUsersAndAvatars = ReactiveCommand.CreateAsyncTask(async _ => { var users = await LoadUsers(); foreach(var u in users) { u.Avatar = await LoadAvatar(u.Id); } return users; }); LoadUsersAndAvatars.ToProperty(this, x => x.Users, ref users); LoadUsersAndAvatars.ThrownExceptions .Subscribe(ex => this.Log().WarnException("Failed to load users", ex));
13. Observable Collections public class TweetsListViewModel : ReactiveObject { ReactiveList<Tweet> Tweets = new ReactiveList<Tweet>(); IReactiveDerivedList<TweetTileViewModel> TweetTiles; IReactiveDerivedList<TweetTileViewModel> VisibleTiles; public TweetsListViewModel() { TweetTiles = Tweets.CreateDerivedCollection( x => new TweetTileViewModel() { Model = x }, x => true, x => x.CreatedAt); VisibleTiles = TweetTiles.CreateDerivedCollection( x => x, x => !x.IsHidden); } }
14. Elm and ReactiveUI comparisons · Both effectively use time-varied values (i.e. Behaviors) · Elm helps you with scheduling and ordering, Rx makes you think about it explicitly · Elm disallows signals-of-signals, Rx lets you use them (but doesn't provide any help with it)
15. Some great things
17. A Consistent Model of Asynchrony · Imperative code has 1030534x ways to model asynchrony (callbacks, events, promises) · Modeling asynchronous code in tests is actually possible
18. A Powerful Model of Asynchrony · Describing complicated policies of asynchronous operations become sane to write · Discrete UI concepts can be written succinctly
20. Expressing Complicated UI Intent // Dismiss the undo flash after a certain amount of time. // NB: The logic here is, "Show the flash for at *least* 5 // seconds. If the user does any UI action after that, *or* // it's been a super long time, dismiss the flash" Observable.Timer(TimeSpan.FromSeconds(5.0), RxApp.MainThreadScheduler) .SelectMany(_ => Observable.Amb( anyUIAction.Take(1), Observable.Timer(TimeSpan.FromSeconds(20.0), RxApp.MainThreadScheduler).SelectUnit())) .TakeUntil(ex.DoUndo) .Where(x => ex.Ok.CanExecute(null)) .Subscribe(_ => ex.Ok.Execute(null));
21. Time Travel Testing
22. [Fact] public void CommitterDateTakesPrecedenceForRelativeTime() { new TestScheduler().With(sched => { var c = new CommitModel(Mock.Of<IAvatarProvider>()); sched.Start(); c.AuthorDate = sched.Now; sched.Start(); Assert.Equal("just now", c.FriendlyRelativeCommitTime); sched.AdvanceBy(TimeSpan.FromMinutes(5)); c.CommitterDate = sched.Now; sched.AdvanceBy(TimeSpan.FromMinutes(5)); c.UpdateCommitTime.Execute(null); sched.Start(); Assert.Equal("5 minutes ago", c.FriendlyRelativeCommitTime); }); }
23. "I have no idea how anyone can write apps without Rx" — Every developer I've met who gets good at RxUI / RAC
24. Some not great things
25. Team Education · Most developers haven't used RxUI / RAC before, every developer we hire has to take The RAC Class™ · New developers will fall back onto writing imperative code
26. UI Frameworks care about Threads (or really, just one thread)
27. Space and Time (leaks)
28. Creating long chains of events is cruise control for memory leaks
30. Virtualizing Lists are really sensitive to allocations Rx loves allocations
31. Long Call Stacks are Scary
35. Tooling can Help Us Out
36. Logging and tracing · Setting up a good logging framework is critical · Log Signals that end in an error · Signals are natively dual-point logging (i.e. they represent a Span)
41. Thanks! — @paulcbetts