This is said often but without much qualification. I've found mutable data structures with change propagation (via observable or what not) to work much better given that the whole diffing thing can be avoided altogether since you know exactly what has changed.
It is my understanding that the DOM is broken in how it handles invalidation/re-rendering (doing it for each modification rather than batching), but again, I don't see how immutability helps fix that problem any better than just doing it the right way with a mutable virtual DOM?
Almost every application (ever) is a list of lists of lists (and so on). Even text can be broken up into paragraphs/code-blocks etc which form the lists. If these structures form a tree, and that tree is somewhat well balanced, then small changes can be found in log(n) time by comparing reference identity without developer intervention (by either a mixin or Om-like system). log(n) ends up being extremely fast for n in the range of graphical nodes in most UI applications. For everything else, a windowing infrastructure can be used (usually baked into a very sophisticated <ListView> component - our ReactNative mobile applications use this approach (special <ListView> component that exploits immutability without developer intervention)).
I work in this field and have written lots of code, both mutable and immutable, declarative, OO and functional, to solve a variety UI problems. I've also written my own language-enabled editors using multiple techniques (see comment https://news.ycombinator.com/item?id=9117234 for the one I'm working on right now, but you can see https://www.youtube.com/watch?v=SAoRWmjl1i4 for an Eclipse-based one I did in 2007), so I've definitely got multiple feet in the game.
I don't work in Web, most of my UI code was written for Swing, SWT (Eclipse), and these days WPF in immediate mode. It seems like React is solving a bunch of JS/DOM problems, so maybe my experience doesn't transfer, but I've found that for my work, it is much easier to just go with mutable data structures that support change propagation, so you make a change that affects a line in a block (my current editor architecture, block - lines - more blocks - more lines - etc...), the change is just...O(1) because the line can be damaged/repaired directly! So why would I give that up for O(log(n))?
In most applications, we have three tiers of time durations.
Tier One: During an interaction/animation you must update the screen every 16ms. Code may not run longer than 16ms (less in practice).
Tier Two: You are not interacting, but may begin interacting with something at some unknown time. Code may not block the UI thread for longer than about 50ms so that there is a guarantee of not introducing perceivable delays into the interface.
Tier Three: Long running tasks which should be executed in parallel with the UI thread/process.
If going from O(1) to O(log(n)) still allows you to meet your deadlines for whichever latency tier you are targeting in whichever supported device classes you want to support, then it's worth it in order to program with better abstractions. Blocking for 1ms is as good as blocking for 13ms in Tier 1. Blocking for 25.5 is as good as blocking for 40ms in Tier 2. (This is helped by a decent event loop/task scheduler etc).
Again, all of this assumes you genuinely value immutability as a programming paradigm over mutability. If you'd really rather mutate, then you should just be using mutations/observables. I would not rather. Sometimes, I still perform mutations for the most critical parts of a UI abstraction (such as a scroller animation etc) - but I am up front about it being a compromise of what I'd rather do.
Memoization is still more expensive than just tracking changes and avoiding unnecessary recomputations directly (it only starts winning when doing dynamic programming). There is a point on accidental coupling but this is more of a correctness rather than performance issue.
In the high-performance computing field, use of immutable data structures is suicidal; even tries are a magnitude slower than in-place mutable data structures. And he only compares against naked shared mutable state, not against managed mutable state with change propagation.
And that gets to the end of the talk: the real reason is they want to avoid using frameworks that already solve this problem by tracking changes, and React somehow avoids that since you can do just the diff post facto. Ok, I get that.
The problem with this though, is that the entire framework would have to be written around this idea, and everyone who uses the framework would have to use these datastructures correctly. That being said, it will probably be both more efficient and relatively easy to use once Object.observe lands in Ecmascript 7.
In the meantime React works with every datastructure out there, which is a big plus.
Immutable datastructures, while more expensive to change, could be very cheap to diff, because you know if two objects have the same pointer, they're equal.
That is the problem, and one of the reasons why I haven't ported Glitch (works in C#) to Javascript yet (instead, opting to wrap it up in a new language).
> Immutable datastructures, while more expensive to change, could be very cheap to diff, because you know if two objects have the same pointer, they're equal.
Yes, I use this property a lot in my own code; I'm pragmatic and use both mutable/immutable data structures. That being said, I find it easy to trace changes and do change propagation on a mutable data structure (that can change) vs. an immutable one (which obviously require diffing since you can't know exactly what changed). Using immutable data structures actually make that problem much harder from my point of view, but I see its utility as an easy way to integrate with existing code and programming practices (something I can't offer).
