If you're averse to React because of JSX, “mixing templates and views” and similar superficial “best practices”, you're missing out. Engineers embracing React are not dumb. You should consider a possibility that they think it's good for a reason, and that reason is something you should learn about instead of armchair-rejecting it.
Try tuning out your inner rule-of-thumb linter for a weekend and really give it a try.
I've been against mixing logic and views for the longest time and I still wouldn't mind if separation would have been introduced, but recently I gave up.
Either I'll be unnecessary stubborn and miss this awesome new tech or I'll shovel my opinion aside and do try working with this approach.... or at least until someone else introduces new framework that comes with code separation.
The react philosophy of the world is that mixing mutable state and views is bad, logic and views is fine. To be honest this makes a lot of sense to me as a functional programmer.
After learning react something conked in my brain. As I rehashed the arguments against logic in views, I realized that they recapitulated FP's arguments against uncontrolled mutable state, just without the nuance. Nothing wrong with purely functional logic/computation itself. And that delineation feels a lot less arbitrary than "logic in views is bad".
To be clear, I'm not arguing you should implement domain logic in your views. That's a dumb strawman.
But the religious fervor that people have against logic in views is absurd. Reexamine why you think logic in views is bad, and I think you'll find the arguments are actually against 1. poor factoring resulting in ridiculously complex views (and hiding logic is a halfass, crappy way to control complexity) or 2. mutable state.
Are user triggered events for a view part of the view, or something else? I think this is where such a separation of concerns comes in... a view is more than just how it is rendered, it is also events for the view. There are also child components, and handling these events, or encapsulating them into events to be triggered by a given view.
When you think of it this way, it makes sense. From a designer perspective it doesn't, and you have to adapt designs into a view, just the same, you usually do anyway.
I think you've got it exactly backwards. Events are not part of the view.
Events happen to be triggered by the view and they affect global state which is then reflected in the view.
This is the separation of concerns you want. Components don't need to know about events beyond triggering the ones they need. Components only worry about rendering state.
It's too bad that you can't separate virtual dom, the innovative part of React, from the rest of React which has its warts. They should receive credit for coming up with the idea, but there are better, more reactive frameworks using virtual doms now. React shouldn't win simply because they were first and have a big company behind them.
And those singletons will bite you hard the moment you try to do server-side rendering because you will now have state shared across sessions.
The approach that raynos/mercury takes where state is fully decoupled from layout/rendering is the way to go. In mercury, all the render functions are composed together to make one large pure function. You give it the current state and it deterministically will always render the same layout. So much better than the react approach. Furthermore, the complete decoupling from the rendering functions and the reliance on using bijective lenses with shallow copying means you get time travel debugging for free (i.e. undo redo is available right out of the box)
I discovered React after using Om, and I believe that Mercury comes with a lot of good practices enabled by default. But it's technically possible to achieve the exact same 'decoupled-state-pure-shallow-etc' using React. It's just not turned on by default and some additional boilerplate is needed, so it's harder to achieve this in React. The only reason I use React instead of Mercury is that React is battle-tested. No amount of testing can replace the real-world situations.I choose a path of higher resistance because I trust React most (due to it's popularity and usage), even though Mercury is better in theory. Ultimately it's a pattern, and that pattern is associated with a level of abstraction over the DOM as well as the way we write code. Mercury and React are two examples of this. I would like to see more competition and libraries to choose from, modular DOM abstractions, with declarative UI's and alternative API's and jsx-esc transforms. That'd be awesome IMO.
Agree that singletons for Flux are bad, but they're totally not necessary (e.g. https://github.com/acdlite/flummox). And if you don't like Flux, don't use it. I don't see how that's can be a demerit against React, since they're completely separate things.
React doesn't have a good pattern for passing information up the tree other than Flux. Prior to Flux they told everyone to pass functions through props which is fine up until a certain point but it doesn't scale to large apps. Flux does solve it, and is a generally good idea, but the singleton pattern breaks when applied to the server.
Ah, I see. I don't really agree, because I think the enforced unidirectional data flow is one of the best parts of React, but that's certainly a valid point of view.
I just want to emphasize again that Flux is entirely possible without singletons, and works just as well on the server if you create new instances for every request. Flummox does it, Fluxible does it (at least for stores). It's just a shame Facebook pushed singletons and then everyone followed their lead.
Honestly I think FB put the flux pattern out before that idea had become fully developed looking for feedback from the community. The flux repo explicitly states that none of the examples are used by facebook or even resemble facebook (client-side) code. The dispatcher, which was released months after they announced flux, is the only component in the flux they admit to using. Immediately after it's release (even before the dispatcher) we began writing client-side javascript code based on arbitrary examples meant to demonstrate a concept.
I also think flux's real-world implementation came from the necessity to build React components within pre-built apps, where they simply didn't have the ability to pass down props because they had to create complete separate components.
You can use Om if you like, right?
Or Omniscient. Or Morearty.
Not everyone is ready to go into FP-land right now.
React lets you take these decisions yourself.
Whether local state is practical for your team or not.
You must be living in a very different world than mine if you think that compared to its rivals (Angular, Ember, etc), React is somehow “encouraging” local state. Sure you could go more functional than that, but take a look at the mainstream frameworks and you'll see it's such a long way to go, that had React not allowed local state, it would not have gotten adoption at all.
There are going to be lots of different types of frameworks that encourage different patterns all backed by this concept. I'm not saying React is the worst of the bunch but I also don't think it's clearly the best.
All I'm saying is that when people praise React I wish they were really praising what you call "declarative component model".
React is receiving 100% of the attention in this space when imo it should be receiving about 75%.
The "declarative component model" is not an innovation, it's an obvious approach that has been used for ages when prototyping. You can achieve the same conceptual model of your interface by just generating the entire HTML of the page and rerendering it whenever something changes. This obviously doesn't perform and has some edge cases where it breaks things (textboxes, for example), but the idea is the same as react's components.
FRP is based around the idea of declarative components, but a lot of developers have an aversion to it because of the academic aura around it.
React's innovation is using virtual DOM to make declarative components usable without delving into FRP.
>FRP is based around the idea of declarative components, but a lot of developers have an aversion to it because of the academic aura around it.
Not just that. I may be stupid but personally I find it mentally simpler to `setState` than to `flatMapLatest`. To each their own I suppose.
>React's innovation is using virtual DOM to make declarative components usable without delving into FRP.
Precisely.
>The "declarative component model" is not an innovation, it's an obvious approach [..] can achieve the same [..] by [..] obviously doesn't perform and has some edge cases where it breaks things (textboxes, for example), but the idea is the same
React is not an academic paper, it's a tool. It doesn't need to have new ideas, it needs to execute on them in a practical way. Which it does.
The mental model is much simpler because your `render` method describes UI at any point of time. Not just the DOM, but your own components too. On the inside, any component can have its own state, but from the outside, it's not visible, and components using it don't care or know about it.
Separating `props` and `state`, component boundaries, lack of two-way binding and predictable top-down data flow make it easy to reason about where any data comes from, and how UI will change over time.
This is true of all virtual dom libraries. There is no point of two-way binding when you have a virtual dom, and I'm not sure it even makes sense since you have to rerender the tree anyways.
So we agree, you're just separating the concept from the implementation and I'm talking about them as one.
There's a difference between “virtual DOM” and components.
In React, components are not just functions that return their own virtual DOM. AFAIK for many vdom-based libraries this statement wouldn't be true.
React components may have local state (as much as some people hate it, some find it useful), they have a lifecycle, can react to receiving new props with side effects, can implement diff bail-out hook. And you can nest such components declaratively.
The other libraries intentionally left out shouldComponentUpdate because they don't think it's something a developer should have to worry about. I don't know of any evidence to suggest that React is faster because of this feature.
And declarative nesting is a feature of all of the frameworks I've come across, I'm not sure why you think that's unique to React. The advantage feature I would credit React for is the size of its community and influential advocates like you.
WPF (even completely ignoring XAML), JavaFX both support declarative nesting with interesting databinding features to support it. They are based on retained scene graphs, however.
If you are interested in academics, I published such a system at ECOOP in 2006:
JSX mixing presentation logic and views seems to me like they have reinvented PHP. PHP is derided for this, but also for it's shaky implementation and semantics. JSX seems like Good PHP and that's a big deal because PHP is wildly successful despite being awful.
JSX isn't really HTML, though. It's a way to describe React components in a familiar syntax that resembles HTML. It's a far cry from concatenating strings of HTML like you do in PHP.
That isn't declarative in the same way... that's markup. The kind of declarative that is used in React is more like functional programs. You write programs, using actual code, that compute ("declare") exactly what the UI should look like from top to bottom, given each possible input.
This is in contrast to a model where you mutate an existing UI model each time something interesting happens.
Declarative is a vacuous word that means anything you want it to depending on context. We used to call functional programming "functional" and logic programming languages like Prolog "declarative." Then declarative started meaning markup, then declarative started meaning...immutable functional code? In PL, we mostly just avoid the word altogether these days since everyone has a different idea about what it means.
React is closer to an immediate-mode UI model: you write programs that compute exactly what the UI should look like on each frame, rather than mutating a scene graph each time something interesting happens (as occurs in retained-mode UI models). Substitute DOM for scene graph, and the distinction might hold.
I think that the reason people started calling React 'declarative' was because 'functional' was interpreted to mean (purely) functional, with no side effects.
But yes, exactly. React is similar to an immediate mode graphics API. Except that also has weird connotations, because people think of things like canvas that are very low-level: all you get are lines, arcs, and fills. React's primitives are at the same level of abstraction as the DOM, you just work with them in immediate mode, not retained mode.
If I understand correctly, the DOM is retained, and React brings abstracts it efficiently back to a immediate mode API with some state retention, which has benefits since things stay consistent automatically.
In contrast, a UI model like WPF uses (declarative) data binding to achieve something similar, but without as much flexibility and with more verbosity.
I'm working on a system that allows for state retention in an immediate mode model, though wrapping WPF rather than HTML:
I know Andy (former UIKit team) was quoted in the intro thread but I'll do it again:
>I say with confidence as a former UIKit author: React's model for the UI layer is vastly better than UIKit's. React Native is a huge deal.
https://twitter.com/andy_matuschak/status/560511204867575808
If you're averse to React because of JSX, “mixing templates and views” and similar superficial “best practices”, you're missing out. Engineers embracing React are not dumb. You should consider a possibility that they think it's good for a reason, and that reason is something you should learn about instead of armchair-rejecting it.
Try tuning out your inner rule-of-thumb linter for a weekend and really give it a try.