First standalone components to get rid of modules and now signals. There is quite some innovation happening in Angular while being a really mature framework.
I enjoy Angular as much as I love React, Vue or Svelte. It's always about picking the right tool. And when thinking of Angular it's that batteries-included framework that is godsend for big enterprises where any decision given by the outside is something you don't have to battle for.
Often I describe Angular as a hidden champion. It's loved and trusted by enterprises while they are not really talking about it. Plus they hide all the package downloads behind their own npm caching servers like Nexus or Artifcactory.
On the other side there are tons of individual engineers working with React or Vue (for good reasons) and writing blog post after blog post. Maybe that's one reason why folks think Angular is dying? I don't see any signs of that though.
Angular is as little innovative for web frameworks as Firefox-ESR is for browsers. It merely keeps copying features from other frameworks - just many years later. It is a chronically outdated framework that always struggles to keep up with its competitors. It would be ok if those were deliberate design decisions, but if the features get copied some day anyway, what is the point? Why not do it the right way from the start?
For example, this update brings us computed properties, an essential feature for any complex performant web application that was made popular by Vue.js 10 years ago [1]. And now in 2023 we get it in Angular, essentially a confirmation by its devs that its lack had always been a design error.
I also cannot understand the "mature" argument. For example, it took five years for documentation on the integral `<ng-content>` to arrive [2]. This is something I'd expect from the side project of a lone programmer, not an enterprise-level framework.
The only upsides of Angular are its "batteries included" approach and the (debatable) default of RXJS, while the downsides are plenty (see other comments).
What's the point?
If you copy after there is some proof that the features work in practice, you are less likely to implement something that's going to make everything worse instead of better.
Frameworks are experimenting with thousands of new features which might improve something, but this doesn't mean that they do in the end.
People rely on Angular, this makes it important that everything is thought through.
I don't want to become proof of concept tester for new features
> The only upsides of Angular are its "batteries included" approach
This basically trumps everything else though. Angular is likely going to have by far the lowest TCO for any projects with a 10+ year lifespan.
Like, sure, you need to watch 20 hours of Udemy videos to understand how to build an app, but that two weeks of work upfront is going to save you millions of dollars and multiple years of stress eating coffee pastries every morning.
Not quite, its webcomponent implementation is ahead of React's, which is still pretty poor. Its PWA integration has been among the simplest and most seamless I've seen. It's easy to cherry pick examples either way.
Computed properties has always been a possibility in Angular, you'd just create a getter[0] based on multiple different values, works like a charm. This new way is probably faster, but I've honestly never had any issues with the other approach.
The problem I had with getters in a fairly complex application (WYSIWYG editor) was handling mouse events, in particular onMouseMove - change detection was triggered on every such event, and the application was very slow.
EDIT: It was slow because there were many getters based on other getters etc and everything was recalculated on each event.
I don’t think a computed property would behave any better. It sounds like your event handling should have been decoupled from change detection and only applied changes when necessary. This would probably be a problem in any framework.
The concept is actually slightly different. With memoization you store a map from inputs to outputs. A computed property doesn't necessarily do this, it just caches the result of the last computation.
The important part is that a computed property tracks its dependencies and is automatically re-calculated ONLY when necessary. If it behaves differently, it's not equivalent to the concept of a computed property in Vue.
Isn't that just semantics? Whatever name you give the feature "lazily computed and cached values in the context of web frameworks", previously Angular didn't have them (built in) and now they do.
Are you aware what comment tree you are in? This whole thing started because GP wrote this:
> For example, this update brings us computed properties, an essential feature for any complex performant web application that was made popular by Vue.js 10 years ago [1].
So obviously the relevant definition is "what the feature does in Vue". What is your issue here?
Every version essentially breaks the previous one.
90% of the information online is outdated or incorrect. By the time you start a project on Angular X, Angular X+2 is out which makes many things complicated, some deprecated.
It's true that it provides a robust set of defaults and libraries but those are generally an overkill for most projects.
On top of that there's a higher compilation complexity in Angular which I have seen every single Angular project struggle at some point.
> Every version essentially breaks the previous one.
This is simply not true. It might have been true in the early Angular v2 days, but since v6 or so, updates are relatively trivial.
https://update.angular.io has your back extremely well. The trick is to ensure whatever third party libs you're using have also been updated, otherwise you're application may break.
Even if we take it for granted that package downloads are hidden, how should we account for the ~40% drop in new questions in the Angular tag on Stack Overflow since its peak in 2018?
In the Stack Overflow developer survey last year, Angular was 52% loved versus 68% for React and 63% for Vue. Can we find a way to dismiss this as well?
On the freelancing platform Codementor, based on rough sampling, there have been about 338 Angular requests in the past year versus 2098 for React (6x).
Sure, Angular won't die outright because the enterprises that have adopted it will need ongoing support, but this seems like a weak metric for growth. For enterprises selecting a tech stack now, choosing Angular chops the hiring pool by a significant factor. I don't see evidence of advantages Angular offers overcoming that.
On the other hand, React has gained new life through frameworks like Next.js and maintained relevance. React's popularity is not just individual engineers making a disproportionate amount of noise on blogs. React hooks are the most established unit of construction for making components at this point across web, mobile, desktop and command line. That won't last forever, but there are too many signs of Angular trending out and React trending upward to dismiss.
The entire library is so spread out, with so many half-assed features.
When I wrote Angular I spent 2 weeks on getting something similar to `props.children` working. <ng-content> didn't work if the elements were dynamic. Outlet system was way too verbose.
The 2 way databinding is anachronistic and belongs in 2010.
The i18n system they shipped with is awful. No way to have translations in code, only in template. This breaks down very fast when building generic components.
Because of the opaque CLI I've spent 2 weeks figuring out why every third dev build resulted in a 150 line long tsc stacktrace, and it turned out to be angular's 1.x types package. Would have been so much easier with a webpack configuration.
The props.children/slots issue is one of my biggest problems with angular. It’s really hard to use composition like you always do in react.
Another thing is, that angular components always wrap their template into a block element (like a div) in the DOM. So if you do some refactoring and split up one big component, the result in the DOM is different and your CSS may not work anymore. If you are using flexbox or grid, it may even be impossible to split the component.
> <ng-content> didn't work if the elements were dynamic
Usually if you run into problems like that, you should use templates.
Honestly, I'm not sure why `<ng-content>` exists vs templates. I'm sure there's a reason (verbosity?), but templates have way fewer caveats.
> Would have been so much easier with a webpack configuration
I agree. It's not hard at all to do a Webpack configuration of Angular. Just add the Angular plugin. (That's what the CLI does under the hood.) I wish they document it though.
This has been a debate since long before React or Angular existed. Keeping code and templates together typically makes sense to programmers while keeping the template as html in a separate file means you can edit it in any html editor, no special tools needed. I've always been partial to keeping them together, but working with designers who prefer it the other way, I can at least respect where they are coming from.
> It uses invalid HTML, with attribute syntax including brackets and parentheses
Brackets and parens are valid in attribute names.
Section 13.1.2.3 of the HTML standard[0] states:
> Attribute names must consist of one or more characters other than controls, U+0020 SPACE, U+0022 ("), U+0027 ('), U+003E (>), U+002F (/), U+003D (=), and noncharacters.
It's my answer to what I can tell you agree is the most important question in software engineering - "What programming language has the shortest 'Hello World' program?"
We're talking about a framework here, not a programming language. I don't think the above poster's assessment about the amount of files needed for hello world offers a charitable conclusion either, given frameworks are not languages. The same file usage would apply to many other popular 'heavy' opinionated frameworks.
There’s a grain of truth in that thought. First impressions matter, frameworks aren’t exempt from that basic psychological fact. If you treat ’complexity of hello world’ as a proxy for learning curve steepness (again, people do things like that subconsciously), you may end with a very wrong assessment very early. It’ll probably tell more about the assessor than the framework but it’s a real risk.
> In my experience people that hate Angular don't understand why they hate Angular.
One of my major annoyances is the templating language. Completely unnecessary and over complicated, compared to JSX. Just another weird syntax to learn with its own arbitrary limitations and quirks.
I think it’s a good indicator of the overengineering surrounding Angular overall.
I see it as the opposite. JSX is a convoluted mess, and does not make for easy reading due to being mixed in with everything else, while Angular promotes clean separation and promotes a template language like many others. It's usually React codebases that I've seen overengineered due to lack of features, while Angular codebases are very same-y - that's a good thing from maintenance perspective.
Obviously there's some matter of taste, but JSX is simpler in my mind, because it's just JS with a tiny bit of syntactic sugar to have nicer function calls for components. That's it. Anything you can write in JS, you can write in JSX and it works. You can debug and use all the tools you'd use for regular JS.
With Angular templates, there's an entire new language, with its own weird syntax ngIf/else, ngSwitchDefault, let-*, pipes and other components that get pulled in based on whether you forgot to set up correctly or not in modules far-far away somewhere in the codebase, brackets of different sorts having different functions and just a lot more stuff to remember. And if it doesn't work, you're gonna get a silent error most of the time with no easy way to set a breakpoint to investigate.
I also don't believe there's any real separation between the HTML templates and the Component class, just because they're written in separate files. They're just as tightly coupled as any regular React component, just harder to follow. I don't think I've ever seen an Angular component, where you could swap out the template, without making simultaneous changes to the Component to accommodate it, which should be possible if they were truly separate.
I haven't had the chance to work with either much, but at the end of the day, a new new syntax just for doing fancy if/else/for seems unnecessary. I believe Angular's hand was forced since JS at the time didn't have as ergonomic syntax as it does now, with fat arrow functions, .map and other niceties.
Having mutability/2-way-databinding makes angular hardly benefit from using TypeScript. State is just so much pain and it is hard to make impossible states not representable in the types.
Angular requires a lot of extra steps to get the templates properly type-checked (and you even need an editor plugin for that). I can't even name which non-standard entries in the tsconfig are necessary to have all templates checked entirely.
Most people use the cli to generate new components. Where in other frameworks, you just create a new file, maybe use a snippet and call it a day, the cli yields a test, scss, html and the actual component code as separate files.
Having worked with angular and react for 5 years in parallel, I'm pretty certain that I'd never touch angular again. The DX is so complicated in comparison.
Despite forcing you to use TypeScript, angular is the framework that benefits the least of it.
I like Angular, but approach of doing the same thing in 3 different ways (promises, rx and now signals) make codebase more complicated and adds additional brain load.
A 3MB JavaScript runtime blob right out of the gate? ("Only" 275KB needed just to first render, like that's laudable on its own.)
Other frameworks have already loaded dependent REST/GraphQL data before Angular has even initialized.
Take someone on a mobile device and put them at the edge of cell signal range either because they're out an about or stuck in the bowels of a building with thick walls. Suddenly 275KB-3MB matters a lot more and the user is staring at a blank screen for 10+ secs.
Then there's the boilerplate, all the metadata you need to add to each component that has nothing to do with your problem domain.
And then there's runtime speed compared to alternatives.
While I'm truly happy to see Angular incorporate signals, now devs need to learn when to use signals rather than rxjs. I know now that it's the difference between needing async or not, but if you go down one path, it's a PITA to tear it down and rewrite for the new path.
THAT's why I don't like Angular, though I'd happily choose Angular over React and Vue, both of which are a prone to reducing codebases to big balls of mud due to their "flexibility".
How about the extra 30 seconds to couple minutes for CI/CD to make a production build for static deployment? V16 added the ability to put esbuild/Vite into the mix, but it still builds WAAAAAAAAY too slow compared to modern alternatives, especially if you're on v15-.
ng serve for development is a pig too. Save-reload cycle is far from instant.
Signals are a step in the wrong direction. They provide extremely little value over RxJS Observables (ex. avoids glitches when you have diamond-shaped dependencies), but you sacrifice so much in return, like control over async operations.
The API looks like it wants to attract React devs, and I fear that it won't accomplish that but instead it will drive away the remaining Angular devs. In no time we'll see SO spammed with questions using `computed(async () => ...` or race conditions when using `effect`.
I like Angular and I think it's still a solid choice for new apps in an enterprise setting, and signals are a welcome addition that solve real problems, but I don't understand the hype. Signals are as old as JavaScript if not older. The diamond problem and sync propagation are not new problems. We've come full circle. Old problems become new problems but with the same solutions. What did we gain with all of these iterations if we landed back on what was essentially the Knockout model of 2010?
I'm reminded of papers such as "Deprecating the Observer Pattern (2010) [1]" which documented solutions to these problems a long time ago.
I guess every large ecosystem has to rediscover certain facts… same thing with yaml and xml on the sysadmin excuse me devops side, nosql getting transactions and scans… everywhere you look there’s a bunch of people who like doing more than reading, and also doing gets rewarded much more than reading. The outcome is inevitable. Can’t wait for the next cycle, really curios what we’ll rediscover next.
I don't think I understand signals. What's reactive about them? I have to wrap every primitive in a custom object and I have to call setters for every primitive and I also have to define a function for computing derived values. None of this is reactive, these are just React hooks with 3x noisier code. The examples in the page imply you don't have to write dependency arrays. These were never a painpoint for me, but per their RFC:
> It is possible to, optionally, specify an equality comparator function. If the equality function determines that 2 values are equal, and if not equal, writable signal implementation will block update of signal’s value and skip change propagation. The default equality function compares primitive values (numbers, strings, etc) using === semantics but treats objects and arrays as “always unequal”.
So just like React, optimization is opt-in, but worse -- React transitioned from `shouldComponentUpdate` to dependency arrays for a reason. Hell, Python devs learned this lesson 15 years ago with the transition from comparison functions to `key` lambdas.
What am I missing?
Sidenote: the second sentence from that quote made my head spin. I've read it three times and still not sure if it's a mistake
I think I also don't understand signals. I expected something like signals/slots but this is something different. I read the example a few times to see if I was missing something, but I don't understand how this idiom fits into the larger picture of writing a component. What problem does it solve?
I don't get why, Angular is really nice and has so much included and is well documented and no need for reflux and redux libraries that might do or not do what you need
I worked a bit with AngularJS. My, definitely over-exaggerated, feelings are that it was a hideous abomination. It's been long enough that I can only remember how awful it was to have to register everything with the right strings. A lot of negative sentiment towards modern Angular is probably a holdover from that.
Modern Angular on the other hand just seem full of pointless boilerplate. From a personal perspective, the way it's typically structured just feels verbose and clunky compared to React.
I hear this critique a lot, but what exactly is that pointless boilerplate? When I write a React application, I have to use a lot of React-flavored boilerplate too... useEffect, useState, useThis and useThat... What makes this so different from the Angular TS annotations and lifecycle hooks?
* Angular classes just feel a lot more verbose/messy than the equivalent functional implementation in React.
* Perhaps it's just familiarity but I have a much easier time understanding what a functional React component is doing just by reading it from top to bottom.
* I have an irrational hatred for separate html template files.
* I'm also not a fan of templating via custom html attributes.
* Two way data binding is sometimes convenient but feels messy.
Pulling a random hello world comparison that starts from a base template from google:
React: 16 lines - 342 chars - 2 files changed.
Angular 34 lines - 808 characters - 3 files changed, 4 if following common html template practice.
React
HelloWorld.js:
import React from 'react';
function HelloWorld() {
return (
<div>
<h1>Hello, World!</h1>
</div>
);
}
export default HelloWorld;
index.js:
import React from 'react';
import ReactDOM from 'react-dom';
import HelloWorld from './HelloWorld';
ReactDOM.render(<HelloWorld />, document.getElementById('root'));
I can definitely relate. I started using Angularjs around version 1.1 and remember being completely flummoxed at the error "Unknown provider: $providerProvider <- $provider". Debugging that left such bad taste in my mouth that I probably will never try Angular.
Angularjs did get better after 1.5 (when they added components and one-way binding), but IMO the model is still fundamentally broken. Vue fits my mental model much better and is much better documented; if I were to start a new project today, that's probably what I would reach for.
I don't know about others but the move from AngularJS(v1) to Angular burned that bridge for me.
Developers have long memories. And the fact google keeps burning good will with other shutdowns isn't contributing.
Besides that Angular is also a heavy framework (gives a lot, but requires you to do things in a very specific way) where hacker news crowd has usually a preference for unintrusive frameworks and ones that give a lot of freedom and leeway.
Look at this way, angular 16 is now introducing signals, vite support and improving standalone components.
Now, let’s compare to Vue, Vue has Standalone components, vite support, signals since v3. And fine grained reactivity since v0.
So, one must ask if here things are so good, why stick with angular and wait for everything to stabilise when you can jump ship to Vue now?
Similar argument can be made for react, though I admit hat Angular to react transition won’t be as easy.
And this is in context of newer advancements only. When you take a step back and look, you will see many cases where certain angular issues are only in angular. While rest of the js frameworks are converging
I wouldn't call reactivity in Vue "fine grained". Every object that gets loaded into component state becomes reactive, and every piece of code that touches it gets dragged into a massive internal dependency tree.
Vue has blacklisted reactivity, everything becomes reactive (and observed) by default and you can opt-out, while MobX features whitelisted reactivity and is much more fine-grained
I say fine grained as changes in an object will invalidate only that object, and only it's dependencies will be updated.
Contrast with coarse grained reactivity I. E. angular, where it must dirty check all "reactive" objects to identify the changed object. As angular can't tell exactly which properties changed.
How do you find the upgrade path. It's been a long time since I was working with Angular 2. I felt a little fear at the pace of improvements. It seemed that for a long lived application I'd be forever spending time upgrading the framework. Is that what it's like or are the upgrades reasonably easy?
I feel like there's a real major upgrade every 3 years, which is reasonable given the pace of web development itself.
The "Ivy" renderer was a big one, but it was introduced in 9 and you had until 12 to transition. Still, these were like 5% of what you had to do from AngularJS to Angular2+.
That's only the rendering, though. If you used other libraries (that's already included in angular) you wouldn't be able to say the same. For instance if you use routing in react, you've probably have had to rewrite your app 6 times by now.
Angular has been pretty stable since the v4 release. I recently did an upgrade of a large enterprise app from v8 tot v15... Jumping 7 major releases, no problem with the migration tool.
They’re different things, though. React has a lot less going on. If you want something with a stable API for the tasks React is suited to, it doesn’t make sense to throw Angular at that job.
I don't think that's quite true anymore. As someone sarcastically remarked on Twitter [0], React has patched native events with synthetic events; has rolled out its own scheduler in order to achieve the suspense effect ("patched the event loop"), and is going to patch the native fetch for reasons that I am not clear about, but that are doubtless also suspense-related. With hooks, it introduced a syntax with a set of implicit conventions that, confusingly, still looks like javascript, but has significant differences from how javascript works ("the rules of hooks"). React provides its own way of lazy-loading modules; its new server-side renderer currently assumes that when React rehydrates, it takes control of the whole DOM tree; and the docs are now strongly recommending people to use full-fledged frameworks (Next, Remix) when starting new projects. React has long outgrown its "it's just a library" stage, sadly.
Yeah, the most WTF thing I've seen React do is to mess with console.log by silencing them (WTF #1), which was particularly appalling since React did it only because it run the rendering twice under StrictMode (WTF #2), and only during development so you're not sure what the real behavior will be until you run it on production mode (WTF #3).
Messing with the console is such a no-go that I'm appalled it even passed the drawing board, making the "React isn't a framework, it's a library" argument even more bonkers than it already is.
Oh, I totally forgot how mad I am at React because, with the new hooks-based approach, it has removed the semantic of "do this side effect just once when a component mounts". Mounting a component is such a crucial part of the life cycle of any web component that I am amazed that the React team does not have a special way of handling this; or that with their new architecture they are not guaranteeing how many mount-unmount cycles a component will go through.
This all makes me want to move on to web components; but of course now the project that I am a developer on is locked in React :-(
Certainly, I don't think React is a trivial library. It's an absolute beast given its narrow scope of application. I suppose my point was that React's narrower scope means it's easier to keep a stable API compared to Angular which has a much broader scope. But maybe you weren't suggesting that's incorrect, and just wanted to point out that React is a rat's nest now (which is totally fair, haha).
My goto example for RxJS are event handlers. Heres an example that implements drag-and-drop (for which the vanillajs / d3js versions aren't nearly as pretty)
It basically translates to
1. Wait for a mousedown event
2. Start listening for mousemove events
2a. On each mousemove event, do something
2b. When a mouseup event occurs, stop listening for mousemoves and return to 1.
More specifically I've found RxJS makes simple one-off calls (eg API fetches) ceremoniously painful (do I need to unsubscribe? do I get the value sync-ly or async?),
but anything that involves multiple values (eg websockets, polling, event handlers) gets much cleaner.
My biggest complaint with RxJS is probably the dev experience though. It makes call stacks / the debugger useless so you have to get comfortable with console.log's everywhere.
Which is one of the reasons I avoid chaining observables and filter()s in general.
RxJS, more specifically its primitive Observables, fills a different space from Promises. For simple cases where only one value is emitted Promise probably works better and is more intuitive, but there are certain things that Promises cannot model - specifically when there is a stream of events over time. RxJS is more complicated because it needs to model more complex things. I do think Angular made a big mistake overusing it in places where Promises would have worked perfectly fine, but when you do need it it does work pretty well.
> I do think Angular made a big mistake overusing it in places where Promises would have worked perfectly fine, but when you do need it it does work pretty well.
Was it really a mistake or were they just using it before native promises were widely supported?
Yes, RxJS is extremely hard to use. I developed on a lot of different technologies, but RxJS is one of the things I never seem to get a hang of.
It’s not just the syntax (so (many (brackets)) and => lambdas), but it’s also really hard to create a mental model for what’s happening.
I know that react hooks are not a replacement for Rx, but really often you can program similar functionality with hooks and it’s just 10 times easier to read.
If by unintuitive you mean “Can I look at it without training and know what’s going on.” But the problems they solve are also unintuitive. Only, those issues can be buried under a tangle of nested callbacks that look correct, intuitively.
Once you take the time to really (and I mean really) understand RxJS it’s incredible. Write your own Observables. Write your own operators. Learn what the built in operators are really doing (subscriptions, etc) reimplement them yourself.
I've seen components rewritten using observables to be sometimes a full 1/4 of the original size.
The biggest thing for me is the realization of the "source of truth". With observables, if they are set up correctly, you can see the exact bits of information that some resultant stream are dependent upon. Operations then can have a single action because all of those sources can be muxed together rather than having several calls throughout a component to doSomeAction().
I have such an issue with this response because it just seems like “you’re holding it wrong” a la the iPhone “scandal” where apple attempted to blame an engineering flaw on its users
I would imagine most people’s use case (mine certainly is) for RxJS boils down to “call a JSON API and receive a response”. That shouldn’t be a hard problem
Imagine if someone complained about the complexity of Git and the answer was to “write your own DVCS”
The entire point of abstractions is that I don’t need to understand what’s going on underneath them
1) You skipped points one and two in my comment: a) Learn the foundations of RxJS. b) Understand how to extend RxJS by creating operators. Which lead to c) read RxJS’s code to understand how it implements operators to better learn to implement your own.
2) RxJS is not Git. It’s a library designed to be extended by users. To use RxJS you need to write code with it. Git is not exclusively a library and doesn’t require you to write you own extensions to use it day to day.
As a casual Git user you wouldn’t get much out of implementing a DVCS but you would benefit from learning to host your own repo instead of relying on GitHub.
If someone was struggling to use Git I’d tell them to learn the foundations, practice them, and then to host their own repos to continue learning.
The same basic track I recommended for RxJS.
A big moment for RxJS users is when they realize that they need an operator or observable that doesn’t exist. It’s really fun to write your own.
I think the most bad rep for RxJS comes from using it when it is not needed.
Parent comment said:
> But the problems they solve are also unintuitive.
Do you consider calling a JSON API unintuitive or complex? If not, then you may be using the wrong tool. If you need nothing else, you are perfectly fine using a promise.
If you need to await extra requests, transform them, and react to other events then you need RxJS. For a simple call, you do not.
> I would imagine most people’s use case (mine certainly is) for RxJS boils down to “call a JSON API and receive a response”. That shouldn’t be a hard problem
Do you consider the following code hard to understand or are you are making requests in a more complex way?
```
this.network.get('<url>').subscribe(response => <do whatever you want here>)
```
Even if we agree to disagree that the above code snippet is hard to understand, you can just convert it to a promise:
No, that call isn’t difficult. What is more difficult are examples given on things like Angular University, where there are pipe, subscribe, catchError, among others, in a single call chain. It’s not obvious to me at all what the order of execution is in this call chain for instance:
If you look at a snippet of code in a language you don’t understand you wouldn’t call it intuitive. Once you learn the language you might see that what what was unintuitive before is intuitive and idiomatic now.
Once you learn how RxJS works examples like the one anbove anre intuitive.
Angular 2’s most egregious crime is that their tutorials try to make it (and RxJS) seem “simple”. They aren’t. They’re powerful.
Very much agree, and even for the more complicated use cases Saga's are much cleaner and I would argue easier to understand. Article below goes through a few of the differences.
I worked for 4 years intensively with RXJS. I love the concept, but it quickly becomes a mess in readability when trying to merge different data sources.
I feel like most of the RxJS functionality doesn't have a real use case. At the end of the day, you are doing an async call, something axios handles just fine. I did a bunch of presentations on RxJS, so I understand it fairly well, but I have never used it past basic API calls "irl".
> I feel like most of the RxJS functionality doesn't have a real use case.
I disagree. I have implemented several use cases using RxJS which would have otherwise been significantly more difficult to solve. (Think timing parallel sequences of animations und other UX effects etc.)
I’ve rarely seen it used beyond very, very basic things in Angular projects (and I’ve always wished I saw more of it, actually). Maybe we just work on totally different kinds of projects.
I am the guy who doesn't use it, but people used to recommend using it for "event-driven architecture on the front end" - clicking a button is an event, typing text is an event, hovering over something is an event, and there could be multiple subscribers. The example use-cases people give are really contrived. The functional crowd also pushed it. Again, to me it's another trend like Web Components that never really took of in the mainstream. Yes, people use Web Components and RxJS' full power, but it's not something to learn when you are looking for practical, widely applicable skills.
If anyone still cares about this, here is a helpful site that you can understand without knowing anything about RxJS:
https://rxmarbles.com/
The biggest benefit, which I haven't found much of a practical need for, is you can continue to emit 1...n times, whereas a Promise only resolves once. Since it's an indeterminate number of times, you need to subscribe and unsubscribe to prevent memory leaks, so it just adds more overhead. Use a Promise if you only need to resolve once.
I'm not sure if you have spent much time in a functional language but I found this helped me understand what is going on with a Promise and an Observable. They're all Monads. Just like List and Map or the optional Monad.
When I first started learning that sort of thing it made no sense at all. But now it's just a design pattern that allows you to organize things. Having that understanding helps abstract away some of the mechanisms in the observable.
Indeed! The concept of Monad is simple and very useful; once you learn it, you start seeing it everywhere.
JS Promises are not completely monadic though, because a Promise that resolves into a Promise is always effectively auto-flattened into a single Promise. But most of the time other monadic properties of Promises can be used despite this.
RxJS really needs a better debugging experience, I would love some visual representation (like a graph?) of all the events and how they relate to each other (who is waiting what, what has been waited, what was canceled by whom, etc). I tried a few plugins and browser extensions that supposedly do that but they all crashed, in their defence it was probably because it was a huge monolithic app.
I haven't used rxjs specifically a whole lot but I am a fan of the general declarative pipeline style that it promotes. A different JS library that I've used extensively that promotes a similar style but for more general purposes is Ramda. It is a very different approach to programming than is common so I understand not finding it easy to grasp. In my experience, the benefit is generally that there is less clutter in your code and the dataflow that is happening is a lot more explicit which helps me build a mental model of what is happening a lot more quickly.
ChatGPT is actually pretty good at exploring these different styles btw. It is pretty good at taking one code example and implementing it in different ways when prompted.
RxJS _used to be_ more intuitive, IMO. Around version 5 or so the API changed to use `.pipe` instead of simply chaining operators. Also, some functions were renamed to be (unintentionally, one assumes) more opaque. Such as `flatMapLatest` -> `switchMap`.
Anyhow, there's definitely a learning curve but what it provides is very powerful. It's unfortunate, IMO, that the API has become less approachable over the years.
Although I see how chaining might feel more intuitive, pipeable operators allow for a lot more intuitive and readable composition, making it so you can create reusable and easily tested operators, or even sets of them.
It’s great for what it’s designed for, but you’re not wrong either. Composition can be hard to manage and navigate where a chainable API can be easy to follow and interpret by simply reading the flow.
I think RxJS is actually amazing and I like the path they chose, but I see why it isn’t more widely adopted too. It isn’t nearly as intuitive as it maybe could be.
There's definitely much to be said in favor of simple function composition.
However, I found the API choices odd because in RxJS the `.pipe` operator is, itself, chained rather than being standalone. Using a standalone `pipe`, such as via Ramda or something home-grown worked just fine, so I found it very strange that they mixed the functional composition style with the object-chainable style.
> I see why it isn’t more widely adopted too
I agree that RxJS is amazing, which is why less adoption is unfortunate. So many projects would benefit from RxJS or something like it.
Have you tried `async` pipe and `takeUntilDestroyed`? I only ever `subscribe` if I need to trigger side effects, like setting a form validator based or a value (probably can be replaced by async validators) or modify DOM outside of the app root.
The problem with RxJS is that it feels like Node did in the beginning. It's hard as eff to figure out where the system goes, what happens when, in what order etc.
Anyone else run into an Angular 15 styling nightmare when upgrading to the latest Material Design (MDC) release?
I attempted to use the migration tool [0] for upgrading an in-house LOB app, but unlike previous upgrades client hated the resulting UI - understandable b/c of major styling regressions everywhere - so downgraded back to 14.
Angular 14 LTS support ends in November [1] and as best as I can tell, if we want to stay on a supported Angular release we're forced into a lot of custom styling work users simply don't care about.
If there's some way to stay on the "legacy" [2][3] components on Angular 15 or 16, would appreciate any links/hints. Cheers
I maintain two Angular applications which leverage Angular Material and where dependencies are usually kept up to date.
The first one is my tiny pet project - https://github.com/Klaster1/timer-5 - that I use daily. Updating to MDC components was straightforward and style changes did not cause much trouble.
The second one is a moderately-sized enterprise app I work on as an employee. Every single component update introduces visual regressions the team had to coordinates the fixes for with the UI designer. We split the workload by similar component types, largest pain points being buttons and form controls. Total estimates are in 30-50 hours range, we plan to chip at the task bit by bit until Angular Material 17 arrives, where the legacy component are to be removed.
On a side note, migrating to Ivy-enabled dependencies was on even larger time scale as dependencies had their own breaking changes we spent a ton of effort on, especially Chart.js 2->3 and ag-grid 26->29.
It is not well documented in the Angular Material documentation, but Angular Material 15 moved the "legacy" CSS to a new file.
This means that, depending on whether CSS or SCSS is used, the legacy theme has to be added to keep using the legacy components without styling issues.
The legacy Angular Material components might be removed in a later version, though.
I eventually ran out of time for my toy application and just left it with styling regressions. I didn't understand the advice I used to get styling before, and the migration instructions left a big gap in understanding around what styling was expected to hang off of.
I guess SolidJS should be very flattered! Signals seem very inspired by it. Overall I'm excited by these changes. Rxjs is great in many ways, but it's always been a bit awkward in Angular (e.g. Components hide that Inputs are Observables, so you can't just pipe between them).
There is a lot of discussion here about how Angular is a good choice for larger enterprise applications, but as somebody with more of a "backend" set of experiences, it seems to me that Angular might be a better choice for smaller teams as well?
My struggle is that when I want to build even a simple frontend with React, if it's been a while since I've done any frontend work I find myself having to re-learn what the current set of libraries / tools / frameworks I should be using with React is. For example, if I want a three page simple website, how do I handle routing, is it still `react-router-dom`?
1. Google is the main corporate sponsor
2. Google is funding both Angular development and Flutter development
3. The Flutter team keeps throwing serious effort behind their web deployment target. It's not at 100% yet, but they're getting there.
4. Google internally saw returns from migrating cross-platform UIs to Flutter
5. Google is well-known for being trigger-happy on cancelling projects
Pure speculation, but what are the odds that Google starts to transition off Angular, thus killing Angular? And therefore, why should people start new projects on Angular?
Every project in google uses angular. Probably the most commonly used tech inside Google is Angular. So rest assured that Angular will be there forever. On the other hand I doubt flutter's future, mainly due to its reliance on Dart, Dart was extremely hated inside Google few years ago. There was even AngularDart at some point.
When Dart finally gets around to releasing their updated web and JS interop support later this year it is almost certainly about to become my new default there too.
Google Ads is the reason why Dart was still around when Flutter came to be in first place.
After Dart's failure, Google Ads had just migrated from GWT into AngularDart, they weren't going to go through yet another rewrite, thus they rescued Dart from being fully rampeddown.
It is also the reason why AngularDart stayed around, when the Angular team decided to adopt TypeScript.
Hopefully ngcc being dead makes build/CI times go down. It's real bad :(
(Also, hopefully in the next release or two their esbuild based builder is good enough to use for everything. these 30 second+ reload times when almost nothing changed are the worst.)
Thank you for the updates, Angular. I don't like working with you at all but your updates kinda assure me that at least I'm not working on legacy software.
Computed properties, Stand alone components, “signals” already exist in Ember in some form to prevent constant recomputing of getters, route data being available in components directly
I just wish Angular would copy Ember Data. I worked in Ember apps for eight years and I’ve been in Angular for the last year… it’s the biggest, most annoying missing piece.
The example includes:
`effect(() => console.log('Name changed:', this.firstName()));`
But the text says "We also declare an effect, which callback will execute every time we change the value of any of the signals it reads — in this case fullName"
I've slowly come to enjoy Angular, although the templating definitely feels clunky coming from React.
My biggest gripe is the lack of nice component libraries: Material just looks hideous, NgPrime has ads in the documentation and generally not well supported.
I have to say v14 was already interesting with its introduction of typed forms, but this brings improvements I didn't even expect.
For better or worse I've been largely stuck with Angular since 2017 (and completely since 2020), and it's not my framework of choice in private projects, so it's all the more exciting to see it catch up at least in some respects.
The other day I was wondering if Google still has the technical chops to introduce a new, modern framework to replace Angular, but perhaps that won't be necessary anytime soon.
I think the top issue in React-land these days is definitely that reactivity is not provided as a built-in. I agree that lots of libraries have achieved pretty decent DX in that regard, because the hooks API is actually very powerful and reusable, but I wish the React team wanted this for us all.
Congrats to Angular team for delivering what sounds like a pretty thorough reactivity implementation. I think their opinionated and provided-by-Angular strategies make it a tempting option more and more over time, even though I enjoy my React stack.
How many apps actually benefit from reactivity, i.e. 1,000 people can post to a feed realtime and the app fetches data from 200 other backend services? I believe for vast majority apps, a basic HTML/JS/CSS is all one needs. It's fairly trivial to make your own SPA in ES7.
If you don't know the use-case for reactivity and are going to advocate for "vast majority apps" being authored in "HTML/JS/CSS", I don't really understand why you're debating me.
Frontends - that are clearly more complex than the ones you use - are state machines managing asynchronously-accessible state stored in external services. Due to the nature of how many resources an application might use, it quickly becomes cumbersome to map out that state machine manually, and it is more beneficial to just write declarative display logic that will automatically execute when the necessary preconditions are present and a change has occurred within the state management.
Now, mind you, nothing I've said really involves reactivity yet; and, even worse, most solutions - especially your custom SPA - will rerender the entire tree of objects relying on each changed resource in order to accomplish what I listed.
The beauty of reactivity is that you as a developer simply write modular code - whether it's state or view, the more modular the better - and then your system ensures changes to the underlying state tree flush updates to the view modules attached exactly to the parts of the state tree what changed. It's optimal performance by default, you run the least amount of your own code each time a change happens.
Of course, it brings its own challenges and issues, but it really means that your general answer to "how can I scale this UI" is just "write it in smaller chunks", and that scales pretty dang finely across a team.
Unlike a hand-rolled ball of JS. Come on folks, it's 2023, put your tired arguments away.
Maybe you can consider that on HN you are talking to experts who in their lives experienced dozens different UI paradigms and frameworks including deep experience with React, its ecosystem, its benefits and its warts. I am pointing out that the problem of "how can I scale this UI" is experienced by a very small percentage of apps. For Meta or Uber it makes sense as they need to coordinate overlapping real-time feeds where dependency graph is insane; why should however most apps bother jumping on this paradigm? There are complex yet performant apps like draw.io that never touched anything reactive and aren't missing it in the slightest.
> Maybe you can consider that on HN you are talking to experts who in their lives experienced dozens different UI paradigms and frameworks
Maybe you can consider that on HN you are talking to people who spend their working lives answering the question of "how can I scale this UI" and think that reactivity is a good approach.
> why should however most apps bother jumping on this paradigm
I stated pretty clearly the benefits above - it has great performance in general due to minimal redraws, and those minimal redraws are accomplished "automatically" / via runtime construction so you don't even need to think about optimizing accesses, it's already the least possible.
More to the point, I don't see why we shouldn't laud Angular for delivering an opinionated reactivity implementation when their opinionated, batteries-included approach has enabled a lot of teams to hop in and build apps.
I also don't think you have any fucking clue what goes into UI development if you propose rolling your own SPA, so there's also that.
That might be true but React has the same issue with larger teams and everyone using a different part of it; the pace of development drops rapidly and apps become complex monsters for little reason.
> Since we introduced Angular in 2016 it has not been possible to get a compile-time error if you don’t specify a value for a specific input. The change adds zero overhead at runtime since the Angular compiler performs the check at build time. Developers kept asking for this feature over the years and we got a strong indication that this will be very handy!
HTML templating is not productive as you might think.
In reality, you want control on every code you write. You can't put a variable in the HTML , which forced you to put those elsewhere, which hurts productivity.
Important new features. But in my opinion too little, too late. To me it feels like Angular is slowly dying. No idea though, why so many companies still use it for new projects.
> No idea though, why so many companies still use it for new projects.
If I had to guess, it has a mature ecosystem, it is still actively developed, and it has already been "battle tested" by various companies. Angular is a "boring" choice, and some people prefer to stick with boring solutions than exciting, new solutions.
It seems the alternatives are trying to catch up to Angular. The entire react ecosystem's embrace of Next is proof that people want a framework and not "just a library" for a lot of their needs.
I prefer Angular as I much prefer spending my time building rather than wading through open source bs finding what works, what plays nice together, what others are using. Version management is as simple as "ng update".
People always say angular comes with „batteries included“, but honestly I don’t get it. If you take vite, vitest, react, react-router, react-hooks-forms, react-i18next and axios you have more or less all the angular features covered. You don’t have to „dig through bullshit“ to do that.
Unless sarcasm, your comment actually explains the part you don’t get quite well. If you pick React for a project, how do you know to pick the other 6 dependencies you listed?
No, it’s not sarcasm, just install the listed dependencies and you are good to go.
For forms and validation you can use the mentioned react-hooks-forms. MUCH easier compared to reactive forms.
I never understood the benefit of angular dependency injection, because of its very limited singleton only approach.
React-i18next is just so much better than all the i18n solutions for angular. If you want to be able to switch the language on the fly, you anyway need something like transloco for angular.
Pipes are just functions for templates, no need with jsx. And directives can be replaced with composition. Much easier and powerful.
Angular material is really dated, look&feel is from 10 years ago, check out the current material guide, and compare it to angular material. And it’s just way too hard to customize it, and it doesn’t come with a lot of features. And technically it’s not a part of angular. There are a few good commercial angular UI libraries though.
I don’t think you understand. For someone who isn’t familiar with the React ecosystem - how would they know to install any of your mentioned dependencies.
Forget which solution is better, apparently there is a subset of users who choose Angular purely because they want features built in & not to go searching for additional dependencies.
My experience with libraries built by the community. Libraries for react have much higher quality than angular libraries. I usually fix the hard bugs in our team, and therefore read a lot of library code on GitHub. Usually react libraries impress me and angular libraries are often rather underwhelming. Highly subjective opinion though.
I don't have much experience outside of Angular ecosystem, but I share this sentiment too - the framework might be good, but the community size and quality also matter a lot. A similar example would be Sublime Text vs VSCode - I use the former, but VSCode has so much more vibrant community and everything supports VSCode by default, unlike ST where you are lucky to get a half-working prototype.
First of all, if you want to compare actual growth, you'll need a logarithmic scale. Second, if you add in alternatives like Vue.js, you'll realize React dwarfs both Angular and Vue.js, yet I don't see anyone on HN claiming "Vue is dying".
I agree. To answer your question it's really just companies sticking with what they know. Just like how a huge portion of the internet is still built on PHP.
> The number of people wanting to keep using Angular has been falling for years. But they are forced to by some external control.
Exactly my experience. I often ask why angular was chosen for the task, and usually the answer is: because the architect put it on the slides, and because we hire angular developers.
Agreed. Stewardship of the angular ecosystem is lacking. Versioning across first-party @angular packages isn’t consistent. Some @angular packages are in permanent beta for unknown reasons. Anything besides the latest release is abandonware.
And a lot of community libraries don’t work for new releases. Often they seem to need an update to work with the current angular version. React usually doesn’t have that problem. And because of their much smaller focus they only have major releases every 18 months or so.
Though the idea of signals has been written about and used in computing for decades, knockout was the first frontend framework to implement them in a form we’d recognize as signals today. I think that’s why people are getting worked up about angular using signals.
Claiming we’re circling back to knockout makes no sense though. Signals are sort of a primitive for modelling solutions; they aren’t the solution. The fastest tools for UI development are presently using signals to solve performance issues such as Solid, Qwik, or even Leptos which uses WASM and still outperforms React.
React maintainers have said that signals could even be used in react eventually, though they don’t want it to be a surfaced tool that you have an API for. Instead it would be generated with a compile step.
Regardless, we’re not headed back to knockout days and signals are not an archaic solution to a specific problem that we already have better tools for — there’s still plenty to revisit and explore to get better performance.
Isn't it easier just to learn ES7, use async/await whenever possible, learn how to do div display=none/block, read some CSS tutorials and forget about learning any JS framework?
It's easy to make your own SPA using div display: none/block, so is routing. Global store is also pretty easy to do including update events. Async/await makes backend calls super simple, one could still use promises for a few more advanced cases as well, or even go full Saga pattern. Why is a framework needed these days?
There is no framework. Just simple JS calls and bindings to HTML with somebody defining REST or GraphQL backend. And some eye candy CSS. The point is that with ES7+ JS can be finally readable and not a spaghetti code monster so anyone can easily understand and extend any code.
I was involved with some deep React apps that utilized all kinds of advanced patterns and I simply don't see the need for most apps I see around. React is neat if you have thousands of concurrent updates coming in all the time where dependency graph is a killer. But only a few companies need it - why then force this on all frontend devs by default? It's like with big data - let's push everything to Hadoop/Spark when our SQL database fits into 20MB etc.
As a long-time Angular user I had the exact same feeling, and was glad when I finally could switch job to one with a React and Next.js stack.
Which is when I truly discovered the state of the React ecosystem, and noped the f*ck out of it. Design systems that recreates the <strong> tag but with a React component, the abomination that is CSS-in-JS and "typed CSS", mixed with some of absolutely brilliant libraries with well-thought out APIs, mixed with "best practice" du jour made by clueless frontend devs rehashing arguments in a truly blind-leading-the-blind fashion.
And I do like React, and might probably use it again, but it has the downside of being the defacto choice for new devs, which creates a huge spread in the ecosystem quality that other "second languages" ecosystems such as Rust or Elixir won't tend to have.
(I know that I'm mixing a framework with a language, my point also stand for the whole JS/node ecosystem, just doubly so when you focus on the React side of the NPM ecosystem.)
I don’t like this suggestion. I mean, it’s a good, solid suggestion. But I don’t like everyone using the same thing. A market dominated by a single entity means less competition which means less innovation.
If we didn’t have competition we’d all still be using jquery.
I agree, but also I think it’s fair to say many FE jobs today are basically React dev jobs. I was really excited about Elm, but it seems like that’s not gonna take off.
I enjoy Angular as much as I love React, Vue or Svelte. It's always about picking the right tool. And when thinking of Angular it's that batteries-included framework that is godsend for big enterprises where any decision given by the outside is something you don't have to battle for.
Often I describe Angular as a hidden champion. It's loved and trusted by enterprises while they are not really talking about it. Plus they hide all the package downloads behind their own npm caching servers like Nexus or Artifcactory.
On the other side there are tons of individual engineers working with React or Vue (for good reasons) and writing blog post after blog post. Maybe that's one reason why folks think Angular is dying? I don't see any signs of that though.