[aardvark1]

I've working at large F100 financial services companies as an engineer and technical manager for 10 years. I've seen front ends built using a myriad of techniques and frameworks.

Leaving bytes and bundle weight aside, in 2020 JS frameworks like React are far and away the best approach for front-end engineering from a large organizational standpoint.

1)Not difficult to hire for. JS frameworks are currently in the sweet spot where they're both widely popular and "cool". Consequently it's not hard to find good candidates. That wasn't the case 5 years ago where React/Angular developers were still uncommon.

2) The HUGE and helpful community of developers. There are so many excellent resources for learning and solving problems with JS frameworks. JS developers are eager to show off on medium and similar sites. I've never seen such a large and enthusiastic community. This will inevitably subside as JS fall out of favor but I hope it doesn't for a long time.

2)Rapid UI development. The JS ecosystem make it easy to build UIs quickly that don't completely suck and look good for the ever-important demos to senior leadership.

3)Opinionated frameworks naturally enforce best practices and ensure relatively consistent architecture. This makes it easy to move engineers from one team to another and make grokking new codebases a lot easier. New engineers are productive quickly and begin creating value for the company. This was not the case with the inevitable jquery monstrosities that resulted after a critical mass of code or # of engineers.

4)Fewer bugs. JS ecosystem makes it very easy to write unit and integration tests.

Companies with IT budgets that are 9+ figures will happily pay for extra bytes across the wire in exchange for those benefits

[fooey]

One of my main things I argue strongly for on any new project is "don't do anything weird". Don't use weird fringe languages, don't use weird fringe frameworks. If the thing you want to do doesn't exist, think real hard about why it does not.

At my office, we've somehow ended up with a critical graphql server that's written entirely in Kotlin, which is cool but completely impossible to hire for or get any kind of community support for it. It feels like absolutely everything you want it to do, you have to do build yourself, whereas if we'd used a normal off the shelf graphql server it all already exists.

I'm so sick of maintaining old weird forks and unnecessary custom code that doesn't need to exist.

[james-mcelwain]

I think the advice is sound, but Kotlin and Graphql are pretty mainstream. This sounds like a local market problem.

[karatestomp]

I think if you're putting out "Senior dev with 3+yrs experience in GraphQL and Kotlin" ads you're going to have real trouble in most markets, if not all.

Of course you shouldn't be doing that, is the thing, but that's how most job ads read.

[fooey]

Yeah, this is what I mean

Kotlin is fine, and GraphQL is fine, but when you combine those, you have a very small niche.

[ptx]

Kotlin is really very similar to C#. Java developers shouldn't have much of a problem either (now that Java finally has lambdas), and I hear Scala is mostly a superset of Kotlin.

GraphQL and (Kotlin or C# or Java or Scala) probably isn't that small of a niche.

Edit: Also, Kotlin does have near total compatibility with Java and can use any Java library (aside from the most magical bytecode-rewriting annotation processors) so I don't understand what you're saying about having to rewrite things that exist.

[erik_seaberg]

I'd put it the other way around, Kotlin is a subset of Scala that the masses might be ready for. I would always prefer the kind of candidate who's excited about stuff like Kotlin over others.

[james-mcelwain]

The nice thing about Kotlin is that you can just bind to the existing Java GraphQL libs. Clojure is similar in this respect. Sure it means you have to also be familiar with Java and the Java ecosystem, but it's actually great to interact with Java-the-ecosystem when you don't have to write any java. ^^

[ricardobeat]

Why are people even hiring specifically for such a niche subset? These are not huge complex systems, anyone can become familiar with GraphQL in days.

[karmakaze]

I would further describe 'weird' as being out of the norm for the company and where hires come from. In general, I have a 'weird' budget of 1.0 thing(s) for non-critical projects.

Recently, I've been spending it on Vuejs/TypeScript (with Go backend) at a company that uses React/TypeScript, GraphQL, Rails/Ruby, and Go.

[ng12]

Absolutely agreed. I have never in my life had a stakeholder get on my case because our site was 3mb when it could have been 1mb. I have had many stakeholders get on my case because we couldn't turn around features fast enough, or the site was buggy, or our UX was worse than our competitors. Those are the things that the people who pay me and my team care about and if I could 10x our bundle size for a 1.5x improvement in development speed I would do it in a heartbeat.

[untog]

That depresses me because it's clear who is being ignored here: the user.

A 3MB bundle size instead of 1MB does make a difference, and if you're working on the kind of site that relies on conversions/purchases to succeed then you're absolutely leaving money on the table. Not to mention that Google factors in page load time to rankings.

> our UX was worse than our competitors

If your competitor has a < 1MB bundle and yours is 3MB then any user on a low end Android phone is absolutely going to have a worse UX experience. Maybe stakeholders aren't aware of all this. Maybe you should tell them.

[ng12]

Most of the apps I've worked on don't have conversions and aren't indexed by search engines. One app I worked on recently did have a mobile version that received a small fraction of all traffic, and even then overwhelmingly from relatively modern iOS devices.

It's like all engineering: use the right tools for the job. If low-end Android users are a significant portion of your target audience you have to design your system around that.

[tiborsaas]

Dev experience and productivity is key, but don't forget that 10x bundle size has effects you can't simply dismiss. It does cause a UX problem since a slower pageload is part of the UX.

SEO. Google also takes into account page / paint times.

These are the things your stakeholders probably don't know about but you should.

[dirtnugget]

As a Frontend Engineer I can assure you there are ways to counter that issues but they're not exactly popular: Server Side Rendering. It adds another step t the build process in which HTML is prerendered, so you get the speed of loading plain HTML. It then loads the actual JS in the background. It's a PITA to implement

[tiborsaas]

With tools like Next.js or Gatsby it's not that bad. We use Next at work on a large scale and it works for us. But do you know what language gives you instant SSR? PHP :) It's funny how we had to invest so much effort to replicate with another language supports out of the box. I've built a blog in a few hours with Craft CMS, perfect tool for the job this case, I didn't even have to touch PHP files, just the templates.

[ng12]

The majority of the projects I work on target users on desktops with modern browsers. You have to get into the massive 10mb range before noticing any lag, and even then only on an empty cache. SEO isn't a concern either, all of our marketing pages (when we have them) are static sites.

It's all about picking the right tools for the job. If low latency is a business requirement by all means don't use a framework.

[thinkloop]

Wait another year or two and your stakeholders will smarten up - users are noticing. Svelte JS, for example, has everything you listed minus the costs. React is very far from the end of the road.

[ng12]

Maturity matters, unfortunately. I've been looking for a chance to try out Sapper but I wouldn't feel comfortable recommending that we start a major new project with it.

[aabbcc1241]

Also don't miss out S-js and Surplus.js. They are helpful tools to build reactive tools without VDOM.

[milesvp]

I was lucky at one point to work with product owners who understood that latency translated to lost revenue and so they helped champion the idea that engineering excellence was actually valuable, and that squeezing out performance was itself a worthy goal. It helped that we could compare our metrics to comparable sites, and could give upper management bragging rights when our pages loaded faster than bigger higher staffed “competitors”.

But otherwise, I think you tend to be right, in a lot of areas even an extra second of latency doesn’t matter, and more features will move the needle more.

[goatlover]

2 more MBs is going to matter to some users with limited data or bandwith.

[ng12]

Sure, that's a trade-off you have to make a decision on when designing the core architecture. If your users are sitting behind corporate T1 connections near an AWS data center bundle size doesn't matter at all. If your users are still on 3G phones it matters a ton. For somewhere in between it becomes a business decision: how much performance are you willing to trade for developer productivity?

[hundchenkatze]

You're comparison implies that T1 is faster than 3G, but that's probably not the case except in really poor coverage areas.

T1's maximum data transmission rate is 1.544 megabits per second. [0] While 3G's theoretical maximum is 21.6 Mbit/s for HSPA+. [1]

[0] https://en.wikipedia.org/wiki/T-carrier#Transmission_System_...

[1] https://en.wikipedia.org/wiki/3G#Data_rates

[entha_saava]

In practice, 3G is quite slower than that.

[nojito]

>I have never in my life had a stakeholder get on my case because our site was 3mb when it could have been 1mb.

As cloud costs become tougher to swallow, this is going to change.

[untog]

The common thread in a lot of this is disregard for end user experience. But I think the biggest reason that these discussions never go anywhere is that people are using the same tools for wildly different purposes.

At a guess, I'd say that a webapp for a financial services company is going to be loaded once then interacted with heavily? Single Page App model. In that instance the time taken to download, parse and execute JS upfront isn't a big deal. Same as Gmail: when it loads, it takes a couple of seconds, then I leave the tab active for who knows how long. Plus in this situation the audience is relatively monolithic: a lot of desktops, and most of them fast.

The problem is people taking lessons learned in that environment and applying them elsewhere. Shopping sites, news sites, essentially anything that people arrive to from a Google search pay a much bigger price for a giant payload. Users can and will leave your site if it's taking too long to load. They'll be running on all kinds of devices, including low-end Androids with very underpowered CPUs.

Whenever I see a site like that implementing a full React page with Redux and who knows what else bolted on, I cringe. It's not an appropriate use of tools. But these days people only look at developer productivity (we can add so many more features so much faster!) rather than focusing on the experience in the user's hand.

[metalforever]

The extra bytes across the wire come at a cost of worse search engine performance (extremely costly for ecommerce) and bad accessibility for people using screen readers. #2 I also don't agree with. The JS ecosystem seems like the opposite of "build UIs quickly". You can't swing a stick without hitting like 5 tools you have to use.

[nitsky]

> This will inevitably subside as JS falls out of favor but I hope it doesn't for a long time.

I don't think JS will fall out of favor for a long time.

Languages that compile to wasm are unlikely to be competitive until native dom access has broad browser support, which could be years. On top of that, compilers will have to be written to do code splitting and compile times will have to drop.

Languages with strong type systems are unlikely to be competitive because TypeScript is good enough that JS projects are now manageable.

[jve]

> 1)Not difficult to hire for. JS frameworks are currently in the sweet spot where they're both widely popular and "cool". Consequently it's not hard to find good candidates

If someone is good at JS why not hot hire him? Wouldn't a professional be up & running with ANY framework you throw at him in a month or less?

[scarface74]

I know JS like the back of my hand, I’ve been working with it off and on throughout the years since it was first in beta for Netscape Navigator. I’ve had to do some relatively complicated JS logic on the front end and back end.

But, I wouldn’t work for any company crazy enough to hire me as a modern front end developer.

[sbayeta]

This is also true for other well established web (not JS) frameworks like Django or Rails

[troughway]

Not only this, but eventually we will ditch JS in favor of something more "isomorphic" with whatever the back-end is using.

Tech like Blazor and related asm.js type thinking will eventually replace JS when it bypasses asmjs completely and ships bytecode that the browser executes directly.

I doubt these frameworks will be much different compared to what we are doing today, because we will still need to deal with layouts, alignments, the color and shape and proportion of things, event handlers etc. The usual stuff.

Ragging on React, Vue, Angular is misguided. These frameworks allow developers to write front-ends as if they are applications, because in a lot of cases, this is what the modern "website" really is. They are training the next generation of front-end developers much like how jQuery did in mid 2000s.

[nitsky]

> Not only this, but eventually we will ditch JS in favor of something more "isomorphic" with whatever the back-end is using.

Why not use JavaScript itself for this? That is the strategy employed by Next.js/Nuxt.js/Sapper/etc. Your React/Vue/Svelte components are rendered on the server and then hydrated on the client.

[bcrosby95]

Why not just have everyone code in Javascript now and forever?

Same reason we didn't settle with everyone coding in C. Or C++. Or Java. Or Python.

The web is just a method of distribution. Not every problem or domain takes equally well to each programming model or language. It would be nice to be able to pick a tech stack that works for my domain and the characteristics of my project, rather than picking a tech stack because of how users access it.

[james-mcelwain]

Because Node.js isn't appropriate for a lot of back-end applications, or for legacy back-end applications that already exist.

[jerf]

Mostly because for all the development done on it, Javascript is still a %50 accurate name; it ain't Java, but it is definitely a scripting language. Node may be popular, but it's lightyears away from being able to say that it's chasing all the other languages out of the server niche, and it's been around for long enough that we can judge it's unlikely ever going to. In 5 years it's going to be a lot easier to get a non-scripting language compiled on to the client than it is to make a scripting language work as your sole backend language.

[james-mcelwain]

That dream is already here if you use Clojure, and it's great! Web assembly makes me very bullish on this kind of setup. Maybe we won't even be targeting the DOM in the future?

[karatestomp]

> Maybe we won't even be targeting the DOM in the future?

Unless the other thing you target instead is similarly well-developed, you're gonna have real trouble with accessibility.

[james-mcelwain]

That's a really good point. Accessibility is super important and a shame that many devs (including myself) forget about it.

[nimmer]

> eventually we will ditch JS in favor of something more "isomorphic" with whatever the back-end is using.

I think you are looking for Nim. It compiles to assembly (through C), javascript, wasm and more.

[zankly]

It’s like Java... tons of orgs will trade off perf to enable an engineering team of mixed quality to deliver projects.

[gautamdivgi]

What is wrong with java performance? I think the garbage collection in Java is probably (if not) one of the best.

Start-up times are an issue. But once warmed-up the JVM is very performant provided tuned well.

[eitland]

Except Java is extremely performant and if your engineering team doesn't know exactly what they're doing there's often a good chance they'll both use more time and create a slower product with a low level language as far as I know.

[to11mtm]

umm wat.

> Except Java is extremely performant and if your engineering team doesn't know exactly what they're doing there's often a good chance they'll both use more time and create a slower product with a low level language as far as I know.

So agreed that it'll probably take more time if the team doesn't know what they're doing. But as for slower product? Ehh I think that was the point GP was making; Yeah, its' an overenginnered product that results at times, but it's got just enough guardrails that you hopefully just get a slow, or at worst buggy product, versus a product that doubles as a loot box of future CVEs.

[eitland]

The point is: for the majority of programmers - including a number who can write both reasonable reasonable C and reasonable Java - they will create better, faster programs with Java than with C and in shorter time.

There are absolutely times when Java doesn't cut it but at that point your options are getting increasingly limited in many ways: who can build and maintain it, what kind of languages and libraries you can use, the hardware you use, tuning etc.

[xeromal]

Exactly. Ensure the average developer can contribute. Of course top tier devs can deliver better products using bare JS.

[scarface74]

It’s completely different. Java is a back end language - even if we accept that Java is 20% slower (just making up a number. I don’t know whether it’s slower or not), hardware is cheap and you control it. You don’t have any control of the client’s computer or their bandwidth.