A modest critique of Htmx
chrisdone.com
chrisdone.com
I especially go back and forth on attribute inheritance (it can be disabled via the htmx.config.disableInheritance option)
Three of the criticisms boil down to the fact that client-side state doesn't always play well w/htmx swaps (especially the simple ones) which is absolutely true.
And events can get crazy. They are powerful, but crazy and at times hard to debug. Such is event-driven life.
The one thing I don't agree with is the default queuing mode: it is not to cancel an existing request and replace it. Instead it is to keep the current request in flight and queue one and only one additional request. I'd need to sit down w/them to see if they were misinterpreting something, using the hx-sync attribute to implement the behavior they mention, or if there is a bug.
I would also like to take this opportunity to market our mug for people who don't like htmx:
https://github.com/bigskysoftware/htmx/blob/1242977d11bebe56...
and that dumps the existing request queue and puts request created by the last event in it by itself:
https://github.com/bigskysoftware/htmx/blob/1242977d11bebe56...
we don't cancel the current request or issue the next request until the current request finishes
but there could be a bug in the code, for sure, it's pretty crazy
regardless if you have a lot of events flying around then updating the UI via hypermedia exchanges is going to be a bad idea, as I mention here:
https://htmx.org/essays/when-to-use-hypermedia/#if-your-ui-s...
This seems to be a scenario where switch/case blocks could make the elif-trees a bit easier to read.
Also, the code could use some care about not going so deep into the Vs:
// request headers
if (requestAttrValues.noHeaders) {
// ignore all headers
} else {
for (const header in headers) {
if (headers.hasOwnProperty(header)) {
const headerValue = headers[header]
safelySetHeaderValue(xhr, header, headerValue)
}
}
}
// request headers
if (!requestAttrValues.noHeaders) {
Object.keys(headers)
.filter(hdr => headers.hasOwnProperty(hdr))
.forEach(hdr => safelySetHeaderValue(xhr, hdr, headers[hdr]));
}
Actually, I guess, then, OP just had an off-by-one error? Imagine requests [r0, r1, r2 ... rN], where r0 is still in flight... OP thought r0..r(n-1) would be cancelled, in fact just r1..r(n-1) will be cancelled (I think?). Or maybe OP understood it but just mis-described it!
I am curious to hear the reasoning/use cases for this choice being the default strategy.
This way an in-flight request has a chance to complete (otherwise if you send requests too often none of them would ever finish), but any request that hasn’t started yet isn’t sent since it’s meant to be overwritten by the result of the later request anyway. Of course if you expect later requests to do anything other than idempotently replace the previous ones you’ll be surprised.
Although that may be because you have a background of Industrial Engineering.
It seems like such an elegant solution I’m surprised more people haven’t started using it. It just works.
because of the event system on the web - things yeah are weird.
And thanks for bringing intercooler.js / htmx as alternatives to a crazy world.
I should have disabled it early in the project, too late now.
I also still haven't figured out how to properly use the "save history to local storage" thing. Often there has been a server state change between the user navigating away and clicking the back button and I see no option other than disabling the thing altogether.
I do believe everyone involved means well and aims for technically strong outputs but my good the end result is still fuckin chaos.
Backend and systems programming has people with strongly held opinions and flame wars too but somehow it feels more like a war between gentlemen and less The Purge chaos.
i think the ideas of hypermedia are fairly stable and relevant, particularly the concept of a uniform interface, even if htmx is just one implementation of them
Who told you to do all those changes? Maybe change who you're listening to, or even better: don't blindly follow what others say/talk about, but listen and think about it thoughtfully instead. Stay on whatever framework/library/platform you want and feel works best for you and your use cases. No one is forcing you to chase the latest trends.
It helps to pick "boring tech", and hope that what has worked will continue to work. But it's still unpleasant to hear about the X killer every week, when X is your bread and butter.
That said htmx is the only remotely intriguing development I've seen in frontend dev in a decade.
I feel like the backend is just an inherently simpler problem to solve, request in, response out. Largely stateless. Of course there are harder things to do like stateful 2-way connections, but the solutions are mostly concentrated on a couple of technologies. Like yeah we have stuff like graphql, but at the end of the day the principle is the same.
The problem is thinking that all of these things are different domains when they really aren't. The web browser is just a program; its job is to fetch resources and render HTML. If you know how HTTP and HTML/CSS work, you can absolutely make great web applications.
But this requires knowledge and expertise, and also maybe not shoehorning an entire application framework into what was designed to be a collection of hyperlinked documents. I suspect this has more to do with money than anything else... the web is dead, long live the web.
The backend and systems programmers are still writing code in a programming language.
So much of FE dev now is stringing together other people's declarative frameworks. When programming is done it's mostly glue. So if you exist in this paradigm you're going to have very strong opinions about whatever framework/tech you are using because it's so hard to operate outside of it. You also gain a vested interest in the survival of whatever you've chosen.
This is why I've been slowly trying to make the transition out of web dev, at least FE. I don't want to deal with this crap anymore!
Interestingly, I think this problem exists in areas of back-end work. I know people who mostly do systems integration, where they're customizing some enterprise behemoth. But since that world is anchored both by the big vendors and by your company's choices of them, it seems saner to me. People may identify as a SAP developer or whatever, but they don't have to advocate for it or have a quasi-religious belief in it.
I don't know FE development enough to speak on it but I agree with your sentiment in general. When I see "declarative," I think "learn by memorization and trial and error rather than reason and intuition."
Thankfully, they don't take over your application code, though they eat a considerable amount of brainspace, especially in the collective discourse.
Discretion and focus remain key skills in tech these days.
I gave up on webdev when this started to set in. Honestly, it felt like a bunch of kids chasing thoughtful engineering out by simply making more noise. Everyone wanted a magic library that relieved them of thinking, but the only way to make decent software is to think deeply about what data flows where, and how it gets there.
I've since rekindled an interest in it after finding some tech that did click with me: Quarkus + Renarde, htmx, Svelte. All of these are off the beaten path and have users that choose to use them, versus those that have to do, and I really think it makes a difference.
> This is why I've been slowly trying to make the transition out of web dev, at least FE.
Systems/backend dev is a lot more friendly to thoughtful engineering IMO. Check out Java, Golang, and Rust and see what libraries/users you click best with.
And honestly, using Rails or other server-side frameworks, you can get very far with way less effort than those expensive front-end teams by simply using Hotwire:
Sending down rendered HTML using Hotwire Turbo requires far less time, and HTML over the wire is in reality no heavier than sending down JSON. If you absolutely need a bit more interactivity on the front-end while avoiding a server roundtrip, it's easy to drop in little Stimulus JS controllers as-needed. From my journeys to and fro in the real world, I've seen most projects do not need more than that, and are arguably wasting budget trying to use heavier tools than that! For most sizeable projects, you can do more in Rails & Hotwire with fewer developers than a 6-person team using their favorite server side language and React. I'm not saying you have to use Ruby on Rails. I'm saying I wish the dev world would embrace this paradigm in whatever their favorite language/frameworks are.
Frontend, you largely have to pick a team. Trying to run multiple frontends in the same page has a number of problems, if nothing other than each of them gets so large that even just one can impact performance, but they will also fight over events, and you can't cross the streams, etc. The whole structure is more winner-take-all, by its nature.
Don't get me wrong, the front-end certainly seems like a mess to me too. Although I think it's actually not quite as bad as it used to be, maybe over the hump one can hope.
Also, and it seems like this has to be pointed out in every one of these threads, the complexity of interfaces we're building on the web now is far greater than what we were building 10 years ago.
I have Python + Tkinter code that’s close to two decades old and still runs great (doesn’t look great tho but it didn’t look great at the time either).
I have some vanilla JS websites that are about as old and still work great.
As other commenters have noted, boring stable choices are there. Ignoring them for the fancy novel brittle stuff is entirely on the programmer.
All of those are good in theory but don't stand against the demands of modern consumers.
If as a technologist you can’t imagine companies making money selling software to happy customers without relying on the brittle trendy framework du jour, my only advice would be to spend less time making silly bets with strangers on internet discussion boards and more time looking at boring profitable companies.
They don't care about React or htmx. They just want something that fixes the problem.
This is actually very freeing: meet a real need and you don't have to believe that software development consists on running on a treadmill of constant changes for their own sake. Just do the job competently.
Most of the time it is dictated by the whims of corporate(by making frameworks/tools as part of a job requirement) or by the mob(teams need to collective decide what technology to use).
You sincerely believe that devs will just add complexity to their work just because?
Meanwhile when I go out in the world I see all manner of front-ends being actively used. It's almost like different users want different things idk.
Backend (and frontend!) can be use stable technology choices if you want to. There are also plenty of bleeding edge libraries/frameworks/languages that come out every week.
As an example, how do you want to deploy your backend? VPS, severless, Kubernetes? What database? Which language? Which libraries?
There are boring stable answers to the above for backend, just like there are boring stable answers to frontend.
Really, I think you’ve just found those boring stable options for the backend and you haven’t done that for the frontend yet.
React has been out for more than a decade and it’s the dominant way of building UIs. And there are pros and cons to this with 1000 solutions, but, really, plain React will work for most things you want to do.
Joking aside, you really do need to solve the problems jsx solves, but also webpack, next.js, react-router and depending on your application, something like mantine or tailwind. These are the minimum I think.
If you have a smallish project you can get by hand-rolling your own solutions, but it doesn't make the problems themselves go away.
In fact, it is exactly the claim in this thread that there are a lot of beaten paths in backend languages with relative stability over time, whereas in frontend it is 1) not clear what the beaten path is (there isn't one) and 2) when there appears to be favored solution, it often changes radically or disappears after just a couple of years.
For example, create-react-app used to be the default way to start a react app, it is now deprecated. But for python, Django has been a stable beaten path for many, many years. Same for Rails. C# also tends to change every couple of years, but there is always just one single Microsoft recommended way to do things.
Sure there are many experimental libraries and contenders in these languages as well, but there isn't really anything with the same maturity and stability in frontend.
<script src="react.js">
<script src="react-dom.js">
<script>
const recognition = undefined
var foo = document.querySelector("#foo")
let bar = ReactDOM.createRoot(foo)
foo.innerHTML = bar.render("<b>eyond<a>ll{recognition}");
</script>Alternatives are Webpack (stable, boring), Vite (starting to overtake Webpack since it's significantly simpler), and 1000 other libraries.
Again, there is some movement here, but Webpack has been in a dominant position for nearly a decade and there's no reason you can't continue to use it today if you value stability. It's the equivalent of choosing to host your backend on a VPS instead of trying out serverless.
Needing a build step is part of the complexity of frontend. I think frontend is getting much simpler with build tools like Vite and the addition of ES modules. But, even with these new things coming out, you can stick to your boring stable choices on the frontend just like you can on the backend.
This is insane for beginners.
> You need at least a router
You only need a router if you want to build a SPA with different routes, not all web applications are that! Tons of practical React apps are just a single screen, or even embedded component.
> some way to manage state
You do not need an external library for complex state management, the React docs have some great guides on how to achieve this with plain React:
https://react.dev/learn/scaling-up-with-reducer-and-context
> some way to manage style
Plain CSS and the JSX style property are entirely enough for most apps. If you are using a build tool like Vite, you can also do CSS modules for scoping styles to your React components for more complex applications. I do agree that styling is not a strong point of React, but I don't think pulling in a dependency is necessary.
I just wanna vomit reading that article (which is in official docs!!!). There is a much cleaner way of managing state with just a single `useState` per component instead of bringing over some Redux dogma for some inexplicable reason:
```
const [state, dispatch] = useState()
// ...
dispatch(doThis)
dispatch(doThat)
```
https://react.dev/learn/managing-state
I was linking this later section about extracting state logic into a reducer, because you suggested that you would need "something for state management", I assumed you were talking about something like redux.
But of course you can get pretty far with a single useState and prop drilling as well. Not sure how it supports your point that 'plain react is useless', it sounds like you are arguing against yourself with that comment?
The Redux "pattern" you speak of, does nothing to manage state better than useState(). So I'm not sure why that API even exists, or why you think that is a good response to "need something to manage states".
Let's go back to that, managing state would be for example error and pending state for every async action. In react, that basic, repetitive task is a whole celebration if you just do plain. Try do do form with per-field back-end validation feedback and see how much boilerplate you need for that.
I 100% agree that React is super boilerplate heavy, a lot of things are indeed 'a whole celebration', I personally choose something like Svelte over React if I have the choice, because I don't enjoy that either. All I'm saying is that plain React is not useless! It's completely fine. I don't mind working with it, it makes a tradeoff of being more ceremony but perhaps less magic and I think the tradeoff is okay.
If you want an actual website with proper routing and possibly SSR and want to use “plain” React with no additional libraries you’ll have to waste a lot of time reinventing the wheel.
It is unavoidable to have a truly gigantic amount of (transitive) dependencies, and its rare you go a month of updates without it breaking your app.
It is true that for backend it is more or less the same, however it is a matter of degree. If frontend is all you do, it may be managable, but for full-stack devs who also need to manage infra it can sometimes be overwhelming.
You really don't need any of this though, you can get super far with React alone, and the new docs on react.dev are great at explaining this and pushing this idea in my opinion.
You don't need any of this (aside from a bundler). That's my whole point. You can use these libraries and be on the cutting edge, or go with something simpler.
You'll have the exact same problem on the backend if you want to use a library for everything.
I've maintained a sizeable react app for 8 years, and this has not been my experience at all. Maybe 10 years ago things would break monthly but nowadays things are quite stable, with very reasonable deprecation schedules and upgrade paths. I have lots of code running that I haven't touched in 5+ years and it is not causing any problems even when I'm keeping the main parts of the appliaction up to date with modern practices.
There are a lot of green screen terminal apps that still work today, and haven't been touched in more than 5 years. As long as you leave the same code on the same machine once it's working, and keep the moths off the vacuum tubes and the power flow steady, there shouldn't be a problem.
But try to change that CP/M parts terminal so it can interface with your new EPR and see how well it "just works".
My point is just that the React ecosystem has become surprisingly stable. Of course if you are going to continuously rewrite things into the current flavour of the month framework, you are going to spend a lot of time doing that, but keeping up with upstream changes in dependencies is not something I have found particularly burdensome.
To take the project I mentioned above as an example, currently it’s running on React 18.x, most dependencies are on the latest major version et.c. Still, much of the code is as it was written 8+ years ago, because the core APIs of React have been stable.
Maybe in 2025. I've learned Svelte 5 and think it is pretty good for the tradeoffs it makes though. And I think MPA is the right default for 95% of sites.
What you wrote is probably true (and the one "see how far you can get in react only" comment is probably a decent path, but the landscape is overwhelming.
E.g., I might think to ask, "Which tool creates the output with the lesser overhead on the client, Webpack or Vite?" But I can't find anything solid about that, since everyone's too busy hyping up the dev experience or whatever.
It's a shame that you have to swim against the tide so heavily, if you value simplicity over immediate kitchen-sink levels of functionality. Personally, I've landed on pure client-side React (with any bundler, or with Babel alone if you're feeling adventurous) since it doesn't try to have any purpose other than updating components according to state. Many of its competitors have too much poorly-documented magic for my taste.
Its only a circus if you're constantly chasing the new shiny thing. Many of us have been productively building apps in react for nearly a decade without significant changes in tooling.
So qualify "building apps in react" with whatever bespoke build system, router, state management, etc combination you are actually using.
Just because you found yourself a tent inside the circus doesn't mean it's not a circus.
Frontend is a mishmash of random components and packages, unless you are 100% sure in advance what are you going to use and and do chances are that you’ll run into with issues with the latest version of package X not working with build system Y due to cryptic incomprehensible reasons unless you make sure to stick with the latest “mainstream” trends (which seem to change ever 1-2 years).
You could say the same about developing in Clojure, or Rust or Golang, but somehow you typically only see this same tired rhetoric applied to JavaScript.
The rhetoric applied to JS is fully valid and justified.
Things also change in all language ecosystems. Does anyone use lein these days? Not really.
This is utterly jaw-dropping. There is simply no way to respond to this, since the Everest mountain representing the JS reality directly contradicts this. Its like saying one does not experience Newton's laws of motion. I think you should definitely hold a poll with the community. You would be the super, ultra minority who would hold this position. The only you would have avoided breaking stuff is if you stuck to plain old JS, avoided use of all frameworks, maintained all libraries yourself and did not use any build infrastructure. There is utterly no other possible way you have gone a decade without stuff breaking. React has changed so much - its tooling, libs, best-practices and idioms in the last decade alone.
Comparing the JS versus Go ecosystem is like comparing a mad-max kraken with ten-thousand tentacles surfing the waves and wild on steroids vs a boring container freight-ship placidly making its way across the ocean.
But yes, it can be fun to surf on the kraken.
React router for routing, since.. as long as I can remember.
This notion that the ground is constantly shifting beneath you just hasn't been true for over a decade.
Use CRA + webpack, now use Vite instead, you know what? Screw CRA you should just immediately jump to Next.js instead (which seemed like a huge overkill initially, but actually seems kind of nice).
Unless you’re working on a single project or continuously following what’s new it just seems confusing and overwhelming. If I got comfortable with CRA and came back after a year or two should I still use it for a new project even if it’s bo longer the default? Will new packages/etc. still work with it? Maybe… who knows. I just know that I now have to waste time figuring that out.
When React "is just a library" but the installation instructions tell you to install some large, VC-backed framework immediately - it just doesn't sit right with me.
But also as someone who does fullstack in a pretty much even 50/50 split, this is such a tired meme by this point seemingly from people who just refuse to actually sit down and think about the FE ecosystem for longer than 3 seconds and end up parroting this same talking point.
You could make a similarly dramatic post about the backend world. There you're even fighting which programming language you want to pick. PHP, Ruby, Go, Elixir/Erlang, C#? What about the recent mess with ElasticSearch and Redis? Where and how do you host your backend? Which DB do you use and why? Mongo, anyone?
At this point in the FE framework world, you've got 3 very stable players that have been there for a decade. Angular, React and Vue. And sure, they've each gone through some sort of reworks in the past, but nobody's forcing you to pin your dependencies `@latest`, you can pretty much always keep trucking along with whatever version you were on. We've got a huge app still running along on Vue 2 (not even 2.7 which backports the very nice Composition API) and we see no reason to bump versions to 3. I've even had some nightmare scenarios trying to update Rails and Laravel in the past that dwarf any problems I ever encountered in the frontend world.
Also, vanilla JS, HTML and CSS still exist and work just fine if you're building some small toy thing, but you'll quickly realize why these frameworks exist in the first place (or you'll do the dumb thing of building your own as some do) if you're doing anything more complex than a brochure website.
And the tooling has only gotten simpler if anything with Vite. Want to start a new frontend project? It's literally just `npm create vite@latest`, and you get a nice TUI with which you get to fully customize what you want and don't want. The README even comes with a nice little list of commands for you to run, and hosting can be done literally anywhere where you can deploy the `dist/` folder to, just like the "good" old days.
For people who always claim superiority for being the genius backenders they are, they sure seem to be having a lot of trouble with something that they will themselves say any coding monkey coming out of a bootcamp can do with relatively few problems.
There's a joke I've told many times in interviews if they ask me to define what a full-stack engineer is, to which I joke "It means I do whatever the senior engineer doesn't want to do, which 99% of the time is front-end". The people I tell it to laugh, because it's funny but also because it's true.
On the front-end you're dealing with more customer/users and it's inherently more of a servile role, while it's much easier to be stubborn, eccentric, 'powerful' in the tech-facing backend.
I'd also wager there's a certain "tech purity" angle too, given the inherent reverence foisted upon lower level languages and programming fields that is seemingly rife.
To play devil's advocate on myself, perhaps it's more that the higher end technical complexity challenges for Front-End aren't commonplace enough. I've had the fortune to work in a field where the level of necessary technical complexity on the front-end part of my job is very high, so ymmv
In the Java world alone: Spring vs EE vs Play vs Quarkas vs Micronaut vs Helidon... Tomcat vs Jetty vs Netty... Jar vs War vs Ear ... Maven vs Gradle vs Ant...
Just don't chase fads. Stick to things that you know are working fine.
https://farmrio.deco.site/ is one of the largest clothing retailers in Brazil and all the frontend interactions use HTMX, alongside a partial rendering strategy we developed.
More info at https://deco.cx/en/blog/htmx-first-class-support
You're artificially making your FCP also be your TTI, which means page navigation, when everything should be cached and fast, feels slow. That's not something e.g. Lighthouse tells you.
I recommend showing the page right away, even if there's going to be jank. Jank/Cumulative layout shifts can be fixed later.
However I think it's a powerful solution to updating the DOM when state changes on the backend and a roundtrip is required anyway. For that matter the examples on the official project site are better to understand proper use cases for HTMX [0].
Personally I use Active Search and Edit/Delete Row a lot. While Active Search does not change state in the backend, I found it's still a valid use case for HTMX when searching large data sets, since the backend is (in my case) just way faster to search through a large data set.
It all comes down to your ability as a developer to find the proper solution to the problem at hand. A counter does not need HTMX.
If you’re using an HTTP request to update a counter, it would be to update the persistent server-side state of that counter (which you’d also do if you’re using React and a JSON API).
No one is advocating for using HTMX for purely client-side state. They’ve been very consistent about this, recommending Alpine, vanilla JS, Stimulus, Vue, and so forth when you need pure client-side state.
For example a like counter wouldn’t be much good if nobody else can see your like.
An attractive future direction might be to re-implement Htmx in React:
• The server sends a JSON blob that React converts into virtual DOM components.
• That would solve the component state problem.
• It would mean we require no special bridge to use React components.
• It would let us use our React-connected web fetching library, and carefully avoid the queuing choices made by Htmx.
• It would solve the morphdom problems and browser DOM input elements problem, too, which is pretty much a solved problem in React.
In this way, we could drop the Htmx dependency but retain the benefits of the idea. That is, given a budget to embark on such a big piece of work.
RSC is still experimental, and the default implementation assumes that you're going to run JS on the server, which undermines some of the point of HTMX.
But someday (likely in the next year or so) the RSC wire format will be standardized, and then any language can generate RSC JSON for React to consume.
> By default, Htmx will cancel requests that are in-flight if you trigger another request on the same queue (element).
This seems like the only default that's reasonable to me
don't know if the author has a specific example in mind, but if a user submits an input, changes that input mid-request and submits again the request needs to be cancelled, otherwise the UI will be inconsistent by showing a response to an outdated input
Just processing one response after the other won't be possible if a response swaps out content with different IDs, so a second response won't be able to be swapped out in the same way
If you queue them and process responses in order, that would also be correct. Probably has fewer failure modes. Some indicator that an element is still being processed would help to prevent user confusion though.
if a response is meant to swap out a div, the first response completing will swap it out - if this response isnt compatible with the same selector (e.g. a #response is not available) a second completing response won't be able to update the UI
Isn't this exactly what HTMX was replacing? The client needing to have a ton of logic and state?
It's easy to put unobtrusively in place, fun to play with, and it worked. Reminded me of using jQuery for the first time. Would I use it for a complex and involved project? Initial impressions suggest probably not.
I think it's one of the best ways to use Hotwire/Turbo. Default to Hotwire to whatever degree of understanding you have, and as soon as it feels like HTMX/Hotwire isn't the right tool, you can easily switch to your framework of choice.
It's a spectrum of interactivity, and we should try to pick the best (whatever that may mean) tool possible for the job - sometimes that's hx-boosted links, sometimes it's a react app
Would love to hear more about this issue. Preserving the state of elements like input or detail is the main function of libraries like morphdom, something must have gone wrong there.
if you look at the algorithms they give up very quickly and simply merge the new nodes in once matching doesn't work at a certain level because of this
there is a new API coming down the pipe, moveBefore(), that will allow much better morphing/preservation in general:
https://github.com/noamr/dom/blob/spm-explainer/moveBefore-e...
htmx supports it today if you enable it in chrome canary:
The examples don’t even make sense, it just sounds like someone trying to use htmx like it’s react and then getting upset cos it doesn’t work how he expect it to. Lack of experience.
I want to dig into this a bit. React of course maintains its own rendered part of the DOM and htmx trying to reach into and change any part of that DOM is not going to go over well. It's just going to be replaced with React's rendering again on the next render cycle.
htmx provides two points at which it can interact with other libraries or frameworks: DOM events and CSS classes. I don't see any problem with classes, but React's synthetic events would probably not work well with htmx (or any other non-React library really). Maybe frameworks like Preact which use real DOM events would work better.
From my very personal and pragmatic point of view (I am a lone developer : I need to create software solutions that are robust and that simply work without the need to spend months to learn a new technology, that probably will be replaced by a new one in two months or so, that no customer is willing to pay), HTMX is the most pragmatic and useful solution.
It is complete in features and you can do literally anything with it and is easy to integrate in existing web applications.
I used it with CodeIgniter and now with Golang.
You could you query string as mentioned in the article, but that's not really convenient when done in a custom way for every single case.
Having said that I think that a way to go could be a community effort to settle on the ways to handle different ui patterns and interactions with some bigger project to serve as a testing ground. And that includes backend part too, we can look at what rails does with turbo
My opinion is that htmx (and similar) approach is different enough to potentially require a different set of ui interactions and it usually hurts when one tries to apply react friendly interactions to it.
Everything turns into spaghetti code really quickly.
Separation of API and frontend isn’t just for fun. It’s for making testing bearable.
Finally, mixing data modelling (database) with presentation (html output) in one place (htmx handler) goes out hand really quick too, making testing much harder.
But real coding cowboys from the golden age don’t write tests, right? :-)
I can only guess that you're trying to put all your logic and structure on the front-end, with the backend only serving what the front-end needs ad-hoc? Don't do that. Build a classic MVC app (Django, Rails, ASP.NET MVC), factor your views into small, reusable components, and use HTMX to replace page elements at that level.
Similar concept and functionality. jQuery plugin attach events like "x-post" and other attributes, sends data to server which always returns HTML.
800+ CRUD pages full of business logic, around 2k routes, single PHP server, single MySQL server, serves up to 3k requests per second seasonally, P95 around 50ms.
Team still adding and changing features every 2 weeks, even after years in production. Stack is custom PHP framework, MySQL, custom jQuery plugin that acts similar to htmx.
Onboarding is dead easy. It was made with a no-build frontend stack. Meaning there's no build pipeline to understand and fight against.
I look at React/SPA misuses and self inflicted pain, and feel sorry for them.
Handling events by writing functions that need to match unchecked strings feels _super_ clumsy to me.
I know elixir processes are very very cheap, but I still don't like the idea of maintaining a stateful connection for each live view page, especially with the lack of typing.
Jumping between my markup and the 10s of different handlers in my live view is annoying too. I like the locality of behavior principal a lot.
Again, I really like elixir, but untyped languages kind of suck for dealing with the string based world of web frontends. That's why so many javascript shops take on the complexity of typescript.
The corollary argument is -- Why don't you have a simple UI?
No use if you want react though...
> The server sends a JSON blob that React converts into virtual DOM components.
I don't understand what is left of HTMX if you do this.
Isn't that just React?
The author should try inertia.js, it has server-side routing and react templates.
I’m building a thing with HTMX and hyperscript. Very excited to run into these issues for myself.
I don’t know, maybe it really solves their problem, but it just feels so… wrong. I don’t know. (Shakes head, gesticulates, exits the scene)
The better we can keep state out of abstractions without losing maintainability or performance, the better.
What an amazing way of handling state!
Or the part where you use data tags for something that should be done with JSON...
On the contrary, Unpoly -which shares similarities with HTMX but is probably more high level- makes it very clear on its front page:
Progressive enhancement for HTML
Get powerful new HTML attributes to build dynamic UI on the server.
Works with any language. Gracefully degrades without JavaScript.
Anyway, as always, tools ad trade-offs...