The benefits, I can see.
JS frameworks move really quickly, and when we're working on a large, long-term project, it sucks when big breaking changes are introduced after only a couple of years. Sticking to slow-moving web standards (which are quite mature by now) increases the longevity of a project.
And the stability also means that more time is spent on delivering features, rather than on fixing compatibility issues.
There is also the benefit of independence. The project's success is not tied to the framework's success. And it also makes the project more secure, from supply chain attacks and such.
Because there is no "abstraction layer" of a framework, you also have greater control over your project, and can make performance optimizations at a lower level.
I feel not using a framework can even make us a better developer. Because we know more of what's going on.
But this alternative of using none... it seems rarely talked about. I want to learn more about building large (preferably web-based) software projects with few dependencies.
Do you have any suggestions on how to learn more about it? Are there any open source projects you know which are built this way? It needs to be large, complex, app-like, and browser based. I'm more interested in the frontend side.
Thank you!
React is a lot more stable than I think you're giving it credit for.
> And the stability also means that more time is spent on delivering features
Frameworks/libs also exist to save you time, thus letting you spend more time on delivering features. And fwiw, the obsidian team seems to agree in principle. Your link goes to a forum post of some kind, in which one may find a link to obsidian's third party deps: https://help.obsidian.md/credits#Third+party+acknowledgement...
These do not include React, but do include:
- i18next - lezer - moment.js
Plus a bunch of others. Why didn't obsidian write their own date lib and chose to use moment.js? Because it saved them time, despite the fact that moment.js does make changes, and many people are moving on from it in any case.
The idea that not using a frontend framework will let you focus on delivering features seems reductive, and the obsidian anecdote doesn't support the idea anyway.
Whatever you're building, it's never a bad idea to deeply understand the tradeoffs that using a library will bring. Obsidian probably couldn't accept the downsides of React due to them needing a bunch of custom renderers for content, which React makes harder. But that is likely a rare constraint for apps in general.
Generally speaking, libs like react exist to save you time and help you focus on delivering features.
Edit: for a more practical example, you don't need to go down the web components rabbithole just for native reactive components, MutationObserver can watch for your template fragments entering and exiting the DOM and run your init/destruct routines on them. Now you can just push HTML back from the server with fetch or xmlhttprequest or whatever, and as soon as your component's HTML enters the DOM it will be "hydrated" by your init routine via MutationObserver. Honestly, add some event signaling and XHR and you're further along to a smooth SPA-like experience than you might think.
The frameworky code is under 100 lines of code, heavily inspired by what I think is a good idea from React: "components are pure function", the simplest example of the about page beeing visible here: https://github.com/mickael-kerjean/filestash/blob/master/pub...
Since the whole project was migrated over from React, I was able to improve performance to a degree that would have been impossible using React, it's a breadth of fresh air to open the performance and memory tab and to be able to optimise things from there to get 60FPS no matter what whilst preventing massive ram usage because those frameworks runs all sort of code that is out of your control. Also because there is no build system, I was able to introduce plugins to be able to patch frontend code via patch applied dynamically by the server to create features like the recall button to recall data stored on glacier when using S3, dynamic patching of the icons to apply branding and many other things that I would have had to fork entirely before
Anyway, I hope our industry would go more often the vanilla route, es6 is ready for prime time and solve all the pain point we had while internet explorer was still a thing
You have been conditioned to think frameworks are necessary. That’s not true. You can build anything without them and it would probably be better in many ways. Would you spend a ton of time? Yes. Would your code be harder to maintain? Yes. But it is absolutely doable and not as hard as you’d think
I wouldn't use JQuery for a new project, as you can do almost everything it does with straight DOM manipulation. But there are still some strategic vanilla JS/TS packages that come in handy, e.g. clipboard, mousetrap, file-saver, split.js.
Web Components with Lit is kinda fun, though you'll have to deal with DOM shadow roots or disable them.
I would challenge that using a framework leads to less security. In vanilla JS you've got to use something like dompurify religiously to avoid XSS bugs, and you're tempted by the shiny candy-like innerHTML attribute.
I'm working with JS for already 25 years. Tried all of the frameworks, and continue on doing it. And every time I try something new, the refactoring flow turns most of them into NextJS (if it's very UI rich or customer facing or something very web-oriented), or Vite+React+Tailwind (client) and Hono (backend) if it's more of a tinker toy needing more custom solutions.
The boilerplate with NextJS is cleanest (compared to all the other frameworks) and API is the most straightforward one, and you can safely ignore the vendor lock in. Its just a pretext to hate on NextJS. They all have some kind of a "vendor" lock in. Be it a vendor-or-a-specific-approach-or-whatever-lock-in.
And Vite+React+Hono — simplest to set up for quick experiments, and very powerful with minimal boilerplate. Will probably create a starter for this one, as I have been using this stack quite a lot lately.
EDIT:
You can pretend vanilla JS is all you need, but then your app grows, then you suddenly need types, and state, and more events and their handlers, and SSR or something else. Thus React has been the most stable bet for quite a while for me now.
And, when a new developer comes in who knows React, they can ramp up fast and even if they don’t, there are plenty of resources instead of asking Bob how something works - if Bob is even still working there.
> I had a mind-blown-moment when I learnt that Obsidian was built without any frontend JS framework.
The comment you’ve linked to is wrong. Obsidian uses the Electron framework for its “front-end” framework.
In fact, it was even affected by the recent “Tahoe” Electron bug.
https://en.wikipedia.org/wiki/Obsidian_(software)#Availabili...
Could need some js library and that is a maybe.
In general:
If is like a eCommerce site, blog, wikipedia, forum? Not need
If is like a Game, Paint app, form builder? Need.
And how you learn it? Not use a framework, and try to stay with things more minimal, like tailwindcss.
2. router https://developer.mozilla.org/en-US/docs/Web/API/URL_Pattern...
The rest should be code organization, not even a build tool.
Can check this example https://github.com/radio4000/components
With the framework, you have structure. It might not be the best structure to use, but you have one. React lasted really long and is still works today.
Going without framework works if:
- you have a really small project
- you are so advanced developer or have specific use-case that using a framework creates more burden than time saving
- you are creating something so valuable that reimplementing it yourself is worth it
I think that regular website will bleed via maintenance costs in the long-run if you don't use a framework. Instead of paved road with framework migration you would be fixing and debugging bugs in your framework/codebase.
So yeah, JavaScript is nicer nowadays, and browser apis are more consistent than they were then.
In the past we used http://microjs.com/ and similar to find small libs to work with when we threw together marketing websites, not sure how maintained that list is nowadays but the idea is that you can make websites like lego, sometimes you don't need a whole box of Eiffel tower to make a little tree.
If your functionality is well thought out and not super complicated, don't have 10s or 100s of devs on the same project then working without a well known framework could make sense, otherwise there can be steeper learning curve when onboarding new devs, scope creep / bloat, etc that will likely cause issues down the road.
To learn about this you could try to replicate for example the mentioned Obsidian by yourself and research solutions for all the issues you run into, e.g. minimal routing, custom view renders, reactivity, performance, etc.
From there one big question is if everything needs to be dynamic or not. You often have menus and navigation that you don't need a framework for and that even something like jQuery is well suited for. On the other hand a framework would probably slow things down for the writing experience in Obsidian, which is the main draw for me to use Obsidian. Once the writing module is written from scratch, it won't seem like such a big lift to also do all the navigation and controls by hand.
That doesn't mean this is the right way for everyone. A lot of times the speed to releasing new features is more important, while the handicraft of the UI/UX is less important and in those cases using a framework is absolutely the right choice.
My priority is code clarity … so while it is not yet tested at scale, https://harcstack.org is aimed to do this on the server side (disclosure - I am the author)
Another determinant is the server side language / approach to templating you choose. With HTMX Python, Rust, Go are all fair game. I’m leading with Raku due to its web framework and ORM.
As you can see most functionality can be implemented in around 100 lines of code. The advantage is that you can build up functions just for your needs. These functions are super easy to understand because they don't have any dependencies and the whole implementation fits on one or two screens. There is very little indirection.
Another advantage is that most code is actually doing something concrete. With frameworks you often have a lot of code that is just pointing to other code, or is boiler plate code to set something up. Not having any boilerplate code also makes stepping through the debugger a lot easier.
The project has no build step in development which also makes debugging easier. It does have a build step in production.
Most people use frameworks to give the code structure. When writing vanillajs you need to bring the structure of the code yourself. This does take some getting used to but I think it is an essential skill for good programmers to have. On top of that many frameworks push you into structure that does not make much sense in my mind (for example seperating the view and the controller if those are always gonna change together).
So to conclude: My advice for building without frameworks is, just start building. Problems will come up and you can tackle those as you go. Do you think you need to use events in your application? A simple implementation is 10 lines of code and can be written in 10 minutes. If it takes me 1 hour to not use a library I will spend that 1 hour. Most frameworks will not bring that much value. Easy things become short and hard things will become impossible. I think most value is actually brought by the language and the basic api's (DOM, CSS, etc)
Going without a premade framework means it's a fight between you and the pile of complexity. Code will tend to become ugly and slow and hard to change the more you add, so you'll have to go into a cycle that's more like add-refactor-add-refactor. Forward then sideways, forwards then sideways. Do this long enough and you'll will end up with your own libraries and frameworks to manage the complexity.
If that's not enough, Lit enhances it a bit such that either LitElement and/or litHtml can accommodate whichever style you want: https://lit.dev/
There is no reason you cannot implement MVVC/MVC or whatever else on JS for reactivity, it’s just that react et al just abstracts that away for you.
You are effectively choosing some method of implementing and abstracting the observer pattern and handling cleanup since you can easily get into a situation where you have dangling references to objects that are never cleaned up (memory leaks) or pretty nasty GC hangs which can make the UI feel bad to use.
However, you'll inevitably end up with your own after some time. Even if you try your best to stay away from it, you'll need abstractions for:
- components and lifecycle
- some sort of templating
- connecting your app's data and components
Even pre-web. I remember, how quickly I could create Delphi desktop app with drag & drop UI builder more than 2 decades ago. If we speak about generic ERP, you could make dozens of screens in a day.
You just split your app to layers.
That might be for example:
1. Just UI;
2. UI logic (validations, preloading selects...);
3. Transformation of data between data model and UI model
All of the things above lived on users' computers.
All of the things below were on servers.
1. API gateway (terrible things like SOAP, XML-RPC or completely proprietary formats);
2. Business logic;
3. Database
- Routing
- State
- Rendering
A lot of frameworks try to cover all three of these, and that's where you tend to see a lot of feature churn. I would recommend picking a rendering library (like react) and writing your own routing and state library, even if it's just a light abstraction on top of an existing library. This will make it way easier to upgrade or move in the future, as you will control the API.
* greater flexibility
* faster execution, My current application produces a large single page application that fully renders visually and executes state restoration in about 105ms localhost or under 200ms across a network
* faster maintenance, I can do massive refactors that touch most files in the application in about 2 hours thanks to thorough use of TypeScript interface definitions
---
Here is how I do it:
1. Just start building your application. Realize that for any significant application the front end UI is probably only about 20% of the application code, or less.
2. Start the UI by writing static HTML. Reserve JavaScript, in the browser, only for handling events and state restoration and execute state restoration only on page load.
3. I prefer to use WebSockets over HTTP as much as possible for absolutely everything I can get away with. WebSockets are full-duplex, where HTTP is not, and have far less overhead. That will substantially reduce the complexity around messaging in and out of the web browser.
4. Everything after that is just execution, planning, refactoring, and feature addition.
5. You can do absolutely all of it in TypeScript now without any compile step. Node.js now executes TypeScript directly, but the browser does not. To solve for the web browser just include the browser code into your Node application using ES6 module import. If the imported code is a function you can convert it directly into a string for transport, and otherwise use JSON.stringify for objects/arrays.
This is what I used for fontofweb.com's snipper chrome extension.
I’ve switched to web components with lit/html driven by AI.
Could not be happier.
For platforms, just go native - it's easier than ever with LLM assistance.
A) getting better at handling graph data structure,
B) learning how to use Regular Expressions as a code editing tool,
C) quit chasing every new thing, and
D) Typescript.
JSX encourages sloppy HTML. Props are are overly verbose. Hooks are a nightmare. Contexts are gross. Portals are a stupid hack for the very, very stupid reason of making it nearly impossible for parent elements to read anything about their children. The whole "props down, events up" paradigm is woefully naive and half of React exists to get around it because of that.
So yeah, if all you know is react, you've already been on the hard path.
My advice is to go and learn a proper architecture (MVVM is a great fit in my opinion for MOST things) and then figure out how to implement that with something you enjoy using. My only bit of advice here is that the closer you stick to the underlying web platform the better off you’re going to be so if you want a concrete recommendation something like Lit here gives you all the primitives you need along with great performance, documentation and developer experience.
Incidentally, many of the issues people have with React are attributable to this mistake: knowing what a framework is supposed to be, one might expect React to take care of everything, and when it turns out you’re on your own and using React in a complex project means (whether consciously or not) coming up with a homegrown React-based framework of your own, or using a collection of disparate extras that fill in the blanks (like CRA), by then it’s already a hairy mess of a codebase in a desperate need of refactoring.
React lacks most of framework traits. The library does reactive rendering and contains a somewhat restricted set of low-level primitives around it. Even its DOM integration is handled by a separate, optional library (ReactDOM).
Someone could complain it feels like with Vercel the direction for React is to become closer to a framework (now with RSC and JSX runtime being part of React), but still it is nowhere near a state where the phrase “a framework like React” rings true—and, in fact, Vercel’s interests are aligned with keeping React more of a library than a framework so that more people use Next and are upsold Vercel’s hosting, and Next is in fact a framework in the full sense of the word, comparable to Django or Rails.
Similarly, Vue, Angular, Marko, Astro, Qwik, etc. are frameworks. They provide you with project structure, scaffolding, and a lot of batteries. The fact that many of them are, in fact, using React under the hood speaks volumes.
you don't have to though!
if you want to do more pure vanilla, understanding signals is really useful — this basically powers svelte's runs and react's hooks and whatever.
I love nanostores, a 286 byte (!) state manager that lets you build highly reactive pages w/o the weight: https://github.com/nanostores/nanostores
flexible tools like tinybase (https://github.com/tinyplex/tinybase) and unstorage (https://github.com/unjs/unstorage) are also super useful
tools like this lets you build highly reactive, engaging sites that load for under 50-100kb
That being said, as many other comments rightly point out, (for most projects,) if you work directly in the DOM you'll end up recreating what most frameworks do for you. It's really only "worth it" when you can strongly justify it: IE, if you're making a charting library that can handle thousands of datapoints, or a rich degree of interactivity that a usual business web site doesn't have.
But for a typical interactive website: Use whatever framework your peers at your company will be comfortable with.
M -> the MODEL is managed by the backend server, or in-browser via WASM
V -> the VIEW is defined by the HTML and CSS
C -> the CONTROLLER can be solely Javascript in the front-end, solely backend server processes, or most likely a mix of the two
You could even write your own browser that uses python instead of javascript. However, it would only be compatible with python html pages. We stick with JS for web development because that is how the "browser framework" is designed. HTML, CSS, JS, etc. are all just part of that framework. If you wrote your own browser, you would need to reinvent all that as well.
I do not miss the early browser days, where different browsers could have been seen a different frameworks. Things go allot better when the browser framework standardized. I still suffer from the IE trauma.
Allot of what we see with JS frameworks ought to baked into the browser framework, but that process is uber slow, because people have to a agree, and people do not like to agree. But it is magic when they do.
A framework establishes a large portion of those patterns upfront. If the framework is a popular one (e.g. React) rather than an in-house one, it makes it easier to quickly ramp up hires and have a lot of documentation/examples online. Popular frameworks also implicitly contain a lot of knowledge from years of feedback from thousands of devs exercising its APIs and providing feedback, so they're typically pretty good at solving common use cases.
Obsidian was initially built by a single developer. One of the best that I have the pleasure of knowing, but when you're one person, it's much easier to create your own pattern and have it work for yourself. They have since hired maybe 2 other devs, with minimal intention of hiring more, so ease of onboarding isn't a significant concern to them the way it would be for most companies (and "large" frontend apps more often than not require larger teams, by definition).
It uses two 500-line libraries:
This 500-line lib lets you use TSX syntax without React: https://github.com/wisercoder/uibuilder
This 500-line lib implements MVC routing: https://github.com/wisercoder/mvc-router
I'm currently working on a Company and Employee finder application with Web Components. I found them a lot easier to work with than React.
Also, I'm using a serverless platform I built from scratch. So it's literally all made from scratch with very few dependencies.
One of the biggest follies I've seen on the web is people calling themselves All of that said: learn your language, learn the syntax, and you'll be able to pick up any framework built for it with a quick read of the docs (unless the docs suck or are non-existent, then it's best to move on to the next option). Do the things that make you productive. All of this "best practices" crap is a farce. There is no best practice, just popular and unpopular opinions. Do what makes sense to you.
If you're having difficulty thinking outside of frameworks, I would suggest you work on a simple project where you don't use anything like Next.js or React. Start with the bare minimum of tools available to you through the browser API and your backend runtime of choice, then add things on as needed. Try implementing your own routing without using a third-party package dedicated to routing. Work on your own component system, state management, or even a rendering abstraction.
I can guarantee that, once you at least near completion on such a pet project, you'll not only have a better appreciation for what frameworks are doing but realize that they're actually quite overrated in a lot of ways. In reality, it's totally feasible to build an application, especially one for the web, with more of your own code and not having to think about it in terms of how your logic fits within a "framework".
At the end of the day, 99% of frameworks is someone else's opinion. This is I think what makes churn in the JS world painful. The majority of changes are based on a developer's opinion of how things should work, but their opinion isn't necessarily better than anyone else's. Your application is not worse because it doesn't use someone's idea of how an elegant state management system should behave. It's not worse because it's doing its own DOM manipulation rather than handing off all that work to an opaque rendering library. The point is to get the job done. You can make a kickass web application with freaking Backbone or jQuery if you wanted to.
It's not that I don't appreciate frameworks, though I do think it's important for programmers to learn how to move beyond them. Frameworks don't have as big a job as many are lead to believe. Their complexity is primarily arbitrary most of the time. It's not that such complexity can't be beneficial, but bypassing said complexity doesn't require a big brain.
The fact that I have been able to build a multi-user collaborative editor experience without a single additional dependency is incredible. I previously worked for a well-established and well-funded React team who had this feature on their roadmap for half a decade but still find it too daunting to implement.
Phoenix was a great reminder that a lot of the "frontend engineering" we find ourselves doing as React developers just isn't necessary with the right backend. It's HORRIFIC to look back at all the yakshaving I've done in my career already. Wrangling types (GraphQL, codegen libraries), wrangling queries and data-fetching (react-query, SWR, server components), fiddling with middleware (serverless functions, getStaticProps, CDNs). I've seen teams outright abandon testing because the hours they invested just weren't catching any of the bugs that mattered.
I'm not doing any of that anymore. I'm spending that time refining the core data model, improving test coverage, thinking about go-to-market and making money.
Phoenix may not be a good choice if your product has reached that level of maturity and product-market fit where you really should care about "microinteractions", fine-tuned animations, or advanced use-cases for an SPA like offline support and highly-optimistic UI. But I would argue that even mature products don't truly need these things. Just look at the GitHub UI. I've spent a truly astronomical number of hours in that UI and never wished I had WYSIWYG text editing, or animated skeleton UIs, or the dozen other things that the React community tells us we need.
If the answer to that question is “No”, then I’m not conceptualizing a SPA and don’t need a framework.
If the answer is “Yes”, I’d ask where I think that interaction will manifest in the experience and if I can isolate it.
If I am not looking at a SPA use-case, then my focus is on using vanilla HTML and Modern CSS with a tiny bit of javascript (as a scripting language versus as a programming language [1]). Then the remainder of the focus is on data modeling (CRUD), auth flows, and business logic development (what the application will do).
Beyond vanilla HTML, Modern CSS, and a tiny amount of javascript, if I need further client-side interactivity, I would consider something like htmx.
If I need interactivity beyond that, then we’re into SPA territory, but at this point the requirements and application have evolved to that point since we initially said we weren’t building that kind of thing.
Overall, the idea is to progressively iterate on the application, keeping the architecture and dependencies reigned in and aligned with the core objectives.
[1] JavaScript as scripting vs programming language: what I mean by this is JavaScript as a language with many warts is not so bad if its scope is kept small. I believe the issues with JavaScript arise when it is pushed to do too much. This then leads to needing to use Typescript as a matter of pragmatism. I aim to never get to the point of using so much JavaScript that I then need Typescript. I would say using it as a scripting language (to fill in gaps that Modern CSS and HTML cannot handle today) greatly minimizes the need for this progression.
Things get unwieldy at a certain scale when you don't have something like Typescript regardless of whether you have a framework, but you could start by building up small bits of the frontend stack from scratch. I learned how to do a bunch of manual raw js stuff from scratch originally, then did some professional work with YUI at a huge scale which was probably as close to an event driven frontend system as you can get. Start by figuring out how to listen for url changes, make a network request library from scratch, store data somewhere.
Imo one of the beautiful things we now get along with frameworks is a host of independent build tools that make the feedback loop of testing changes nearly instant and loosely coupled from your code.
I wrote a small internal mini-framework to follow the MVC pattern and Web Components for reusable elements. I also used external libraries: three.js for 3D rendering, sql.js for handling the 3d's models meta-data in a performant way, and @tanstack/virtual for virtualizing large lists and tables.
The biggest benefit was finer control over performance. The main downside was a less comfortable developer experience — it’s harder to find polished, ready-made vanilla-JavaScript components, so you implement more yourself.
I don't know why people continue to refuse to do this.
- Svelte seems to be fast and simplifying complexity - Laravel is likely the most complete end to end ecosystem I can't rule out. Their hotwire tech is impressive and all built in. - Rails has something similar I believe
It depends on your app too.
React definitely can be heavier than needed especially if the client experience doesn't heavily benefit from it or there's other ways to achieve it.