What some folks are missing is that SPAs are great for web applications & unsuitable for web pages. There is more nuance than “SPA bad”.
Then dealing with a lot of dynamic content, piping thru a virtual DOM DSL is 100× nicer for a developer than having to manually manipulate the DOM or hand write XML where it’s easy to forget all the closing tags (XML is better as a interchange format IMO & amazing when you need extensibility… also JSX just makes it worse). That developer experience (DX) often can lead to faster iteration & less bugs even with a cost to the user experience (UX). But it’s not always a negative impact to the UX–SPAs can be used to keep things like a video or music player on while still browser & using the URL bar as a state reference to easy send links to others or remember your own state.
It’s equally silly that a landing page whose primary purpose is to inform users of content takes 40s to load & shows “This applications requires JavaScript” to the TUI browser users & web crawlers/search indexers that don’t have the scale of Google to be executing JavaScript in headless browser just to see what a site has to say.
The trick is knowing how & when to draw these lines as there’s even a spectrum within the two extremes for progressive enhancement. React isn’t the solution to everything. Neither is static sites. Nor HTMX. Nor LiveView. Nor Next/Nuxt/Náxt/Nüxt/Nœxt/Nอxt.
I don’t agree with this hard split between SPAs and MPAs anymore (ie. SPAs for apps, MPAs for websites/content). In my opinion SPAs are simply a progressive enhancement for MPAs which allow even faster page navigation. All frameworks now come with SSR solutions and if a website still requires JS to show content that’s a skill issue.
Looking at Astro the line between SPA/MPA is getting really blurry. Just slap a View Transition element on your page and you got a MPA which acts like a SPA when JS is enabled.
In my opinion SPAs are simply a progressive enhancement for MPAs which allow even faster page navigation.
While I agree that there is a spectrum (hinting at that with the last paragraph), this is where I hard disagree. To construct something like this, you are making an application massively complex by trying to re-implement everything on both ends. Using something like Astro is only hiding that complexity but it’s still there, & probably full of bugs & tons of JavaScript that most developers wouldn’t even understand their stack or know how to jump into the Astro code. The amount of time saved is largely minuscule in most cases with the assets cached when navigating to a new page. In fact, I just tested two of their showcased sites which loaded slower with JavaScript enabled & the content was pretty obviously 95% static. There’s probably some niche use cases for this, but it’s not a good default IMO.
What is a web page vs web application? The web is so complex with features these days that pretty much everything is an application.
Gmail is a (bad) web application. A marketing website or even an ecommerce store are not.
I admitted it was a spectrum, but this recent article in particular does a good job explaining the axes of static vs. dynamic : online vs. offline. I think you will appreciate it. :)
I was kindof chief architect for a project where I worked. I decided on (and got my team on board with the idea of) making it an SPA. Open-in-new-tab worked perfectly.
(One really nice thing about it was that we just made the backend a RESTful API that would be usable by both the JS front-end and any automated processes that needed to communicate with it. We developed a two-pronged permissions system that supported human-using-browser-logs-in-on-login-page-and-gets-cookie-with-session-id authentication and shared-secret-hashing-strategy authentication. We had role-based permissions on all the endpoints. And most of the API endpoints were used by both the JS front-end and other clients. Pretty nice.)
I quit that job and went somewhere else. And then 5 years later I reapplied and came back to basically the exact same position in charge of the same application. And when I came back, open-in-new-tab was broken. A couple of years later, it’s not fixed yet, but Imma start pushing harder for getting it fixed.
Skill issue
Building “applications” out of HTML documents – a single one or otherwise – is the sort of thing that belongs in one of those “stop doing X” memes, unironically.
HTMX is great and is the only frontend development tool I don’t absolutely loathe. It enables lightweight SPA development, and provides a very simple and efficient mechanism for doing HTML over the wire.
Unfortunately it also kicks Content Security Policy square in the nuts and shoots a giant hole right through your website security, so if anyone on my team brings up using it I inform them it’s an instant security fail if we so much as touch it.
It’s a cute idea but horribly implemented. If your website has any security requirements, do not use htmx
HTMX comes with a variety of CSP options, though…
Doesn’t matter, the entire implementation principle of how HTMX works and what it does inherently bypasses CSP. There’s no getting around that.
You fundamentally are invoking logic via HTML attributes, which bypasses CSP
how HTMX works and what it does inherently bypasses CSP
Well, no, not really. All HTMX really does are AJAX requests to remote resources, which are performed by interpreting attributes in HTML. You specify the type of request and the target for updating. Those requests can sometimes contain parameters, of course, but any API that accepts any kind of conditional or user generated input has to sanitize that input before doing anything meaningful with it. This requirement isn’t something particular to HTMX.
You fundamentally are invoking logic via HTML attributes, which bypasses CSP
This is not true, though. You are manipulating the DOM via HTMX, but CSP has nothing to do with dynamic content manipulation. CSP is more concerned with preventing the injection of malicious code. If what you’re referring to, however, is the possibility of someone maliciously injecting HTML with HTMX that performs some nefarious action, then I have to ask (again) why you didn’t properly sanitize user input or limit the possible connection sources in your CSP.
If you have a specific example, however, of a way in which HTMX by design violates CSP that can’t be dismissed with “you coded your website poorly,” I would love to know.
why you didn’t properly sanitize user input
This is like someone pointing out that blowing a giant hole in the hull of your ship causes it to take on water, and you respond by asking “well why aren’t you bailing out the water with a bucket?”
You do understand why Content Security Policy exists, and what it is for… right?
“We don’t need a watertight ship hull for the voyage, just reinvent and implement a bunch of strapping young lads that 24/7 bail water out of the ship as it sails, it’s faster and more efficient than doing something crazy like building your ship to be secure and water tight.”
“Wow, these screen doors really suck. I’ve stuck them on my submarine, but they just don’t keep the water out at all. Some people are going to say that I’m a fucking moron and don’t understand the technology I use or that I’m too goddamn lazy to actually take the necessary steps to keep water out of my submarine, but I know they’re wrong and it’s the technology’s fault.”
In all seriousness, HTMX is a tool designed for a specific job. If you have an API that has either non-parameterized endpoints to hit or an endpoint that accepts a single integer value or UUID or…whatever to perform a database lookup and return stored values to be interpolated into the HTML that endpoint returns, then, great, you’ve got a lightweight tool to help do that in an SPA. If you’re using it to send complex data that will be immediately and unsafely exposed to other users, then…that’s not really what it’s for. So, I think the core issue here is that you don’t really understand the use case and are opposed to it because to use it in a way that is beyond or outside the scope of its established convention is unsafe without extra work involved to guarantee said safety. It also implies you are running a website with a content security policy that either explicitly allows the execution of unsafe inline scripts or which does not care about the sources to which a script connects, which is the only way you could realistically leverage HTMX for malicious ends. So, ultimately, the choice to not adopt comprehensive security measures is one you are free to make, but I wouldn’t exactly go around telling people about it.
That’s not broad enough.
If you in any way have functionality that handles anything remotely requiring security, do not use HTMX.
This goes way beyond “parameterized endpoints”.
Listen extremely closely and pray to God anyone dev with more than 2 brain cells groks how serious th8s vulnerability is:
HTMX enables arbitrary invocation of ANY api endpoint with cookies included, through html attributes, which inherently can’t be covered by Content Security Policy
This is deeply important for any web dev worth their salt to understand.
Sanitizing User input should be your LAST layer of defence against attack vectors. Not, NOT, your first and only
It’s supposed to be your “break in case of emergency” system, not your primary (and only remaining) defense layer.
Can you elaborate on that? I haven’t used it, but just assume if you host it on your own domain you can have it play nicely with csp, there are docs in their site about it. Where did it fall short for your use case?
CSP allows you to whitelist/blacklist arbitrary Javascript, and ideally you completely blacklist online js from being executed at all, such that only .js files of same domain can be invoked by your website.
This serves the role of locking down injection attacks, only your explicitly approved Javascript can be invoked.
HTMX enables invoking of logic via HTML attributes on HTML elements… which CSP can’t cover
Which means you re-open yourself to injection attacks via HTML. Attackers can inject an HTML element with HTMX attributes and no amount of CSP will stop HTMX from going “Okey doke!” And invoking whatever the attributes say to do.
This effectively shoots even a completely locked down CSP config square in the nuts, totally defeating the entire point of using it.
It’s a cute idea but what is needed is a way to pre-emptively treat HTMX as a template file that transpiles everything out so the ajax happens in a separate .js file
If we had that, then it’d be safe and secure, as the whole “htmx attributes on elements” thing would just be a templating syntax, but when transpiled it wouldn’t be supported anymore so attackers can no longer inject html as an attack vector
This demonstrates a profound misunderstanding of HTMX, and how websides in general operate. So much so that I would not hesitate to describe this as somewhere between a baldfaced lie and just malicious incompetence. You can’t “invoke logic via HTML attributes,” but you can describe it. HTMX is a client side javascript library that parses custom elements you define in your HTML and uses the data described by them to initiate AJAX calls via the fetch() or XMLHttpRequest browser APIs, which CSP explicitly covers via the connect-src directive: https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Content-Security-Policy/connect-src. It’s literally just a javascript library that parses HTML and uses it to parameterize AJAX calls. If HTMX were somehow able to bypass CSP, then every single piece of clientside JavaScript in the world could violate it.
You can’t “invoke logic via HTML attributes,”
Oh boy a semantic argument
Proceeds to describe how you can use HTMX to invoke logic via HTML attributes
Whatever you want to call it, trigger, invoke, whatever.
You can leverage HTML attributes to automatically cause arbitrary Javascript ajax calls to happen by extension if those attributes being present.
Trying to argue the semantics of this is stupid.
You put HTML attributes on shit, and the presence of those attributes in turn causes arbitrary Javascript client side logic to fire off purely due to the presence of those attributes.
That’s like, literally it’s entire shtick.
And any web dev who remotely understands the point of CSP and why it was created, should instantly have alarm bells going off at the concept of triggering arbitrary ajax via html attributes.
“HTMX doesn’t bypass CSP! It just (proceeds to describe the exact mechanism by which it bypasses CSP)”
It’s bonkers how many people don’t grok this, SMH.
Oh boy a semantic argument
It turns out the language you use can be semantically ambiguous or misleading if you phrase it incorrectly. Today you learned.
And any web dev who remotely understands the point of CSP and why it was created, should instantly have alarm bells going off at the concept of triggering arbitrary ajax via html attributes.
Oh, did you finally manage to fucking Google how HTMX works so you could fish for more reasons to say it’s unsafe? What you’re describing is not a particular concern to HTMX. If an attacker can inject HTML into your page (for example, through an XSS vulnerability), they could potentially set up HTMX attributes to make requests to any endpoint, including endpoints designed to collect sensitive information. But, and this is very important, this is not a unique issue to HTMX; it’s a general security concern related to XSS vulnerabilities and improper CSP configurations.
Do you know what the correct cure for that is?
PROPER CSP CONFIGURATION.
“HTMX doesn’t bypass CSP! It just (proceeds to describe the exact mechanism by which it bypasses CSP)”
Do you genuinely not understand that CSP works on the browser API level? It doesn’t check to see if your JavaScript contains reference to disallowed endpoints and then prevents it from running. I don’t know how you “think” CSP operates, but what happens is this: The browser exposes an API to allow JavaScript to make HTTP requests - specifically XMLHttpRequest and fetch(). What CSP does is tell the browser “Hey, if you get an API request via XMLHttpRequest or fetch to a disallowed endpoint, don’t fucking issue it.” That’s it. HTMX does not magically bypass the underlying CSP mechanism, because those directives operate on a level beyond HTMX’s (or any JS library’s) influence BY DESIGN. You cannot bypass if it if’s properly configured. Two very serious questions: what part of this is confusing to you? And, have you ever tested this yourself in any capacity to even see if what you’re claiming is even true? Because I have tested it and CSP will block ANY HTMX issued request that is not allowed by CSP’s connect-src directive, assuming that’s set.
I felt like I had a good understanding of both htmx and csp, but after this discussion I’m going to have to read up on both because both of you are making a logically sound argument to my mind.
I’m struggling to see how htmx is more vulnerable than say react or vue or angular, because with csp as far as I can tell I can explicitly lock down what htmx can do, despite any maliciously injected html that might try to do otherwise.
Thanks for this discussion 🙂
For React, you can use React Router. That doesn’t mean you’ll do it well though.
It’s tough.
It’s fascinating how some SPAs come about. Often consultancies who win some bid to implement X features. Since “good user experience” is hard to quantify/specify, it ends up being a horrible end result.
Zalaris is one such that I’m in complete awe of. Set up user flows that are expected to take 30 minutes to complete. Yet, don’t keep track of that state/progress withing your own SPA. Click the wrong tab within that SPA, and state is reset.
It’s, just fascinating.
I used to work in a consultancy that was run by a designer, and this was painfully true. I had one memory of working for a big American client on a SPA for primary mobile use, and the SPA we designed for them had several animated videos in the background, had overlays for every product, and for “speed” we loaded it all in at the start - for the dozen or so languages supported
In 2016, if someone were to tell you that visiting this product page on mobile would eat 80MB, you’d probably be fucking fuming once your data bill comes through. The designers and frontend engineers didn’t seem to care, because “it loads fine on our devices”, all hooked up to WiFi…
I think SPAs really only work for offline PWAs.