Well, then something's wrong. I click on different pages in the documentation and the whole page gets rerendered. Seems like it's not delivering what's promised.
It's more like ReactJS/SolidJS (but in Rust) rather than a component library like Bootstrap. Although I definitely agree the home page can do a much better job of explaining this.
But Sycamore does have ambitions to have native GUI support as well. I'm currently looking at GTK, Iced, and GPUI and see if it would be possible to add Sycamore support. This would make it possible to create GTK, Iced, or GPUI apps using building blocks from Sycamore.
I'm personally not to big of a fan of the Elm pattern for UI. Although it can be quite elegant, most of the times, it ends up being quite verbose even for simple things.
I feel like combining the drawing layer from one of these existing native UI frameworks with Sycamore could be interesting in reducing some of the boilerplate with GTK, Iced, GPUI, etc...
It's the gzipped size that matters a lot more. Also wasm is significantly faster to parse and run than Javascript so although 500kb seems like a lot for JS, for WASM it's not that much.
A general website will likely need a lot more data in the form of images and other media so all in all this is not too bad.
The reason why bundle size for JS is so important is that the browser needs to first download the JS, parse the JS, then JIT compile it before it can start running. For WASM on the other hand, the browser can in fact parse it while downloading in parallel and then run it almost immediately since WASM is much lower level. So for WASM the main bottleneck is downloading whereas JS, it is parsing and compiling.
Disagree. The bytes have to be read and extracted and the bigger it is, the more you are putting onto clients to load your page.
I'd also point out that this doesn't appear to be just simple runtime stuff like I thought. The fact that there's a 2x size difference between a simple webpage and a simple todo blog indicates this might be quiet expensive. That doesn't bode well for more complex applications.
I'd also point out that I just looked at the wasm. The bindings javascript themselves took up 37kb. That's bigger than react just for bindings.
> A general website will likely need a lot more data in the form of images and other media so all in all this is not too bad.
Yes, and a general website can function without those images. The part that makes the webpage functional is what I'm focusing on.
> The reason why bundle size for JS is so important is that the browser needs to first download the JS, parse the JS, then JIT compile it before it can start running. For WASM on the other hand, the browser can in fact parse it while downloading in parallel and then run it almost immediately since WASM is much lower level. So for WASM the main bottleneck is downloading whereas JS, it is parsing and compiling.
Before the browser can run a wasm blob, it has to download the javascript bindings which it has to parse and jit compile. It can be compiling the wasm while it's parsing the javascript, but it's not a free lunch. The wasm cannot start running until after the javascript is finished.
The browser is also free to start parsing the javascript while it's being downloaded. There's nothing special about javascript syntax that stops a browser from starting parsing while it's in flight (other than it might end up being dead).
I looked briefly, but is anyone aware of the differences between Yew[1] and Sycamore[2]? Presumably they are both Elm-influenced(?) Rust web UI libraries named after trees, but it's unclear to me why I should use one versus the other.
They differ in a similar way to how React differs from SolidJS.
In react when state changes the component functions that depended on that state are rerun to compute what the component should now look like. Then react diffs the new output with the previous to touch only the parts that changed in the DOM.
In solidjs the component function runs only once (when the component is instantiated), when state changes signals will trigger and cause the specific parts of the DOM that depended on them to change. This is generally more efficient.
The website mentions "giving you full control over performance", what are those knobs and levers exactly? What does those knobs and levers influence, and what sort of tradeoffs can you make with the provided controls?
Unlike other UI libraries, I would say Sycamore has a very clear execution model. If you've used something like React before, there is all this thing about component lifecycles and hook rules where the component functions run over and over again when anything changes. This can all end up being fairly confusing and has a lot of performance footguns (looking at you useRef and useMemo).
In sycamore, the component function only ever runs a single time. Instead, Sycamore uses a reactive graph to automatically keep track of dependencies. This graph ensures that state is always kept up to date. Many other libraries also have similar systems but only a few of them ensure that it is _impossible_ to read inconsistent state. Finally, any updates propagate eagerly so it is very clear at any time when any expensive computation might be happening.
Dioxus originally was more like ReactJS and used hooks. However, they have since migrated to using signals as well which makes Dioxus and Sycamore much more similar.
One remaining major difference is that Dioxus uses a VDOM (Virtual DOM) as an intermediary layer. This has a few advantages such as more flexible rendering backends (they also support native rendering for desktop apps), at the cost of an extra layer of indirection.
Creating native GUI apps should also be possible in Sycamore, and something I'm interested in although there is currently no official support. However, I think one of the big differences with Dioxus would be that Dioxus supports "one codebase, many platforms" whereas I think that is a non-goal with Sycamore. Web apps should have one codebase, native apps should have another. Of course, it would still be possible to share business logic but the actual UI code will be separate.
How does it compare to leptos? Leptos is roughly based on Solidjs and uses signals, to enable fine grained reactivity and avoid a vdom. Why sicamore over leptos?
With Tauri you also get the freedom of choosing frontend frameworks and can reuse existing frontend code/skills. Yes React has issues, for example Svelte handles reactivity in a much better way. I don't see real benefits of re-implementing the whole thing in Rust.
A word to the wise: similar to how foam is mostly air, Tauri is mostly marketing. Most of those 15MB "lightweight" bundles expand to 2 GB+ RAM in practice. Of course, devs still shamelessly (ignorantly, in all likelihood) call the apps "lightweight", while taking up, say, 6 GB of RAM for a 5 button UI. Tauri have also proven reticent [0] to correct the record. One supposes the sole advantage of sharing memory with other Tauri apps is not a sufficient sell to overcome Electron's single-browser-engine advantage.
A pure Rust app takes up ~60 MB for the same UI, with a large portion of that going towards graphics (wgpu table stakes).
You can't fit browser JS ergonomics into Rust and expect zero friction, because once you wire up a stateful UI with the kind of component churn you get in React, you spend more time satisfying the type system, and you also give up hot reload plus a decade of npm junk for odd corner cases.
I really like these projects but missing from them is genericity. If you're taking the time to build a WASM app in Rust it would be nice if that app could compile to something other than WASM. For example, looking at the sycamore website's source I see p, h1, div, etc. What I'd rather see is "row", "column", "text". In their source I see tailwind what I'd rather see is "center", "align right", etc.
In other words, elm-ui but for these WASM Rust apps. Building a mobile app, a desktop app, and a web app, in my mind, should be accomplish-able given the right primitives (without requiring a JavaScript runtime be bundled). Rust's multi-crate workspaces make it a really great candidate for solving these cross-platform problems. IMO of course.
Are there native frameworks which use XHTML? Regardless, a document language being used to construct complex, interactive GUIs is incidental complexity. XHTML can be a compilation target but it does not need to be a development target.
If your only target is web then there is no benefit other than a reduction in complexity.
For example, a "row" is not just a "<div>" tag. Its a div which horizontally fills its container. Centering contents with a "center" style attribute abstracts flex-box, browser compatibility, version compatibility, and the cascading behavior of CSS.
You move the incidental complexity of the web platform into the compiler which will always do the right thing. And in exchange you get the option to compile to a native or mobile app for "free".
I think I much prefer semantic elements like <section> over something like <row>. Calling something a row bakes in presentational information. Something that's a row on one screen size might be a column on another.
I agree. For my blog I don't apply CSS and prefer to let the browser's reader mode perform the styling for me.
But there are categories of application where that is not acceptable. The presentation is a tightly controlled aspect of the application's functionality. If you're designing an application with leptos or sycamore my suspicion is you would fall into the latter category rather than the former.
No browsers support streaming parsing of XHTML so you are slowing down pageloads. Especially for longer pages.
Also the ecosystem is really not there for XHTML, it never really took off. In practice it is close enough to HTML that it probably mostly works, but you are going to have problems.
The advantage is also very small, your emitter is simpler (you don't have to special case void elements and whatnot) and if you need to consume your own pages the parser is simpler. But that isn't worth much for most people.
It does make me sad, because parsing and even emitting HTML is a nightmare. But it won, so at the end of the day I find it easier to just accept that.
Yes I plan on adding doc search. Although I'm not sure if I should try to build one from scratch (never tried building full text search before) or using something prebuilt like Algolia docsearch.
If algolia is close to deepwiki as I suspect, that does not replace the original doc site: it needs to index an existing doc site before.
So adding (even a simple) search to this site would be worth it imho.
I think if you're going to use Rust on front end you're probably going to use it on back end too. In that case, I would just use Dioxus and get the e2e typing for free. What would be the benefit of Sycamore?
I wouldn't recommend e2e Rust generally yet though. I think server/API + web could work, but mobile is just boiling the ocean and will never be as good as native. You might think you can just use it for server/API + web, then do native mobile apps, but actually the escape hatches in all the frameworks I've used are not great.
Sad to say but "just use React" remains the good advice.
i've had my shot at sycamore a number of times. IMO leptos (leptos.dev) has far more fine-grained capabilities, and dioxus (dioxuslabs.com) is overall more hand-holdy but also powerful. comes with tradeoff for speed. wasm still isnt there yet (yet..) but a lot more web frameworks (including smaller rust ones) can be tracked here: https://krausest.github.io/js-framework-benchmark/current.ht...
What is "next gen" about it, is it "just" fine-grained reactivity? Or is this opposed to prev gen of Rust web UI libs, which were...? Couldn't find it in the book quickly, and it seems to not even have search...
The website is an entirely static website, and the frameworks main pitch is how good it is with reactive websites. This website could be entirely the same with html and css.
in case you don't understand what GP is suggesting: your website does not actually describe what you're providing. A "next generation Rust UI library powered by fine-grained reactivity." could mean a UI for native apps - something like egui or Dioxys - or it could mean a way to use rust to output HTML, CSS, and javascript. Or a bunch of other things. And, regardless, there's no way to look at your website and determine how to get that output using sycamore. I can inspect and see your HTML or your CSS, but there's no Rust code for me to compare that against without going and looking it up somewhere.
To be more succinct: you don't even have an image of your UI running on your websites landing page. Not one single image of the library which is, again, a UI library. People have an interest in knowing "does this look and feel like I want it to?" as well as "can I use this in the projects I'm working on?". Both of those questions should be answered by your landing page. For me, at least, it doesn't do that.
There are also plenty of other examples as mentioned in the comment above. Also many other projects using Sycamore which you can see by looking at GitHub’s reverse dependency page.
Every UI or GUI 'new fancy' package should include a very obvious Demo or link to Demo on the Home Page.
You're trying to sell me on some slick looking stuff -> Then Show It. Make it obvious. The market is crowded, people don't have time to hunt around and download it to try.
Don't make me read a manual before convincing me I should spend time on it.
What kind of examples were you expecting? There are plenty of examples in the examples/ folder on GitHub as well as plenty of other projects using Sycamore as can be seen from GitHub’s reverse dependency page