Does it not have the following potential drawbacks:
The GUI has to be written in JS (or compiled to web assembly)
The data has to be copied back and forth between the app and the front-end hosted by Webview.
The communication between the app (back-end) and the Webview part has to go via a network port (loopback), which is potentially slow (and hackable).
Good questions!
Writing GUI directly in HTML/CSS/JS is one of the very few ways to provide fully consistent UIs with responsiveness, among completely different devices, resolutions, DPIs and operating systems.
It works consistently on every minimally capable device in the world, today and in the future.
Having an independent layer for the UI is a good step in direction of separation of concerns, thus, it’s a good architecture practice (like front-end apps that communicates to back-ends through their APIs).
Also, it helps with mobile app encapsulation through Cordova or Capacitor. It can be extremely lightweight and fast as native when (1) reusing the mobile OS web engine (as webview on dekstop) and (2) when using a good processing model on JS side, like Svelte does with real reactivity.
We can use the same strategy of Tauri (see security), where they use the webview library but with no server nor open ports, using only the two-way language bindings for communication.
It’s fast and secure.
Also, the Crystal webview port from naqvis (GitHub - naqvis/webview: Crystal bindings to Webview library) provides this two-way binding as well. So, no server is needed.
Would it not be possible to generate JS or TS or Wasm from Crystal in order to have only one language server side and client-side ?
Does Crystal compile into JS or Wasm ?
How would you pass the data back and forth : JSON or something similar ?
If you use a framework like Axino (Getting Started With Axino. A crash course in app development using… | by Serge Hulne | The Startup | Medium) or Mint (https://www.mint-lang.com/), it will minimise the number of languages to deal with (client-side), otherwise you’ll have to juggle with Crystal + HTML + CSS + JS.
PS: The advantage of Axino (my experimental TS framework) is that an app written with Axino is 100% TS, so:
But as I said in my previous post, it would be even simpler if it was 100% Crystal (if crystal could generate the JS or Wasm code needed on client-side).
Since Webview for Crystal also has an eval() function in addition to the bind() function, it is theoretically possible to skip JS entirely by generating the dom elements one needs from Crystal, as follows Question : example of use of : webview::Webview#eval · Issue #6 · naqvis/webview · GitHub
This would make it possible to create apps in 100% Crystal (no HTML, CSS, JS, etc. needed for the users or the library).
In other words, it is possible to create 100% isomorphic apps (i.e. using Crystal on the « server-side » as well as on the « client-side »), hence resulting in a true app development kit 100% in Crystal.
To to put it another way : by using eval() and bind() the dom and its behavior (responding to events) can be written in Crystal, provided that the one creates a library which creates all the needed DOM elements from :
Crystal + eval() + bind()
Another way to look at it is : All that is needed to achieve an isomorphic app développement kit in Crystal / Webview is a an equivalent to Axino (or something similar) translated from TS to Crystal (which is a bit tedious to do, but not complicated).
This an interesting and also technically valid approach :)
I did the same thing for Golang (build a isomorphic GUI kit for Golang, which wraps JS code into Go functions):
And it works as expected.
(I have not published that code, because it is not ripe, but as an experiment it works)
I will try to test the concept with Crystal, as soon as I can install Webkit for Crystal on my Mac (does not work at the moment: I filed a bug report).
Actually Webview for Crystal works fine (I was using an outdated version of Xcode)
Little late to the party but here it goes.
As of 2024, I’ve been working on a GUI library from the ground up.
It has fairly different design goals than other GUI frameworks that I’ve seen and the layout rules are quite different.
Features include
The library privileges the ability to easily create components as opposed to providing them. There is also support for font clamping/wrapping.
I’m in the process of learning a lot about a lot of technology I’ve taken for granted (a lot of ground to cover), so conversations are preferred over PRs and development will be a medium-slow burn.
If anyone wants to learn more, I put together a demo for a friend. (Thanks to nuclearbananana from reddit for compressing it). There are a few misteps and inaccuracies, but you’ll get the gist easily.
https://file.skinnyjames.net/hokusai-demo/index.html
Project is on codeberg
Why do you define UI and layout through templates (loosey strings, full of typos, which you need to parse at runtime?) and not through some similar nice DSL (pure code, checked by compiler)? What led you to this design? Thanks!
Hi, thanks for the criticism. I’m not sure what you mean by loosey strings full of typos, but the templates are parsed on mount. once. Decoupling the template from the code means that the code can focus on behavior and the presentation requires much less typing. Does this help?
Really happy to see non-html non-JS GUI work happening in crystal!
I hope you continue to have fun and keep working on it!
Another one that is not really a GUI framework but rather a step below it is GitHub - yxhuvud/wayland_client: A basic wayland client library. , which provide wayland client bindings with a thin convenience layer on top.
A post was split to a new topic: Following users
I’m not sure what you mean by loosey strings full of typos,
I guess i can use some doc for lucky (it write template use purl Crystal) for explain what pfischer means.
Lucky uses Crystal methods for rendering HTML. The Crystal methods map as closely as possible to how HTML is used.
Using Lucky HTML adds an additional layer of type-safety, is auto-formatted with Crystal’s formatter, and can be much easier to refactor with regular Crystal methods as HTML gets larger.
Reading this thread, I can see the history of GUI development in Crystal.
At first, people asked for large and advanced toolkits. Later, expectations shifted toward smaller and simpler tools. Many GUI projects do not last long. Developers who use programming for work need the best tools to stay competitive, and they don’t have time for Crystal. So in the end, only the projects that can survive remain, even if they look a bit rough.
What I want is something small — just enough to build a remote control with a few buttons. For that reason, I rebuilt the old libui and made a library called UIng. There have been many libui bindings before, but I added some improvements:
Box, even for APIs that expect structs with function pointers.postinstall.Because of libui’s limits, UIng is not the easiest library to use. But for the right cases, it is “good enough.” The real challenge is to keep maintaining it, and that is the hardest part.
(Translation using ChatGPT)
