Flutter even had a DOM renderer which should in theory have fixed approximately all the issues (I don’t know for sure, I never knowingly encountered anything using it), but they’ve killed it off fairly recently, doubling down on the fundamentally unfixable direction that is the pure-canvas approach.
Flutter is unique because it is a cross platform canvas renderer with huge number of components built on top. Sort of, there are libs like QT but nobody wants to deal with C++ for GUI apps, and even the bindings leak C++isms last time I checked.
Rendering Flutter to DOM sounds like a huge effort bound to fail on some level - just use React and React native if you are leaning in that direction. Im sure Flutter can add some shims to get basic stuff like navigation working to be more web like.
That gives you a distinct accessibility tree in the DOM, which is kinda dumb as an approach. One of the reasons for going pure-canvas on the web was supposed to be performance (I don’t think it was ever true, though—only if comparing with a bad DOM implementation), and if you have to enable this accessibility DOM, you’re guaranteeing you’re doing double work. And no, on the web you can’t just do it when you need it: you can’t poll if it’s needed. All you can do in that case is to present screen reader users with a “this page is currently empty, press this button to unbreak it” button. (Aside: I’ve seen that done once. And the button was broken.)
But the problems we’re talking about are far broader than “does my screen reader work”. Do my links work properly—Ctrl+Click, middle-click, long press, hover? (This one is fixable with only mild compromise. Those that follow are not.) Does my text render correctly? Can I select it and use my browser’s context menu or other similar tools? Does content scroll properly, at correct rates, with correct inertia, without jank or at least a frame’s latency? It’s these sorts of things that Flutter’s pure-canvas approach cannot fix, and they affect, in smaller ways, a lot more people.
I’ve written more specifically about the problems here on HN quite a few times. Search and you’ll find ’em. I really should get down to writing a detailed article about it all at some point… it’s been quite a few years.
No, performance and fidelity are definitely better with the canvas renderers, especially skwasm.
Flutter is very inspired by React, but instead of React which does tree diffing to minimize rebuilds, Flutter tries to make rebuilds as cheap as possible instead - for example through its single pass layout algorithm.
The issue is that this doesn't map well to the DOM.
You can't rebuild cheaply on the DOM because you'd be constantly updating the DOM. And you don't profit from your simpler layout, because you have to implement it in JavaScript and use absolute positioning on all DOM Elements to lay them out.
I do agree that this all absolutely sucks for websites, but if you're building an App that is supposed to run in the browser like Rive or Figma, where you're going to override all click handlers anyways, or where where what you're rendering would be too much for the DOM, Flutter Web is pretty ok.
You can make text selectable, but it still won’t behave at all like normal text. Selection word boundary things won’t match the local platform behaviour, touch behaviour cannot possibly work properly, things like context menus won’t work properly, and more.
The only way you can get the proper behaviour for these things on the web is to use actual DOM text. Some parts of it could in theory be made possible, but I doubt that anything meaningful will ever change—it’s too antithetical to the nature of the web.
"Proper behavior"... You've defined this term as the precise behavior of the technology that Flutter aims to supplant.
Indeed, Flutter will never do everything "properly". But that's not its selling point. It doesn't need to, as long as the number of users reached from a multi-platform release outweighs the users lost because they were offended by a missing native feature. (Assuming you're not some massive company developing a parallel app for each platform)
Fair. If that's the deciding factor on whether you use a computer program, that's your prerogative. I assume, then, that every app on your phone allows you to select every bit of text in it?
Cool. Was hoping to discuss the different expectations we have for a program obtained through a web browser than one obtained through an app store, but I guess downvoting is easier. ¯\_(ツ)_/¯
I'm sure this is tried (I've done it myself), but a hybrid approach would take best of both worlds; heavy rendering on a canvas, the rest as SVG/DOM elements.
Even this relation can be inverted: to speed up SVG interactions, I pre-render complex path and text elements at a sufficient resolution, which are shown during transitions/user interactions, but replaced with the SVG original elements once the render loop settles down again.
> But the problems we’re talking about are far broader
No. You're fundamentally misunderstanding the broadness of the web platform.
On a broad platform even niche use cases are common. If I'm developing an application that will be used by 50 users then none of your concerns are relevant. I know my user base and I know what I need to achieve.
If compiling to WebAssembly and using canvas makes the most sense then that's what I'm going to do, especially if it means I can make use of existing business logic.
It is precisely the broadness of the web platform that makes this possible.
