> All GUI code I've seen relies on inheritance and dynamic dispatch to structure APIs.

Rust has dynamic dispatch, by using trait objects (https://doc.rust-lang.org/book/trait-objects.html). It also has something similar to interface inheritance ("trait Foo: Bar" means that everything that implements the trait Foo must also implement the trait Bar), and one can always do "C-style inheritance" (have the parent be the first field of the child) for structs.

Given that, I'd say building a large, complex GUI framework in Rust is possible. I also wonder how a GUI framework written for Rust (instead of Rust bindings to a framework for another language) would look like.

> I would love to see something like the Flutter Engine[1] written in Rust, which is a layer tree that runs in an OpenGL context and includes text rendering, shape drawing, gesture recognition, scroll layers, a layout engine, etc. – all components necessary for modern mobile app development.

WebRender and Servo together contain all of these.

There's Conrod: https://github.com/PistonDevelopers/conrod

QML has these. Also the properties/dataflow system is a really good way of building guis. Like react but much more terse and declarative.

It's not nearly as complex as that, but I can comment on a GUI program I've been developing for the last few months using the GTK bindings for rust [1]. I'm comparing rust today with C++/Qt as it was about 8 years ago, so bear that in mind in what follows. Before this project, the last big GUI I worked on was about 100k lines of C++, with a very intricate GUI involving both standard 2d components and quite a bit of animated OpenGL stuff. The program I'm working on now in rust is much smaller (about 20 different windows and dialogs, with various normal features along the lines of entry completion and search, and is about 5000 lines of code), but I think it's big enough that I have a feel for what it would be like to develop a big app in this environment.

After the initial bumps and scratches getting used to rust, I've found the experience to be better overall than I remember it being back in the day with C++ and Qt. I actually ended up implementing my own tiny version of signals and slots as I remembered them from Qt to handle interactions between the user and the stuff on the back end (mostly sqlite at this point). That in combination with a certain relaxed attitude about using Rc<RefCell<_>> for GUI stuff has put me in a place where I'm pretty confident that any normal GUI operation is pretty straightforward with rust.

The amount of boilerplate code seems pretty similar between the two. I think that a GUI framework specifically designed for rust might actually be a big win over other options here, but so far there isn't one that I know of that meets my needs. As it is, using the GTK components from rust is easier than using them from C (in my opinion), and comparable with my memory of using Qt in C++.

I've also found that the type system and the error handling are great compared to what I used before. They help not only in catching problems, but also in ensuring that the design is complete before I get too far into it. The code I'm writing now is much more reliable in terms of handling errors and covering corner cases than what I did back then in c++/qt. There is a price for this upfront in terms of aligning your mental model with what rust wants, especially if you're coming from the same background I was. Once that is done I don't think it's any more difficult to get stuff done here than it was there, and the results seem to be quite a bit more robust.

It wasn't all good times though - the initial unlearning/relearning curve was pretty rough for me. I've been programming professionally in various languages for about 15 years, and rust has taken longer for me to get comfortable with than any other language. That's including a wide variety of languages from various paradigms including constraint/logic programming (prolog and ciao), functional (lisp, ocaml, haskell, etc.), imperative (python,ruby,c/c++,javascript) and other more obscure ones. Something about the familiar features of algebraic data types, pattern matching, and type inference combined with the strict memory model caused a lot of mental interference that took me a long time to sort out.

The build system is actually one of the biggest reasons I stuck with rust through the pain of getting started. I really love cargo, to the point now that I can't stand the thought of going back to makefiles, cmake, or any other C++ build system I've tried. Cargo seems to just work, which means I spend less time fighting with obscure linker and compiler flags and more time getting things done. I'd put up with a lot from the language just to keep the build system, to be honest.

Overall developing a GUI app in rust has been a decent experience. If you're already comfortable with the language, I'd recommend spending the time to learn the GTK bindings. Once you internalize their basic operating scheme, the time required to get GUI stuff done with rust is comparable to other languages and toolkits, but you get the peculiar combination of benefits that come from using rust itself.

1. http://gtk-rs.org

It was a bit of a struggle, but I was able to get this going for Mac w/Homebrew.

after `brew install qt5`, I had to `brew linkapps qt5` & `brew link --force qt5`. Then `ln -s /usr/local/opt/qt5/mkspecs /usr/local/mkspecs && ln -s /usr/local/opt/qt5/plugins /usr/local/plugins` which is super not ideal.

I had some error and ran `brew install doxygen --with-qt5` which may not be necessary, but after these steps it built.

Edit: scratch that. As soon as I try the example in main.rs, it fails to build again. 'ld: framework not found QtCore'

Yeah but I feel like HTML GUIs are the antithesis of Rust. Granted, it may be the best option at the moment.

This project is "dead" for quite some time now. Don't know if they really restructure the project (this was the last active conversation i witnessed in the issues section https://github.com/cyndis/qmlrs/issues/28 ) or if its just abandoned now. currently i use this for playing around with rust + qml https://github.com/White-Oak/qml-rust this is the one used in the example, too

C++ is notoriously hard to bind. The only high-quality Qt bindings I know of are Python ones (Pyside).

Not true at all! Have you tried PyQT? It was in my experience an absolute pleasure to use, and I think satisfies your main criteria. No reason why there couldn't be similar bindings for any given language you prefer, though I understand the merits of standardizing on one.

But also the point is that the Qt runtime would be part of the browser, which is usually all C++ or Rust or whatever anyway - not the web programmer's responsibility.

What makes me sad is that Qt for a very long time has had a very well thought out ways of handling the complexities of UIs, including separating view from layout from business logic, doing layouts properly, they even have a pretty and lightweight DSL for defining UIs, which you can then style with a CSS-ish language. And a non-terrible UI builder on top of that. It even does responsive UIs these days! Granted, how well this would handle something as complex as current webapps is an interesting question, but they've always kept up with the changes, no problem.

Also they did proper threads and event loops, data storage and query APIs, serialization, networking and sockets, highly performant 2D and 3D graphics (GPU-accelerated stuff!), an embedded browser, and all this was (and is) all cross-platform! It really was miraculous magic that we had this circa 2005-2007, and it's frustrating to see bits and pieces of this just recently getting web standards. If in this day and age I see one more proposal for how to do javascript promises or Commonjs / AMD / ES6 modules or something I'm going to die of frustration and exhaustion. :-(

Theres some truth around it. While QML may be nice I found the total experience of the QT stack lacking. It's still required to do large parts of the application in C++, since QML is mainly the View part. The impedance mismatch between the c++ and QML parts is huge compared to an environment that only uses a single language. I was part of some embedded system projects that used Actionscript (Adobe AIR) for some former projects and moved to QT in newer ones. For me personally it felt like a step backwards, from being able to implement the whole application in a single language with reasonable properties into a split world where I still need to use native code with all it's downsides. I don't want to say Actionscript is great here (it's for me somewhere between plain ES5 JS and Typescript) - I would also prefer a uniform Java or C# experience (as available on Android or Windows platforms) compared to a mixed C++/QML setup.

I see the main proposition of QT/QML in providing a framework that is not tied to a particular platform. The main thing that comes close are browser engines, but many people will find their huge footprints too large for using them for embedded applications (or partly even standalone desktop apps).

Using a document description language for general application development is and remains INSANE.

The browser is an atrocious, non-sensical platform that was invented to display Hypertext, not to run apps. Now look what we got? 5 new JS frameworks per month, because of some minuscule, often fabricated productivity gain. Various competing improvements to make CSS somehow manageable and an untyped, arcane and hard to optimize language on top of it. And we still struggle to make decent looking and performing apps. It's stupid.

> Imagine if we had permanently made the mistake of saddling the Web with manual, error-prone memory management.

Sounds like WebAssembly :)

Quickly skipping through the thread, this definitely sounds like it! I'd not heard about this. It looks like this definitely addresses part of my concern around HTTP being suboptimal as an application delivery mechanism, not sure how Qt ties into this. I'll have to dive in further. Thank you!