Rust: State of GUI, December 2022(kas-gui.github.io) |
Rust: State of GUI, December 2022(kas-gui.github.io) |
If you want to choose a single language that can create native GUI's, Rust is probably your best or only choice. Other languages are lacking a good story for either the web or for iOS but Rust works well on both.
https://terhechte.github.io/twitvault/
Here's a review of the experience of using it:
https://www.reddit.com/r/rust/comments/zegv2e/comment/izb6nl...
Started life as a rendering layer for FlowBetween so I could put in whatever looked like it was 'winning' later on but wound up writing my own renderer as there wasn't anything quite there yet. Still has that design so another unique thing is that it's possible to use the same API with whatever rendering layer you want.
Speaking of FlowBetween, one thing I have wanted to do for ages is to get rid of the platform-specific GUIs and use something universal. It should be easy because FlowBetween sends straightforward instructions to an independent GUI layer, but I keep bouncing off for a few reasons:
- it's a big ole task so I definitely want to pick something that's stable and also lets me hedge my bets in terms of being easy to migrate away from
- most commonly, FlowBetween needs pressure data from tablets and a lot of frameworks just don't do that (this is also in a terrible state in browsers)
- lots of GUI crates are designed as frameworks and so try to dictate the entire design of any app that uses them, which is no good for FlowBetween which tries to keep its internal design choices independent of its choice of GUI
At the moment, I suspect that some sort of imgui framework is best along with an entirely manual implementation of tablet pointer data: fits with my existing design and isn't 'contagious' in a way that could make it hard to migrate to something else later on.
The listed library only uses the ancient win32 api
I use macOS, so apps should be written with AppKit. For Linux they should either be GTK or Qt, depending on desktop. Recreating widgets from scratch with GPU rendering is doomed to feel wrong to users.
You can also checkout the "feature/ui" branch in GIT [1] to get the UI framework prototype.
[1]: https://github.com/Lichtso/contrast_renderer/tree/feature/ui
UWP is deprecated, althought it keeps being the one mostly used on Windows 11, as WinUI still isn't up to its game and keeps collecting issues across all their repos.
Windows App SDK is not a GUI framework, rather the new marketing name for Project Reunion, the porting of UWP runtime infrastructure on top of standard COM without sandoxing and application identity.
[1] https://azul.rs/
I think it's much more valuable to look at this from the viewpoint of the Rust GUI ecosystem, and not compare absolute numbers where nothing comes close to JavaScript .
I'm all for RIIR but it will take 30 years before enough has been rewritten for a Rust-only GUI to be viable IMHO.
It seems Rust will be famous for the number of GUI libraries.
1. Rust is growing in popularity, and a compelling GUI solution is needed
2. These solutions take significant time to incubate, so many of the ones we're seeing emerge have been incubating for months or years, started at a time when there were fewer options.
3. The implications for cross-platform app development with Rust are profound, and offer us a way to break free from the pain-points and shortcomings of the local maximum solutions du jour: Web + Electron and Phonegap-style WebView mobile apps.
Conversely, an appalling advertisement for the state of GUIs with Rust.
I'm not sure this is the prevailing opinion. It may be something of a "vocal minority" opinion, but it does seem more common in the Rust community than elsewhere. The closer you get to traditional native app designers and developers, the more you'll hear the opinion that native look & feel (not to mention accessibility) is important to embrace.
IMO an ideal solution is to offer both: 1. to wrap native toolkits for things like text and UI controls, and 2. to enable canvas-drawn virtual controls, for those who would prefer to create their own widgets. These approaches can be enabled in parallel through an extensible widget/component system.
Writing effective cross-platform application is harder than one would think. Some time ago there was a discussion on HN where a WxWidgets maintainer stepped in to explain why some WxWidget programs were apparently inconsistent across O/S, and the explanation was that they were developed naively (I think one needs to use specific spacing abstractions, but I don't remember with exactness).
For my personal development, I use non-native (which is easier to develop), however, on large scale, it'd be a tough decision.
WxWidget was my favorite in the past (again, for personal development), however, the bindings are very spotty, which prevents adoption for several languages (wiki page: https://en.wikipedia.org/wiki/List_of_language_bindings_for_...).
It got replaced with C++/WinRT, which basically requires one to code like in the good old days of Visual C++ 6.0 alongside ATL, editing IDL files without any kind of VS tooling support and manually merging generated code.
Even the eldery MFC has better VS tooling support and COM authoring APIs than C++/WinRT.
Rust/WinRT is even years away from the C++/WinRT "productivity" in its current state, let along what .NET tooling provides, which by the way is also worse than UWP with .NET Native (e.g. no designer available, and no AOT support).
The reality is that only WinDev folks actually care enough about WinUI, the rest of us have moved on burned by these rewrites, and only a few hardcore advocates keep showing up their community calls.
I prefer desktop Windows programs to be gdi/win32, they are often the most thought out and optimized for actual desktop paradigms (i.e mouse+keyboard+high res screen) without giant margins, padding and whitespace everywhere and makes use of proper desktop class widgets like treeviews, listviews, tabs, etc.
See also: https://github.com/microsoft/windows-rs/issues/2153#issuecom...
If a person wants a native gui I guess QT would be worth investigating as well (unless one is concerned by the license).
But yes, it means C# UI with cumbersome Rust interop, if you rely on a Rust SDK.
So following your argument, pretty much all apps I use on macOS are 'wrong' as far as GUI goes.
Off the top of my head from what I use most of the time: Maya, Houdini, Blender, Fusion360, Resolve, Darktable, Slack, Discord, VSCode. Not a single one of them uses native Cocoa widgets. And I couldn't care less.
Some of these are top of the line apps for 2D/3D content creation on that platform.
If the vendors of these apps can afford to not care about UI nativeness, why should any single one or small group of developers working their asses off on GUI crates for Rust? Mostly unpaid no less.
Meanwhile I do care. My main reason being: macOS offers fantastic facilities for inspecting and scripting the native GUIs, think using the web inspector or GreaseMonkey, but across the entire OS - but of course it breaks e.g. on Electron apps. Other people will cite help menu integration, custom key shortcuts, accessibility (not only for the disabled), and - yes, resource usage. I remember being productive on a system with 256M of RAM, and before that - 4M, and before that - 64k. It's frustrating to see so much progress wasted, I shouldn't need to close ALL of the chat apps just to run StableDiffusion more smoothly.
MacOS apps feel pretty wrong to me.
Some don't remember which screen they should be on, and always go to that screen at a predetermined size and geometry (tkdiff).
Others expand incorrectly to full-screen (macvim) when moved to a different screen.
VSCode doesn't behave like the native apps.
The list goes on, I'm short on time.
> And I couldn't care less.
You caring or not doesn't make the problem go away - native widgets always feel less wrong than non-native widgets.
Also Firefox and Chrome and their forks (e.g., Brave).
As a Linux user I really don't care if you use GTK for the UI. I've long given up on customizing the look and feel of my system to the point that application integration matter.
If a smart user here today says that recent network-centric apps are OK with him visually, then maybe they are OK?
Even such "simple" tasks as how to composite a window, a video in that window, and a floating menu over that window are not very well specified in any OS (try resizing that window and watch the fun). Or, for example, your floating menu--should the parent window resize itself and paint with transparent pixels in the extra area or should that menu be a separate "window"? And, what does being a "window" even mean?
Add into that the fact that we really should be making multithreaded GUI systems at this point and it's very clear that GUIs are stuck in a local minimum that's really hard to get out of because GUI systems require so many lines of code.
There are Rust GUI toolkits like Druid with custom controls that have elements outside the bounds of the main window.
* The first time any particular window is opened, Swing draws the contents but then changes its mind about the window size, recalculates the layout and then redraws the contents again slightly differently.
* When Swing creates a window, you can sometimes observe how creates it in the wrong spot and then moves it to where it was supposed to be created.
* Alt-tabbing quickly between two windows in a Swing application doesn't always work. It sometimes just glitches and leaves you in the window you started with. (Confirmed just now on Windows 10 with Java 17. The bug has been there for many, many years.)
* When opening a submenu of a menu, Swing does nothing to handle the problem of the menu closing again as you're moving towards it but accidentally mouse over an adjacent item. Platform GUI toolkits solve this either with a delay or by tracking the direction of the movement. IntelliJ implements its own menu bar to make this work.
* Try to find an example of a window that is resizable in only one direction or only up to a certain maximum size. As far as I can tell, this is not possible in Swing, and applications handle this limitation by designing all UIs to be resizable even when it doesn't make any sense.
And there are non-native behaviors in IntelliJ too. For example you can’t close windows by double-clicking the window icon on Windows (a feature of Windows since Windows 1?).
If the toolkit is good enough it's entirely possible that a platform or platforms may adopt it as their native toolkit. In any case, these are really competing with electron which is already non-native. If Rust toolkits can get to electron-like quality but with lower resource usage and better hooks into the underlying platform then I'd consider that a success.
All of them I glady ignore on private owned computers, or use the browser version, they are anyway Web apps.
Use Electron and you'll get OS-native buttons and other form controls.
Pretty sure that's the wrong metric. Users don't care if the interface doesn't look exactly like the rest as long as it's good looking and feels familiar.
As a heavy VSCode user for example, I don't think it feels out of place on a mac for example. The UI is consistent in itself, and the UX is consistent with the global UX of the OS (more so than some native mac apps like Finder for instance).
The main reason I hated Java Swing apps was that they were ugly as hell, not that they didn't feel native.
You can blame a variety of things for this, but it's the reality. Apple doesn't really make it easier with the newer UI approaches creeping in.
That's bad... ideally there should be a runtime choice between GTK / Qt.
side note... I know many people that would prefer apps on MacOS to behave like Windows/Linux (mostly about the keyboard)... and also the reverse (those are louder to complain, but a minority in my case).
> Recreating widgets from scratch with GPU rendering is doomed to feel wrong to users.
Yes, and no. When these widgets try to mimic platform-native widgets we have the uncanny valley problem which makes everything feel out of place like you describe. But if we drop this constraint and design widgets with their own consistent style within the same app, we don't have this problem. Apps like the JetBrains IDE's and Todoist show that this can be done.
To me the middle-ground solution is a UI toolkit with its own widget style, that has some tie-in with platform-native things and conventions like windows, menus, notification trays and keyboard shortcuts. Essentially GTK or Qt without the focus on pixel-perfect matching of platform-widgets. I think this is what Druid is doing. Rust is in a good place for this as it is a modern language that works well on all of these platforms, with no native UI toolkit yet and a package-manager/build-system that supports multi-platform library configuration.
There will always be a place for platform-native apps but cross-platform apps have different requirements. I don't care that my todo list application or IDE does not look exactly "native" on my Macbook. However I do care that these applications look and feel similar when I use them on my Macbook vs on my Linux machines or Windows laptop. Preferably without dragging in a full web-based rendering engine a la Electron.
Style is just one of many things. And it's extremely hard to properly code your own consistent UI toolkit.
On top of that every platform has a myriad of platform-specific behaviors that people expect and that you will get wrong in yours. Accessibility is the big elephant in the room. But even things like secondary focus in prompts/dialogs on MacOS (that Apple's own Catalyst and Swift forget about) is a bitch and a half to get right.
Which means "unsexy" things like accessibility get left by the wayside. And your users suffer.
Writing a UI kit is a huge task. It's one of those things (like so much else in our industry) where you can fairly quickly climb "stupid mountain" by getting a pile of shiny widgets on the screen; but then you look out over the valley below and realize that, holy, crap there's a whole other mountain range of things that a UI kit has to do.
It's also a tough story for Rust, in particular, because copying what other people have done with other toolkits won't really cut it. Rust's ownership semantics and general opinions don't necessarily accord well with the highly object oriented and event-loop / object-tree structure of most existing GUI toolkits.
Aside: I still don't understand why things like Electron based apps are so bloated. They seem to static link whole chunks of their own (forked) Chromium. But both Windows and Mac OS ship native webview components as part of their system which can be used instead. I've done this myself (used Edge's Chromium webview component in a VST synthesizer, and equiv webkit stuff on Mac) and binary sizes were entirely reasonable. Linux was a slightly more complicated story.
For example, QtWidgets is native and qml is not. WxWidgets is, electron is not.
Pax composites a layer of native text and form elements on top of a canvas drawing layer. This solves accessibility across platforms, as well SEO on the Web.
This approach also enables a lean runtime footprint (<100Kb, particularly useful when targeting Web) because the runtime doesn't need to reinvent text rendering, selection, input, etc.
Still early days — have made very little noise thus far; it's pre-alpha — but you can see what's cooking at https://docs.pax-lang.org/ and https://www.github.com/pax-lang/pax
My comment said that rust had a great cross-platform story, not Dioxus. If I was doing cross-platform rust I'd use cacao for iOS, Dioxus for web front end, et cetera.
As far as I know Electron is based on Chromium. And Chromium doesn't use native controls, it has its own custom controls that sometimes look similar to native controls.
With C code often being unreadable, I found that Vala applications often have readable examples of how to use UI components. However, tooling support for Vala is extremely limited so experimenting with existing code may be a annoying. That said, I've had good experiences with opening these projects in Gnome Builder (once it got updated to work on my computer again).
If your component of choice doesn't have a Vala example, I'd resort to reading the documentation. Reverse engineering the API from the docs is often quite doable if you first look at how a component you do understand.
I've had similar problems, but worse, when I tried mixing Rust and Qt. I've never used Qt before so that doesn't help, but I found the API even less predictable in many cases, with some docs and examples several versions out of date.
I'd seriously consider just shipping a WINE'd Windows binary (using a tool like https://winepak.github.io/) for Linux targets with the current state of Rust GUI platforms. The amount of game UI/custom UI libraries out there is impressive but they all feel terribly wrong when I try to use them, in the same way most Flutter apps are terrible to use on desktop.
Granted these applications boast a ridiculous amount of features and are incredibly complex putting them on the opposite end of the spectrum compared to simple GUIs and chat/productivity applications.
To me, it's nice that platforms provide native toolkits* to provide this level of integration. But I don't view any app as "native" unless it comes from the party making the platform, as they are really the only ones that actually have to adhere to their human interface guidelines to the fullest extent.
* On Linux this is more up-in-the-air. I personally don't consider GTK and Qt as native toolkits, but the foundational pairing for things like libadwaita and KDE Frameworks which provide the associated platform widgets and HIG. Using the toolkits directly is fine and will mostly work, but it's not the same as a "native" application.
This is why Rusts' "Are we GUI yet?" and posts like this are a good thing. Eventually some new toolkit will arise that ticks all the boxes. We are not there yet and it will require a huge amount of work.
Rust is a nice language for that since it is already a "next-generation" language in terms of memory safety and its momentum is still building. Projects like Druid, while far from ready, are a good starting point.
I don’t really see the reason for multithreaded UIs either. A single UI thread works perfectly fine if you move your business logic off to a separate thread. Most people are just too lazy to do that, and start developing their app with the business logic running synchronously on the UI thread; which works for a while, but eventually reaches a point where it starts blocking the UI event loop; then they try bandaging it by moving certain pieces to a background thread and they shoot themselves in the foot with thread-safety issues. The way to solve this is not by making the UI framework multi-threaded… Simply be disciplined with UI ↔ business logic separation from the get go and all these problems go away. Yes, I know it’s so tempting to call the database directly from the button click handler - but don’t do it.
Anyone that is still around beyond WinDev themselves just has too much sunk down cost into WinUI to switch for something else, even pre-Windows 8 stacks.
It's amazing the amount of work JetBrains has put into this.
https://www.thurrott.com/windows/windows-11/260092/hands-on-...
On the topic of Rust GUIs on Mac, this was recently in the news:
https://github.com/emilk/egui/commit/e1f348e4b24c2fa83d25c6a...
It looks mostly ok, but on Linux it doesn’t support smooth scrolling (even though this is supported by gtk). You also can’t use the meta key (windows key / cmd) as a modifier key for keyboard shortcuts. So I can’t configure intellij to use all the keyboard bindings I’m used to from macos. Again, this isn’t a problem with other gtk apps. It’s just (apparently) a platform limitation of whatever Java toolkit they’re using.
So in my experience it’s 95% of the way there. I certainly prefer it over Xcode, but it has issues that native apps don’t have.
First of all, up-to-date official packages are now published to Maven Central as part of the release process, so you can just add it as a dependency as easily as with any other library. Until recently you would have had to either fiddle around with Eclipse's alternative package management system, download it manually or use someone's unofficial Maven artifact. But that's now resolved.
Secondly, if you bundle a Java runtime customized with jlink (as is recommended these days), an SWT application is actually smaller than a Swing application. When you don't use Swing, you can exclude the entire "java.desktop" module, which is slightly larger than the SWT libraries.
Basically with the HTML frameworks that use macros I'm getting incremental builds 10 seconds or more. Did you experience the same?
I have an M1 Pro Max though, which is quite a fast machine.
OT: it seemed like an awesome application so I ran it but a few seconds into importing I got
"Rate limit for Tweet Replies reached. Waiting 18446744072038294844 seconds"
I'm not going to live nearly long enough for that!
ADD: and there's no cancel button or anyway way to stop it.
And if you are on macOS doing any kind of design work, this is your mode of working. People using Photoshop or any 2D/3D DCC etc. You use all the screen space real estate on any number of screens you have. But even if I write code I run VSCode in full screen. If I have a dual screen setup I usually have one screen with two code editors, mini map & folders + terminal on screen two. But that's it.
But even assuming you run apps side by side/stacked whatever so you could actually visually see how non-native/different they look. Is that the real issue on macOS today? Personally the main gripe I have is that -- and kindly pardon the tangent -- macOS native window management is shite. One of the first things people that come from a *nix desktop buy/install, when they switch to macOS, is a proper (possibly tiling) window manager.
But back to widgets/look: in DCC apps the main sort of dialog you interact with are attribute editors where you change properties of an object or node.
I like to refer to an issue on the egui crate which has some of the best out-of-the-box support for this kind of widget composition[1].
Apple Human Interface (AHI) guidelines do not have any handrails on how these attribute editors should look or be done. The best you can do is go with dialog AHI guidelines and these fall short for this sort of window. I can go into great detail here why but it's beside the point. If you read my comment on the egui issue I linked above (same handle I use on HN) you may get an idea where I am coming from.
Most Human Computer Interface (HCI) guidlines for various platforms are based on typical late 90's early 2k desktop apps.
They were never updated to reflect on newer paradigmns for UX that evolved in recent years. Blame the fragmentation of the desktop and Electron & co.
IMHO this is the first problem the OS vendors need to solve before we can blame developers of UI libs to not make them look 'native enough'.
can you provide some examples on desktop? Touch maybe, but that is not used on desktop, afaik.
I mean stuff that is common enough to deserve a defintion and a best practice laid out in a human computer interface (HCI) guideline document for a platform/OS.
Off the top of my head:
Tab closing behaviour on Chrome/FF. Close a tab, other tabs shift but don't change size immediately so you the next tab's close button is directly under the mouse and can bel clicked to close another tab (and another, and ...).
Another one: the typical chat app with servers/groups/contacts on a list on the left and the chat on the right (and possibly threads on the far right).
There is no good standard for this, everyone cooks their own soup so stuff you learned for Slack doesn't apply in Discord etc. I.e. it's a pattern of some sort but there are no Windows/macOS/Qt/GTK interface guidelines covering this case and/or suggesting best practices. This includes e.g. platform standard keybindings to select stuff in such apps (at best Tab switching works).
Another one is use of (some) Markdown-inspired stuff in chat apps. I.e. `monospaced` ```monospaced multline``` or even ```<language for syntax highlighting> ...```.
In Slack ``` works but ```<lang> doesn not (but does in Discord, used to work in Slack in the past, I think).
I think there are many such patterns that are maybe not as groundbreaking than the invention of tabs (90's?), combo boxes/menus (also 90's? When there are too many choices for radio buttons) etc.
But beause they are used in some form (or another, unfortunately) by many apps now, the OS vendors should define a best practice for implementing their end-user facing functionality/behavior in their HCI guidelines.
Even if the platform itself doesn't provide a ready-made widget for each such case.
Hope that makes sense and explains what I meant.
1. Conditional templating, like:
<SomeLayout>
if $is_android {
//Only render the back button for Android
<BackButton />
}
/* ... */
</SomeLayout>
<SomeElement some_property={self.get_platform_specific_value()} />
Or any mix of the above.
That native Android button will be affine-transformed, clipped, and occluded so that the two layers together act as a single coherent screen, and the developer can simply position / transform / layout that `<Button>` alongside, on top of, or underneath virtual / drawn elements as if they were on the same canvas.
This "dual layer" approach is already implemented for Pax's `<Text>` element on macOS (native SwiftUI `Text`) and Web (native `<div>` + HTML text) — see: https://docs.pax-lang.org/intro-example.html
In its current state is only useful for CLI or services, unless you feel like doing WinUI team's work for free.