https://github.com/zed-industries/zed/tree/main/crates/gpui
See the discussion on a collection of elements ontop
Which is great, Zed's stack is pretty damn good. But like anything not yet widely used outside of its original use case, it can be rough around the edges, and just plain doesn't support anything that is not needed by Zed.
I'm betting on Slint. I kind of hated QML but it looks like they have learned a lot of lessons.
I really want to love gpui and would like to avoid using webviews, the advantage of web-based apps is that accessibility is (relatively) easy to do if you’re attentive to it. At the time I looked at it, gpui did not support screen readers etc at all.
I suppose I understand why they made that decision given their business priorities, but it’s hard for me to tell my vision-impaired friends and colleagues that there’s literally no way for them to use my app.
I love how Slint works on embedded devices too.
Egui is easy to get started with, and has a lot of widgets that work out of the box. The performance story isn't great, layout options is terrible (largely due to being immediate mode), and it is not very customizable. You take what they give you, and you like it.
Iced is a phenomenal and well-thought out API that is a wonder to work with. When it does what you need, which it often doesn't because it's not nearly as complete as the alternatives. But if you know that a monad is a monoid in the category of endofunctors, Iced might be right for you.
Slint is like Qt, but Rust. I don't think I need to say any more - if you have used Qt before, that sentence either immediately sold you on it, or convinced you to never go near it.
GPUI is a wonderful, extremely fast, stable, perfect API. If you are making a cross-platform native text editor. If you are doing anything else, sorry :(
There are a slew of projects bridging rust to run packaged browser apps which have some sort of DSL for specifying HTML front ends. People seem to like this for some reason. Some people also like being chained up in leather and whipped. I sometimes wonder if these are the same people.
Ratatui really is the best of the bunch. It has layout, accessibility, nice and easy user interface. It's a wonder to work with. Only problem: it is a TUI, not a GUI.
> Inspired by the gaming world, we realized that the only way to achieve the performance we needed was to build our own UI framework: GPUI. [0]
GPUI is not widely used because they only recently started publishing in crates.io [1] and still AFAIK their documentation is not that great yet, I suppose people will start using soon enough it, I was thinking on using it over Iced that was the other GUI that I feel that have future (because it's used by Cosmic Desktop).
[0] https://zed.dev/blog/videogame [1] https://crates.io/crates/gpui/versions
// Vidrun, born of the sea-wind through the spruce
// Vidrun, green-tinged offshoot of my bough, joy and burden of my life
// Vidrun, fierce and clever, may our clan’s wisdom be yours:
//
// Never read Hacker News
// - Aphyr, "Hexing the technical interview"
if (document.referrer.startsWith("https://news.ycombinator.com")) {
document.location = "https://upload.wikimedia.org/wikipedia/commons/d/d4/Human_fart.wav"
}
This topic comes up often, so I wrote a blog post explaining why I think a DSL is a good fit: https://slint.dev/blog/domain-specific-language-vs-imperativ...
Why Rust? Because I like its power and with recent developments such as subsecond the compile times are absolutely negligible.
Whereas when kovid made the kitty terminal protocol, it got implemented by the next 3 or so new terminal emulators. You don't get this kind of modernization from any other UI solution save browsers. And no way do I want to do web development for my own stuff. I suppose I could target Dillo or something like Servo, but I'd be more inclined to consume the base X11 protocol, as Vidar H did.
My Ruby terminal protocol implementation has smooth abstractions from methods performing the bare VT control calls up to semantic methods with better naming, to flexible and extensible text field abstractions. All the concerns are separated nicely and Ruby lets me scope everything just right, so I can call a debugger anywhere as the necessary logic is scoped module-wide.
I simply cannot imagine having any kind of fun doing this in a static compiled language. And I'll be able to build on top of this to make things that aren't editors.
It made the post more amusing, actually. I sighed when I saw "Windows Narrator" suddenly.
I think writing a useful GUI has considerable overhead no matter which technology you use. In addition they cause other difficulties, like testability, i18n, l10n and accessability.
This is why people often resort to command line tools, rightfully so. There are cases, however, where a CLI won't cut it and I believe TUI's are a nice and lean solution that sits right between CLI and full-blown GUI and isn't going anywhere.
There's no standard to communicate TUI semantics to assistive technology, and whatever few standards actually exist (like using the cursor to navigate menus instead of some custom highlight) aren't followed.
With GUIs, those standards exist, and are at least somewhat implemented by all major (non-Rust)UI frameworks.
C| Column Wide |
o| |
l||
1|2|
> This is absolutely not true for a GUI which could easily just be an opaque rendered canvas. Are you not aware of OCR? Besides, GUIs have special accessibility tools, which almost none of the TUIs have, so your opaque canvas isn't universal.
> The fact remains:
That's a myth, not a fact, and you fail to establish "the fact" even in the most basic example
What’s actually hard with screen readers isn’t getting text (that’s been easy on most GUI systems for decades) but communicating things in the right order, removing the need to see spatial relationships or color to understand what’s going on.
TUIs make that harder for everything beyond mid-20th century-style prompt / response interfaces because you don’t want to have to reread the entire screen every time a character changes (and some changes like a clock updating might need to be ignored) so you want to present updates in a logical order and also need to come up with text alternatives to ASCII art. For example, if I made a tool which shows server response times graphically a screen reader user might not want to hear an update every second and if the most interesting thing was something like a histogram I might need to think about how to communicate the distribution which is better than rereading a chart every second only to say that the previous one has shifted to the left by one unit and a single new data point has been added.
Those are non-trivial problems in any case but they’re all harder with a TUI because you’re starting with less convention and without the libraries which GUI interface developers have to communicate lots of context to a screen reader.
There's no protocol that tells a screen reader to say something different than is actually displayed on the screen. The best you can do is having a whitelist of screen reader process names and changing how your TUI works if one of them is detected, but that's brittle and doesn't work over SSH. You'd also have to think about how to do container escaping and interfacing with the host system when you're running in WSL, as the screen reader is almost certainly on the host side.
Most TUIs don’t have these either. So I don’t see this as a difference between TUI/GUI. If you want to make a GUI and want to ignore these things, you are free to do so.
An old school DOS TUI reads keyboard input one character at a time from a buffer, doesn't clear the buffer in between screens, and is ideally laid out such that a good part of the input is guaranteed to be fixed for a given operation. They also were built without mouse usage.
So an operator can hammer out a sequence like "ArrowDown, ArrowDown, ENTER, Y, ENTER, John Smith, ENTER" and even if the system is too slow to keep up with the input, it still works perfectly.
Modern GUIs almost never make this work near as well. You need to reach for the mouse, input during delays gets lost, the UI may not be perfectly predictable, sometimes the UI may even shift around while things are loading. Then also neither does Linux, I find that the user experience on DOS was far better than with ncurses apps that have all kinds of weirdness.
I find egui far easier than Ratatui.
Haha!
In the linked article the author had trouble building Qt for windows and because they wanted to go through 40+ GUI libraries they skipped it instead of pushing through. That's why it doesn't show up in final short-list of good options.
In my books (writing Rust since 2013, but haven't done any GUI work in it), Qt and Slint are the best options out there.
Every GUI I replace with a TUI is one less flow-breaking obstacle where I have to suddenly reach for the mouse.
1. Consistent font and sizing. Pick the font you find easiest to read and set it in the terminal, now all TUI apps use it at the correct size.
2. Consistent theming. TUI apps use the same color scheme as your terminal, as your vim, etc. A consistent "desktop" is less distracting. Especially when you don't have to deal with crappy Electron apps displaying animations and ads (looking at you Discord).
3. Efficient on screen real-estate. Most TUI apps don't waste a lot of space on "padding", "giving elements space to breathe" or other "correct UX patterns". They tile nicely.
4. Never and issues copy and pasting. I used a GUI chat app recently that wouldn't let me select text, seriously.
5. Did I mention they are fun to use and relatively easy to develop?
And copy / paste, "hm does this TUI intercept mouse clicks, ah it does, oh what was the key combo for my terminal emulator that allows to skip that? Crap I pressed ctrl-c instead of ctrl-shift-c". Or worse when you want to select text in a column-based TUI and your terminal emulator doesn't have any sort of column-selection handling.
By default, TUIs have bad discoverability unless the developer puts in the effort. But at least you have keyboard navigation and run in a terminal.
By default, GUIs have bad keybindings unless the developer puts in the effort. In the worst case they aren't even kb navigable. But at least they tend to be discoverable.
TUIs have to be navigable by keyboard. They could be garbage in every other way, but at least they have that.
GUIs might have keybindings as polish if the developers put in the effort. Otherwise they either don't have keybindings, or they rely on generalized keybindings that come with the OS/UI toolkit that can be bad/impossible in arbitrary ways, like if tab is swallowed by an input so you can't even tab past a component or dismiss a modal. Or the only GUI is inside a web browser.
> They could be garbage in every other way, but at least they have that
Not really, they could also not implement keyboard input for some interactions and only react to mouse. You could have a click-to-open-url in your terminal without a keyboard fallback setup. Of course, you could set it up, but that would be "polish if you put in the effort" There is no limit to garbage.
> Also once your workflow is increasingly keyboard based (like when using a tiled window manager),
Exactly, for example, you've set your Windows key to be reponsible for all window-based movements, so you have it move/resize app windows and want to have the same key help you move/resize "inner windows/panes/columns" in a TUI app. Well, tough luck, the terminal doesn't even recognize this modifier!
Again, "as polish if you put in the effort" you can achieve anything via some external remapping to supported keys, but "otherwise" you don't have keybindings, and don't even have generalized fallback common (though not universal, again, there is no limit to garbage) to GUIs.
Or you'd be unable to bind some key combo because in a terminal that key combo is equivalent to regular typing and you need to timeout to differentiate between the two, so a permanent friction.
> tab is swallowed by an input so you can't even tab past a component
Of course you can, use your keyboard to move the mouse pointer and click outside of the input field. All pure organic keyboard-based interaction! "They could be garbage", but at lest you have that keyboard-based option (yes, only "polish/put in effort")
I'd probably put them in a third, discarded, mutant category where they are mouse-first applications that run in the terminal. Thus like GUIs, keyboard-support is bolted on.
So a better distinction here is mouse-first vs keyboard-first, but I think "TUI vs GUI" is close enough to that and it kinda stands in for that distinction in most cases.
Vim, tmux, and weechat would be examples of keyboard-first TUIs where mouse support is bolted on as polish, but never required.
Where you lose me is that every example you can think of where a bad TUI might not perfectly mesh into a keyboard-only workflow, the GUI counterparts to that example is much worse. So we would just agree that it's a bad TUI rather than anyone going "you're right, the average GUI is much better if keyboard-driven control is my priority".
I don't understand some of your examples though. Your super key is always at a global level that you wouldn't expect apps to read, that's the point. And using timeouts to handle keybinding collisions or something? Dunno what that looks like nor why you would do it. The examples are like if I said GUIs suck because it's a dreadful experience if you insist on using your mouse upside down.
It's indeed a better distinction, but no, TUI vs GUI is nowhere close, specifically for your explicitly mentioned use case of "workflow is increasingly keyboard based" because that's exactly where GUI shines as it has much fewer fundamental limitations in keyboard handling.
> than anyone going "you're right, the average GUI
That's your strawman framing, why would I care about "the average"? It's not a limit for anyone, one can always choose a better-than-average app. My point is that at the top there is no comparison because TUIs offer nothing and have much bigger fundamental limitations. So of course with the wrong framing you can find worse examples.
> And using timeouts to handle keybinding collisions or something? Dunno what that looks like nor why you would do it.
Terminals send Esc for Alt-X combos, so to differentiate those combos from a single Escape press you need to track timers (if no extra key signal, then it's Esc, if another key signal, then it's a combo). And Esc is one of the most used keys in e.g. Vim
> Your super key is always at a global level that you wouldn't expect apps to read, that's the point.
Of course I would! What I wouldn't expect is for the app to override any global OS function bound to that key, but I would definitely expect to be able to not waste a whole thumb-conveniently located modifier just because
That's very close to my point - you're used to some trivial keyboard workflow, so wasting the whole modifier is just fine, you can't even comprehend why someone wouldn't want that, it's like a mouse upside down!
While most gui ones also have hotkeys, most people only remember a tiny selection and have to rely on navigating menus and dialogs to access stuff they can't remember which makes their usage very slow.
Menus are also not slow by any measure, they're only needed for discoverability, can be navigated with the keyboard as well, and aren't that different from the menus many TUI apps have (e.g. Midnight Commander). Moreover, there are real novel ideas like Microsoft's ribbon, which basically combine vi-style input and discoverability. Hotkeys is just one way to drive a GUI app using the keyboard, out of many.
2. specifically for "keyboard-centric tech users" that's a big fail since the terminal platform is not capable of supporting advanced keybindings presisely because all they do is "target the lowest common denominator" (as defined in 1965) of keybinding support!!! So your cool setup from a GUI code editor is simply not transferrable.
3. this is very rare, I mean, just look at the screenshots, a lot of them couldn't even add non-gapped borders=== ---!
2. Again, it's not about the possible capabilities. Objectively, GUI programs contain a superset of possible TUI programs (since your GUI program could be a terminal emulator). It's about it being a shibboleth: I can be pretty confident that the TUI application will have a keyboard-based workflow, and I can be pretty confident that the TUI application will be targeting a technical user base, because those are the only people building and using TUI tools. The "lowest common denominator" in that sentence was referring to the users not the systems.
2. but what's the value of a shibboleth if it's not reflected in capabilities??? Why would you care that a TUI app targets "tech user base" if you also know it will be less capable in your 'keyboard-centric' workflow due to platform limitations? And LCD for users just doesn't make sense, what user quality are you min-aggregating?
I am building tooling around "Verifiable Credentials" (the W3C standard, OpenID standard, and all the standards around and below it).
So me and my co-workers need tools to check JWTs, resolve DIDs, generate Proof of Possession, .well-known contents and services discovery, do OpenId Connect flows, interpret "offer requests", "presentation requests", QR-codes, etc. It started as a bucket of random commandline things that random devs whipped up when needed (in typescript, python, bash, rust, some PHP), and now slowly consolidated into a consistent "toolkit". I am currently still porting most of these tools to Rust and make their CLI interface and IO consistent.
But from there I'll be adding a TUI very quickly.
By no means meant as "end-user" software - those exist and called "wallets". But for developers and devops working in this niche. These don't need a GUI. I'd even wager these users don't want a GUI but prefer a TUI, but in any case, the TUI is just so much easier and more accessible for me as a dev that it's the choice between a TUI+CLI or no UI, just CLI.
My reference in this are Delphi .dfm files, which you rarely have to edit manually because they're handled by the IDE form builder. It is still a fantastic IDE to build native UI apps and I don't see a reason why Rust couldn't be used the same way.
I have to agree with this, but for a lot of different things other than GUI as well. My friends all want me to use Rust, but I moved back to C++ after trying to do a lot of different projects in it after finding every crate I needed to be a WIP or abandoned state. There are the massive crates that are super popular of course, but it seems that once you get off the popular beaten path the available crates becomes far worse than any other language I have ever used.
Of course comparing Rust to something like Python or C++ is unfair to Rust, since Rust has had less time to develop those packages, but I think its more about the community at this point. I just don't see any enthusiasm for working on crates/frameworks. Most of the time when I find a crate I need, its sitting at version 0.3.0, last updated 1-3 years ago, with the documentation simply stating "Reimplementation of X in Rust, go read X's C++ documentation on how to use."
GUI is kind of unique because it's a really hard problem - both a ton of work and also a bit awkward to make ergonomic in Rust.
> last updated 1-3 years ago
That's way too low a bar. I bet most of these niche C libraries that don't have Rust equivalents are similarly slow-moving.
GNU Make regularly goes 4 years without a release but it's still alive.
"I like the A tool in category B. System C doesn't have tool A; system C is dreadful in category B".
I think you could have made the acute part of this argument about GUIs in rust 3 years ago, but not today!
I need to try Slint and GPGUI. I love EGUI, but need to do a proper comparison; could be missing out. This seems to be a concern of a different nature from the one you describe!
Because it is quite easy to do.
There are also TUIs in other languages, just Rust is more cool.
It's no less or more equipped than C or C++, so not really sure what to make of this comment given the username lol.
[1]: https://upload.wikimedia.org/wikipedia/commons/d/d4/Human_fa...
Rust is simply not meant for GUI-based data design but I still want Qt in Rust. That's it. Not QML or Slint. No markup at all. None of the immediate mode things. No other languages. Definitely not GTK. I'm worried it will never happen for Rust and it will be such a missed opportunity.