search

emilk / egui

egui: an easy-to-use immediate mode GUI in Rust that runs on both web and native
https://www.egui.rs/
Apache License 2.0
21.61k stars 1.56k forks source link

Tracking issue: improve Key input #3653

Open emilk opened 9 months ago

emilk commented 9 months ago

The keyboard input in egui is limited in a couple of ways:

How can I help?

Requested keys

Prioir art

Relevant issues

Physical vs Logical keys

Consider someone owning a physical QWERTY keyboard:

qwerty

However, they have remapped it to use Dvorak (perhaps even repainting the text on their keyboard):

dvorak

If they start hitting Caps Lock and then moving to the left, the keys they will be hitting will be

The logical keys makes sense for most keyboard shortcuts - if you say "Press Cmd+S to save" in your UI, the user will be looking for the logical key "S" The physical keys are mostly useful for games, where e.g. te physical WSAD on QWERTY moves a character, no matter what the users keymap is.

rustbasic commented 9 months ago

First, store the input: winit::event::KeyboardInput from egui-winit on_keyboard_input().

Then, how about providing the input value to someone who wants detailed keyboard input information in pub struct InputState, pub struct RawInput, or pub struct ViewportInfo?

emilk commented 8 months ago

@rustbasic egui is backend agnostic, and must not depend on winit

samitbasu commented 8 months ago

We use egui extensively at work, and the lack of support for the numpad has been noted (and complained about). I will take a look at this issue and see if I can help out, unless anyone else is already doing it.

CC: @emilk @twitzelbos

samitbasu commented 8 months ago

@emilk - Just to clarify... Splitting Key into a logical and physical key will necessitate breaking the integration API, correct? Since e.g., Key will become a struct with 2 fields instead of remaining an enum? Or did you envision having a separate event that sends physical keys from one that sends logical keys?

emilk commented 8 months ago

I'm not 100% convinced we should split the current enum Key. It is very convenient to just have one enum, and physical keys are also a pretty niche use case.

HactarCE commented 6 months ago

Physical and logical keys are fundamentally different things, and in every successful system that handles both they are treated separately. They just happen to align most of the time on QWERTY. I'm working on an application where many users want to bind numpad keys separately from the number row.

I would love to see egui use keyboard_types, as it looks very complete.

For an example of converting winit events to a shared representation, see my key-names crate. You can try it online in Hyperspeedcube (Help → Keybinds reference to see keys in real-time, or Settings → Puzzle keybinds and click on one of the keys to see it decode scan codes). key-names also detects the user's keyboard layout on Windows and Linux, although it's not possible on web or macOS. I don't expect this from egui but it's just so that I can show the keybinds reference with the correct labels.

TicClick commented 6 months ago

for visibility, https://github.com/emilk/egui/issues/4081 also belongs here

KarKarbI4 commented 6 months ago

@emilk @TicClick

The logical keys makes sense for most keyboard shortcuts - if you say "Press Cmd+S to save" in your UI, the user will be looking for the logical key "S" The physical keys are mostly useful for games, where e.g. te physical WSAD on QWERTY moves a character, no matter what the users keymap is.

The thing is non-latin keyboards have both latin and non-latin symbols printed. And when user wants to, for example, save document, they presses CTRL+ physycal "S" key, which logical key could be is some non-latin symbol.

I've checked web demo, input shortcuts (CTRL+A, CTRL+Z) don't work when using Greek layout.

To fix the issue, I would consider using physycal keys fallback for shortcuts processing for non-latin layouts as described here

Speaking of issue Physical keys not reported on the eframe web backend, I can make a PR, which maps web_sys::KeyboardEvent.code() to physycal_key for web backend.

After that PR is merged, I can create PR, which add physycal_key fallback for shortcuts processing for non-latin layouts

parasyte commented 6 months ago

The thing is non-latin keyboards have both latin and non-latin symbols printed. And when user wants to, for example, save document, they presses CTRL+ physycal "S" key, which logical key could be is some non-latin symbol.

It isn't quite the physical "S" key that is used for "Ctrl+S", though. It's the virtual "S", which happens to commonly map to the physical location for "S" on Western keyboards. And therefore, also shares the same scancode. Have a look at the Bulgarian keyboard layout, for instance: https://kbdlayout.info/kbdbulg; VK_S maps to "U+044F CYRILLIC SMALL LETTER YA". And its scancode is 1F (Just like the US layout: https://kbdlayout.info/kbdus/scancodes).

You might question, then, "what's the difference?" The difference is that VK_S is not always positioned on the left-side... For instance, here's one of the Turkish layout variants: https://kbdlayout.info/kbdtuf/virtualkeys; VK_S is positioned on the lower right! VK_OEM_4 sits in the position where you would expect to find "S". And if you check the virtual key map, sure enough, the physical scancode 1F maps to VK_OEM_4, also as we would expect.

The clear thing to do is exactly what was proposed:

The scancode is very unlikely to be what you want for the former case. The linked SO answer is wrong, as I have demonstrated with just a few samples.

myandrienko commented 6 months ago

@parasyte Turkish F is a latin layout though, so the logical key will be used, and Ctrl + S will still work, even though the scancode is 32 and not 1F. The problem only arises if the user has a keyboard with both Turkish F and non-latin layout (like ЙЦУКЕН), and presses Ctrl + S while in ЙЦУКЕН layout. But dual layouts like this are so rare, they are almost not a thing in practice.

As a dual layout user, I think @KarKarbI4's suggestion makes a lot of sense. An overwhelming majority of non-latin keyboard users have a keyboard with QWERTY + their non-latin layout printed. This will provide a good enough support for them, instead of no support at all which is the current state.

parasyte commented 6 months ago

Changing the layout may change the virtual keys reported. It is unclear why there is any confusion on this.

To add to this comment: any user may opt for the Turkish F layout, regardless of what is printed on their key caps. That is inconsequential to the layout.

myandrienko commented 6 months ago

Changing the layout may change the virtual keys reported. It is unclear why there is any confusion on this.

@parasyte There's no confusion about this. My comment is related to the proposed solution:

Actions that want the virtual representation of a letter or character (like "Ctrl+S") should use the virtual key.

In general, I agree with it. But the proposed solution doesn't take into consideration the users of non-latin keyboard layouts. For example, the user of the Bulgarian keyboard layout will not be able to invoke the "Ctrl+S" action.

Now, most users of the Bulgarian layout will also have US QWERTY or a similar layout set up. However, it's bad user experience to switch to another layout to invoke a keyboard shortcut. So I propose to make the following addition to your solution:

Actions that want the virtual representation of a letter or character (like "Ctrl+S") should use the virtual key. When a non-latin virtual letter is received, the action should be invoked as if the virtual letter corresponding to the same scancode in QWERTY layout was pressed.

I think that addition makes sense because on most localized keyboards QWERTY layout is printed on keycaps alongside the local layout, so users usually expect that keyboard shortcuts work in the way I described.

I hope the proposal makes more sense now. What do you think?

parasyte commented 6 months ago

For example, the user of the Bulgarian keyboard layout will not be able to invoke the "Ctrl+S" action.

There are two things going on. On the one hand, an example of a keyboard shortcut was provided for illustration without regard to i18n or l10n. And on the other hand, we are discussing keyboard events, which is only tangentially related to i18n and l10n (and accessibility in general).

We can’t really discuss i18n, l10n, and a11y without also bringing up IME. And I think you’ll agree it is a large topic without a great deal of overlap with handling keyboard layouts.

I would probably make an adjustment to the example keyboard shortcut that included common “save” commands in other languages to disambiguate what is really being proposed.

But I am adamant that scancodes (aka physical positions) are never what is wanted for normal human interactions outside of niche cases like games.

myandrienko commented 6 months ago

@parasyte Hmm, I have a feeling we have a misunderstanding here. The topic of keyboard shortcuts has very little to do with i18n. You're saying:

I would probably make an adjustment to the example keyboard shortcut that included common “save” commands in other languages to disambiguate what is really being proposed.

That's not how keyboard shortcuts usually work. No matter what system or interface language you have enabled, the "save" command is always "Ctrl+S". I use two keyboard layouts (QWERTY and ЙЦУКЕН), and I probably switch them a thousand times a day. It would drive me crazy if all the keyboard shortcuts would change depending on my current layout.

So what do you do if there's no letter S in your local keyboard layout? Well, your keyboard probably has QWERTY layout engraved on it along with the local one - so just find the letter S there and press it.

image

On this keyboard with ЙЦУКЕН layout you will still press "Ctrl+S" to save. The virtual key in this case will be Ы, and the "save" action will not fire as expected. Not good.

Take a look at this screenshot: even though Chrome UI is in Russian, keyboard shortcut hints are still referring to the latin letters, for this exact reason:

image

That does not mean however that we should just rely on scancodes. Like you said, in most cases that's not what normal humans want :) If I'm using Dvorak, or Turkish F, or any other non-QWERTY layout with latin letters, I still want to press the button with the letter S on my keyboard and for it to invoke an action for "Ctrl+S".

So, to reiterate my previous comment: when a non-latin letter is pressed, we should find the corresponding letter in QWERTY layout and invoke an action for it.

parasyte commented 6 months ago

I see that my interpretation of keyboard shortcuts in various languages is far from perfect. If it is always the case that "Ctrl+S" is for saving, then there is little reason to internationalize it. But there is a misconception that the "S" is in relation to QWERTY layout.

This is precisely why the Windows virtual key system uses, e.g. VK_S. When identified by virtual key, the key is simply called VK_S. Even if the text that will be entered is Ы, it's still VK_S for purposes of the virtual key map. The W3C's UI Events specification (and by extension winit, which is based on UI Events) is a weird mixture attempting to unify platform differences for virtual keys and text entry, and largely pays no mind to scancodes.

I think that the discussion is off the rails because the UI Events semantics have been implied on one side and platform-specific virtual key mapping on the other. Virtual key maps don't deal with Unicode at all, there isn't a Ы virtual key! That's a concern of text entry (which encompasses IME, composition keys, etc.)

parasyte commented 6 months ago

@rustbasic I'm not sure it would be wise to just jump right in, considering how much confusion there already is with existing terminology. The differences are subtle between physical (0x1F), virtual (VK_S, kVK_ANSI_S, ...), and logical (s, S, ы, Ы, ς, Σ, and any other character that can be input with this one weird key), but it's important to understand the role that each level of abstraction plays.

psethwick commented 5 months ago

I have a need for the super key (specifically I need to filter out the key combo I use to start my app .... egui starts too fast)

my first thought was this: /// On Windows, the Windows key. On Linux, the super key (often.. the Windows key) /// On Mac will always be false pub super_: bool, on Modifiers

but now I'm wondering if there's a reason not to rename maccmd to `super` send it on all platforms?