Raw keyboard handling in Unix terminals | Hacker News Reader

Raw keyboard handling in Unix terminals

sw.kovidgoyal.net

115 pointsdrran52 comments

This feels like the computing equivalent of epicycles in medieval physics.

Will we ever just admit that the vt100 tty is just no longer suitable as a basic OS abstraction, and needs to be torn out and replaced? Or in 50 years will simple interactive programs still not be able to detect keyup without importing a bunch of bloat from x11/wayland?

I suspect in 50 years that we will have lost control of the local compute end of things. Our new "dumb client" will be some future iteration of a web browser. And you won't be allowed to run anything outside of it. You'll be able to see keyup, but most of everything else will be on somebody else's server.

Over the last 50 years there have been a few cycles between thin and thick clients (in different contexts) so hopefully it's not all one direction from here.

I agree, though this time there's big money behind taking local control away.

Which is why, as long as there's hackers, there'll be another cycle in the other direction.

They aren't having much luck with the iPhone...

The are! It's just not an everyday occurrence - and so is worth much more - so most zero days end up in nation state hands.

Take what i say with a grain of salt, (I’m not an iPhone hacker and I follow their work semi-loosely but from my understanding there are a number of available and documented tethered exploits. I wouldn’t doubt theres a number of untethered exploits that are waiting in Apple’s bug tracker system, or are simply too abusable for the general public to know about.

I wonder if these cycles have been studied.

The thick clients, though, seem heavily tied to mechanisms that still prevent local control.

If anything the vt100 ttys are just as suitable as ever. Command line interfaces are superior even today because the step from interaction to automation is as minimal as possible. I predict that the futuristc VR GUI of tomorrow will still be just a bunch of vt100 terminals floating in space.

Can't we dream bigger though? For instance, I feel like templeOS's command line is unironically conceptually superior than the linux commandline or any other unix. I should be able to mix text with 3d models or images or sound, pipe them between programs, etc. I should be able to pause a running program, save it to disk, and resume it later. There are all sorts of things that I should be able to do but can't, because of the deliberate unix fetishization of 1970s era computing.

>Command line interfaces are superior even today because the step from interaction to automation is as minimal as possible.

Agreed completely. In fact, the OS should force GUI programs to all speak a coherent language so that you can automate them and pipe data between them. And if the machine is going to serve the user, the user should be in control of the GUI, they should be able to move buttons and UI elements around into the configuration they prefer. This would only be possible if GUI programs had minimal control over actual UI layout, which would make webdevs angry. But given the state of web UI, I think that's an acceptable sacrifice.

Much of the above was possible in lisp machines. But sadly I don't think it would be possible to create such an OS today (unless the "OS" is running in a linux VM and using linux for the actual hardware interface) due to the sheer diversity of hardware. And the closed source nvidia driver would ensure that you never got cool graphics.

In my opinion the mistake was to move away from the terminal in the first place. Many gui commands like "draw button in bottom right corner" could be done via escape codes and creating a gui program would be as simple as writing a bunch of commands to stdout. As far as I know the earliest prototypes of X worked just like that.

Well, X developers must've abandoned that idea for a reason.

X was meant to be a lowest common denominator for Unix workstation vendors to base their proprietary software, window managers and desktop envrionments on. Too much standardization at that point would have meant less distinguishing features for their products. Today all major unix workstation vendors are dead which shows they were clearly wrong.

There is a library for doing things like that; for some reason, it's called "curses".

I believe that's how the MGR window system, from Bellcore, worked. I fiddled around with it briefly on an old Sun Workstation many years back.

https://en.wikipedia.org/wiki/ManaGeR

I think the problem is conflating "dreaming big" to "no restrictions." A large part of the strength of the current terminal approach is specifically its limitations.

That is to say, a large part of what makes the automation capabilities of most terminal applications so strong is specifically that the idea of character streams over raw binary streams being the default.

Can I dream of a lisp machine that lets me do what you are suggesting? I mean, sorta. Early HTML based apps came pretty close on this, but folks regularly refuse to limit themselves to a base set of constructs. Is why it would be easy to user style HN, but basically impossible to do the same to gmail. (gmail, amusingly, created their own theme capabilities...)

> no longer suitable

No way, it's way more popular and useful than ever.

We just need to ditch vt100 and settle on some sort of xterm variation as the new standard.

This won't have much effect, since xterm is an DEC VT emulator.

A superset though. It's perfectly fine to support ANSI and VT-100 if we can add new escape codes to switch to a standardized raw mode, just like has been done for 8-bit and 24-bit color, and for mouse support.

If I read this correctly, this proposal aims to solve just that kind of stuff while remaining compatible with all the legacy software in use today.

This is awesome. For a long while I have wanted to write character-based applications with greater control of the keyboard; think "a terminal tetris that can recognize key combinations such as Alt-I, Ctrl-J, Alt-Ctrl-PgUp, Meta-Ctrl-F14, etc." This should work OK whether running locally on the console, on a (serial) terminal, on a graphical (X Windows / Wayland) terminal, on a remote (SSH) terminal, etc.

I wish something like this was also created for the mouse, so that you could use your mouse everywhere (including the console without any graphical environment running), remotely over ssh, etc. I see no reason why a similar protocol could not transmit mouse events over ssh and allow you to interact, using your mouse, with a remote character-based application.

I would love for such proposals to gain traction and be accepted everywhere.

There are already several standardized mouse protocols available some even with pixel precision. The most complete implementation of those protocols has xterm. They can be tested with a program called "vttest".

Have you heard of gpm[1] - general purpose mouse? Slackware offers to install and enable it by default, even these days.

It's a terminal based mouse server. ncurses supports it so it's theoretically possible to do all the same things in a TUI that you would in a GUI with a mouse.

[1] https://en.wikipedia.org/wiki/GPM_(software)

GPM is for console, not for terminal. I used it a lot in virtual console, before switching to X11 based terminal emulators.

The thing I would like to see is this making it into terminfo entries, meaning that it is possible to detect that terminal supports this behaviour, otherwise while great it is terminal emulator specific.

Excellent to see, the ESC ambiguity problem is one that has especially bugged me.

I hope there's a plan to standardize this with the relevant organizations! I worry that's the only way to see widespread adoption.

Yes, I can imagine a mostly-backwards-compatible, but much improved, standard; let's call it VT-2021.

If you are native English speaker, then start with kind of "caniuse" site, but for terminals and VT, please. It will help a lot if we will be just aware of available features in existing terminals and their compatibility with standards.

Not a web site, but terninfo and termcap databases, normally present on a Unix system, have this information.

I would hazard to say that some basic subset of VT100 commands is "universally" supported, even on Windows if one uses Windows Terminal.

Various obscure high-end features from 1980s, like programmable font bitmaps, or sixel support, much less so.

Also, modern terminals support Unicode, including emoji (as double-width symbols), which was unimaginable in 1980s or 1990s.

> if one uses Windows Terminal.

This is also true of the Windows Console as of Windows 10. Any VT100 (also VT220, VT4xx, VT5xx, xterm ...) support we add to Terminal eventually flows out to the console host integrated into Windows. :)

How to display terminfo in a consumable way?

What about non-standard features, like drawing, sixels, sound, mouse, touches, clipboard, fonts, true color, transparent background, color emoji, double size text, etc.?

And yet, I feel like that database doesn't succeed in replacing a website that tells me "the following terminals support OSC 8 and with what awkward limitations", which is knowledge I find both useful and important for web browsers despite being able to do runtime tests to determine the support of most features. "Oh, it turns out 78% of my users wouldn't be able to parse OSC 8 even if I went to the trouble of using it... maybe I'll work on a different feature today".

"Emit the escape sequence CSI < u at application exit or just before leaving alternate screen mode to restore the previously used keyboard mode."

What happens if the application crashes without emitting this? The user is dropped in to the shell and the terminal and shell might misbehave together?

I guess same thing that happens to any application that sets weird modes. You should use 'reset' command which just emits escape code to reset terminal and terminal emulator hopefully sets everything back correctly.

If the application switched to key press mode, you're not going to be able to type reset.

Well, from what I've read all alphanumeric can be typed just fine, they go as is to terminal, problem is with special keys that some of them emit escape sequences that confuse shell or combinations of those.

But not to trust what I just read I've installed kitty and tested it and you can type in the "reset" just fine. I hit a bit of a problem with "Entering" it, as apparently if you have numlock enabled (which I do on login) pressing "Enter" sends escape code: "^[[13;129u", but if you disable numlock it is typed in as per documentation: "0x0d - for the Enter key", which in any ascii table is Carriage Return.

And yes, issuing reset restored normal terminal operation.

If you enable all key events there's going to be a "key release" event between the r and e and all subsequent letters of reset.

If the terminal is in a completely invalid state, closing it is always an option

Some terminals can issue a reset from "the outside", e.g., iTerm2 has this under ⌘K. (In fact, `reset` inside iTerm2 fails to completely reset the terminal…)

Every now and they I'll want to do ^K in Vim and inadvertently hit ⌘K instead, which is fun. (`:redraw!` sorts things back out. Though weirdly it occurs to me that now that the terminal seems to somehow remain in alternate mode…)

Yes, in such a case, use the terminals native reset capabilities. And hopefully someday shells will learn how to interpret the full keyboard protocol, making it possible to type reset in any mode.

The usual name clash rant. Why kitty? KiTTY is a fork of PuTTY and has been around since at least ~2007 or so

This is written by the KiTTY developer. My guess is that when you open up a kitty session, you're not trying to open a subsequent one inside. So running `kitty` inside a session will give you a preview of the new protocol being proposed here

Are you certain? This looks like it's written by the author of the newer kitty terminal emulator, not the older KiTTY ssh/telnet client derived from PuTTY.

You know what, you're right. It's the terminal emulator, not the putty fork. I guess my mistake proves softblush's point

Question: How does this differ from Paul Evans's solution in libtickit, as discussed here[0]?

(Yeah, yeah, I can read the whole thing and see for myself, but I'm hoping someone has a quick answer.)

[0] http://www.leonerd.org.uk/hacks/fixterms/

The first couple of parapgraphs tell you that it is based on that and list all the issues with fixterms that it fixes.

I see, thanks! I missed that.

This should be an RFC.

How hard would it be to turn this into one?

There is already a de facto standard for extended codes: Xterm's modifyOtherKeys.

Applications:

https://vimhelp.org/map.txt.html#modifyOtherKeys

https://github.com/emacs-mirror/emacs/blob/d68f2b8681f8eeb6b...

https://github.com/akinomyoga/ble.sh

https://github.com/mawww/kakoune/commit/47c0d2038807757207cd...

https://github.com/tmux/tmux/wiki/Modifier-Keys#extended-key...

https://github.com/magiblot/tvision/commit/a50569f549465c9ed...

Terminals:

xterm (obviously)

https://github.com/gnachman/iTerm2/commit/ac1aa575011028ed86...

https://github.com/r-c-f/foot

https://github.com/kovidgoyal/kitty/commit/d360d077d1f7723b1...

https://github.com/tomszilagyi/zutty

https://github.com/mintty/mintty/blob/master/wiki/CtrlSeqs.m...

Reply on news.ycombinator.com