Ghostel.el: Terminal emulator powered by libghostty
dakra.github.io
dakra.github.io
baokaola and I actually wanted to do a "Show HN" next week, but looks like someone was faster submitting the link.
Have a look at the GitHub repo which is a bit nicer for a quick overview: https://github.com/dakra/ghostel
To add some context, Ghostel is a terminal emulator for Emacs powered by libghostty-vt.
There's a feature comparison vs vterm and eat: https://dakra.github.io/ghostel/#ghostel-vs-vterm
And here is a gist with images to compare performance and correctness: https://gist.github.com/dakra/4a0b76ebcf5d52338e134864378465...
But for me personally, it has not only replaced vterm/eat but also any other external terminal like kitty/Ghostty.
Having your terminal text just like a normal Emacs buffer opens up so many possibilities and extension points that are just not available on any other terminal.
Even simple stuff like searching in the scrollback, then navigating and selecting+copying a paragraph only with the keyboard. For every Emacs user that's so natural and fast in Ghostel while often cumbersome in other Terminals where I just reach to the mouse because it's easier.
Happy to answer any questions and also like to hear feedback positive or negative.
If you're an Emacs user and tried Ghostel and are still using Ghostty (or another external Terminal), is there something Ghostel is missing or is it just because you want some processes to run outside of Emacs?
baokaola and I are also very active on GitHub, so feel free to open an issue if you have any.
> is there something Ghostel is missing
eshell allows me to manipulate text as I would in any other Emacs buffer. If I have a function which wraps a word in quotes, and bind it to a key, I can be confident it will work in eshell like it does anywhere else. It's a real killer feature. If I use evil-mode, or xah-fly-keys, or simply want to use ispell to correct the spelling of a word, it all works.
Unfortunately with Ghostel none of this works. It's not integrated in the same way. There are extensions like evil-ghostel-mode, but they are limited.
Are there any plans to improve this, or is it a limitation Ghostel has to live with?
A Ghostel equivalent of eat-eshell-mode would be amazing.
There you could type on the prompt line and then call jinx or your quote wrapping function etc as it's just a normal Emacs buffer. You can't edit the scrollback buffer though, but I don't think that's possible in eshell either.
But line-mode has it's own set of problems. Since we don't send anything to the shell, you could have some problems with autocomplete or similar things that change the text depending on each typed char. Similarly we automatically disable line-mode when you enter a TUI (alt-screen) app, as line-mode doesn't make too much sense in e.g. vim. But that's configurable and you can still force line-mode, it really depends on the TUI apps.
We try to support as much as possible and work around things like fish autocomplete etc. But please try and report any issues you find.
Anyway, unfortunately that is not possible in a Ghostel buffer and most likely also will never be. I'm open to ideas though how we could improve or replicate your eshell workflow.
But also, eshell is awesome and Ghostel is not a replacement for it. It's more a replacement for term.el, maybe shell.el (with line-mode) and other terminal packages like eat and vterm.
Ghostel looks really nice though!
I didn't even think of the use-case that then you can edit output from finished commands in the scrollback. But it makes sense and is another +1 for that feature request.
For libghostty-vt, since you're targeting a terminal TUI instead of an external subsystem (for example; for ghostty, you hit libghostty-vt -> GPU rendering, which is external), you still have to buy into terminal semantics. in my experience, since I was trying to replicate mosh with libghostty-vt as the parser, what happened was that my optimized re-rendering kept getting increasingly coupled to terminal semantics (and the UDP state update model too), otherwise I'd have to send the entire terminal grid over the network like, every time.
What are the tricks for making this both performant and not like, utter cancer? You have a harder issue here too (similar to tmux) in that certain optimizations are just not available to you, or you have to translate (literally geometrically) certain instructions
For Ghostel, libghostty-vt is the source of truth and the architecture is essentially that we serve input to the PTY and the PTY serves output to libghostty-vt which builds out the state in the form of a terminal screen structure. The goal is then to keep the contents of an Emacs buffer up to date to this terminal screen without replace the entire thing every time we redraw. We make use of mainly two things in order to do as little work as possible: - Scrollback is immutable and thus never has to be modified unless it's evicted, alt screen is activated, dimensions change etc. - libghostty-vt maintains row level dirty flags that we scan to make sure we're only replacing lines that have actually changed.
So for the rendering part, we're only diffing the grid state against the buffer, not doing anything based on terminal semantics per se, parser events that draw to the screen are passed straight to the terminal handler. But of course, certain things we need to hook into such as directory and title changes, of clipboard events etc.
Might also add that we're using the direct Zig API, not the C API, which means we have access to things that aren't exposed in the C API.
Awesome project. Been using with doom for a while. How do you manage to get scrolling programs to work (eg Lazygit or Reasonix) where other emulators fail? Is it something special in your implentation or library that makes this work?
What language do you have in mind?
If you're worried about performance, Common Lisp is a compiled language. It's about the best performing dynamic language there is, and not too far behind Zig, Rust and the likes.
But for those who do choose Emacs for its supreme hackability, I got a hunch that their profile is very similar to those who choose lisp for its unmatched meta-programming ; they're also very similar to the OG lispers who actually rejected M-expression, the normal, more mathy syntax with less parens, and embraced S-Expressions, precisely because (= code CST AST) is what gives lisp its simplicity and powers in the first place. It's not that we are stuck with parens anyway, the initial vision of lisp, with far fewer parens, has been implemented in Rhombus for example; and we don't even have to change language, 1 or 2 days of writing macros is all it takes to get your favorite lisp to understand many expressions without brackets, and even remove many brackets from existing code; for example, I'm using something close to https://www.reddit.com/r/lisp/comments/18b6hf/comment/c8d8wt... to write your example as (Too $ many $ god $ damn $ brackets), which btw, doesn't use any less brackets in an Algol language: Too(many(god(damn(brackets())))) (but yeah, in many circumstances, we do use more).
This is why I have a hard time imagining an "emacs spawn" not written in lisp. Before even considering that many Emacs concepts are most easily transcribed in a lisp, who would be interested in writing and using it in the first place? (genuine question)
Thankfully for people disliking lisp, and you are many, there are also plenty of great editors which share some of Emacs strength, while being neither written in lisp nor Emacs copies https://www.reddit.com/r/emacs/comments/kzymtd/are_there_oth...
Edit: apparently, (some) emacs lovers disagree with me, saying emacs can be emacs even without lisp: https://www.emacswiki.org/emacs/Emacsen
And even provide a list! https://www.emacswiki.org/emacs/EmacsImplementations
I have never used WezTerm but from a quick look, it seems you can quickly copy text ace-jump style.
In Emacs you could use e.g. avy to do the same, but there is probably a bunch of other similar packages that do the same.
And additionally you're not restricted to just select/copy a word. You can freely move around and use e.g. expand-region or similar to quickly select more.
It's Emacs after all, the possibilities are endless ;)
Two things: - defined f1 to toggle between semi-char vs copy modes - I ask myself "Is it ok if Emacs dies or I nuke it?" If not, then I execute a command in the normal terminal
this is useful anyway but just in case you didn't know, we automatically switch to copy-mode when it makes sense.
E.g. when you activate mark, click somewhere with the mouse or when point leaves the current prompt position.
That means that with isearch or consult-line etc you're automatically in copy-mode (because why else would you search something and then jump to that position).
For other commands like avy or flash jump packages I have this in the documentation:
;; A package that runs a hook after jumping (e.g. flash): (add-hook 'flash-after-jump-hook #'ghostel-maybe-leave-input)
;; A package without one (e.g. avy) — advise its jump action: (with-eval-after-load 'avy (advice-add 'avy-action-goto :after #'ghostel-maybe-leave-input))
Personally I use copy mode a lot but I rarely have to "manually" activate the mode.
That being said, there are still some rough edges. Sometimes it fails to properly clear the terminal, leaving junk at the top of the buffer before the currrent prompt line. And on a couple of occasions it has totally frozen, with no fix other than killing the buffer and starting over.
Overall, it’s very promising and totally usable as a daily driver, but it needs a bit of polish and bug fixes before I would consider it mature.
The junk at the top of the screen sounds like it could be https://github.com/dakra/ghostel/issues/495 and it should be fixed on later versions. But maybe you're seeing another bug. The tricky part is replicating the libghostty-vt internal data into an Emacs buffer while only replacing the parts that need to be replaced. We have property based tests to exercise this a lot, but sometimes things slip through.
The latest released version as I'm writing this should have improved lifecycle handling, so maybe it also fixes some of your issues.
As you say, the project is still in the early phase so hopefully, we can iron things out over time.
I do see a similar issue, where when I switch to the ghostel buffer and it wasn’t visible before, the text is scrambled. I’ll check if I can find a way to reliably reproduce it.
When are you mostly seeing this? With agent TUIs?
Yes, that sounds like the same issue. I’ll update to the latest version and see if it’s resolved. And thanks for your work on this package, it’s been a real game-changer for me!
And I would also prefer if the link went to https://github.com/dakra/ghostel instead of the documentation which is not that helpful if you don't know what the project is.
It's not the same as expecting the reader to know what a terminal is, just in case anyone was thinking of going there next.
It's a weird flex to insist that the titles should always be extremely descriptive. It is already meaningful for the respected audience. Otherwise you can endlessly argue that a florist may not understand what "terminal" and "emulator" are, therefore it should be more descriptive for an occasional florist vising HN. For anyone who doesn't know what .el stands for, they'd click, see it's about Emacs and if not interested - close the tab and forget about it. Takes maybe a second. Not even a smallest imaginable deal - not big at all.
It also installs a repeat-mode map, so if you see 3 url or file links as output you can just `C-c C-p p p RET` and it will open the first link.
I use that feature all the time.
Can someone help who's used both help me, what's the elevator pitch for this, why is it worth checking out?
The only feature of Ghostty that I now miss in Kitty: setting the text size of each window pane independently. Oh and the slightly smoother quake terminal, even though I don’t use that much.
Ghostty makes some choices that I personally find bizarre. For one, on macOS, cmd+, opens config in textedit instead of in $EDITOR. Another is there isn’t a key action for moving panes; you need to use the mouse. At least half of the config options are platform-specific. And weirdest of all scripting is limited and can only really be done with (get this) AppleScript.
So, like I surmised (and I could be wrong, of course), Ghostty is just a nice terminal that lives alongside your other apps; Kitty is powerful enough to practically act like a second desktop environment (that is also cross-platform).
Of course, this post is about libghostty, the underlying term emulator library. And that is an amazing contribution to the ecosystem! Zmx, herdr, soon neovim, now emacs, hopefully someday Zed – so many things can now embed an amazingly performant terminal
I have opened right now about a dozen Ghostty windows and about 20 tabs in each window, i.e. more than 100 shell instances.
I have started in as many of them as I could, before becoming too bored, a "ls -lR" on a file system with many millions of files.
I could not see any problem, much less any crash. I have been using Ghostty for a few months, very intensively, all day long, and I have not seen any crash or other suspicious behavior.
If you have seen a crash, perhaps there was either some specific version of Ghosstty that had a bug, or, more likely, some weird interaction with some other software that you have, and which might be buggy, e.g. the GPU driver. (I am using an NVIDIA GPU.)
But like people coming to Emacs from Neovim may get confused why is nvim +term has only two modes, but this one has 5, what's the point? Without clearly understanding the problem, the knee-jerk reaction might be "this thing is an over-engineered BS", while in truth Ghostel isn't more complex because it's over-engineered - it's more complex because it solves more of the problem - the extra modes are opt-in tools for tasks nvim simply doesn't address. But it's not super clear how in practice use that leverage efficently.
I know you know them but I list the 5 modes quick here for readers:
- semi-char: The default, where most keys go to the terminal but some common Emacs prefixes (M-x, C-c) go to Emacs
- char-mode: ALL keys go to the terminal. This way you can run e.g. Emacs inside Emacs.
- copy-/emacs-mode: This makes the whole buffer a pure read-only Emacs buffer and all keys go to Emacs. The difference between copy and emacs-mode is that copy freezes the terminal output (comes originally from vterm), while in emacs-mode new output keeps coming in (adopted from eat).
- line-mode: It's like `M-x shell`, everything goes to Emacs but it's not read only but you can type text on the prompt. But nothing is send to the terminal until you press return.
> why is nvim +term has only two modes, but this one has 5
I haven't used the neovim terminal, but I guess you can compare vim insert-mode with semi-char and normal-mode with copy-mode. And surely they have a char-mode as well, or can you not run nvim inside nvim (without having to press some quote key all the time)?
That would leave only line-mode as the odd one out.
> But it's not super clear how in practice use that leverage efficently
As with all Emacs things, that's highly personal.
Currently, I use semi-char mostly and switch to copy-mode to select/copy stuff.
Very rarely do I use char or line-mode.
Why? Keep it a part of distribution.
That means we would have to check in the module binary for all platforms (>10MB together) if we want that it comes with the distribution.
Also looking at e.g. jinx, another popular package that uses Emacs native modules, it does it like vterm and offers to compile on first usage.
So as a Emacs package author, for a user friendly installation you can realistically only offer to download or compile on first use.
If you do not want to or you cannot install the terminfo data, there is the easy workaround to put in your shell initialization script on the remote computer something like "export TERM=xterm-256color".
Ghostty aims to be completely compatible with xterm, so everything should work fine after setting thus TERM, only the newer features of ghostty will not be available.
As it copies stuff to a remote host, automatic ssh injection is disabled by default but you can enable it by setting `ghostel-tramp-shell-integration` to true.
For manual setup you can just copy a few lines from the manual in your shell: https://dakra.github.io/ghostel/#orgfea0bed
Unfortunately, millions of lines of scrolling text are no longer unusual, especially when you frequently compile big software projects. The use of high-resolution monitors has also been normal for many years.
Instant window rendering is addictive, so now I would never return from a fast terminal emulator like ghostty to an older video terminal emulator. The last terminal emulator that I had been using before ghostty was kitty, which was also pretty fast in comparison with traditional terminal emulators, but I like ghostty more.
I explained why it's not in the package in https://news.ycombinator.com/item?id=48881722
For all Emacs package updates, it's best of course to check the source what's changed and then you can compile yourself.
PS, I also think Emacs is one of the few ecosystems where people actually check what changed. It's not like npm where you have a million of deps unreviewed pulled in.
Personally I use borg and always at least quick check what changed, and most Emacs package managers have a similar feature.
I didn't have time to fix it so downloading the binary module has been the only option.
I had the same problem with vterm when I first tried it. The C module failed to compile, with a compiler error about a line in the source. As there was no downloadable binary module I fixed that one.
This is likely the problem. Latest Zig is 0.16 and Ghostty (and therefor Ghostel) require exactly Zig 15.2.
I just updated our code to have a better error message and make it clear that the problem is the Zig version. See https://github.com/dakra/ghostel/pull/541
jFYI, Ghostty issue to support 0.16 is here: https://github.com/ghostty-org/ghostty/issues/12228
So for your homebrew zig you probably have to:
brew uninstall zig
brew install zig@0.15