Sending single key to the shell when with a modified

[liaowang11]

This bug is quite hard to describe and I'm not sure whether it's ghostty or fish shell, but at least it's not happening in kitty.
I'm currently using fish shell with hybrid keybindings, which means that left_bracket in normal mode maps to history-token-search-backward. And in insert mode, I like to press Ctrl+left_brakcet to exit insert mode.

When I press Ctrl+left_brakcet, the previous history item flashes and get cleared immediately. Ghostty seems to be sending left_bracket alone to fish shell.

The video uploaded may do better than my description.

Screen.Recording.2024-12-27.at.16.45.05.mov

[mitchellh]

Convert to discussion for issue triage.

[liaowang11]

Using fish_key_reader and input Ctrl-left_bracket in kitty gets:

But in ghostty:

[00-kat]

Likely the fault of fixterms, see #3151 (reply in thread). TL;DR: Ghostty implements fixterms, which many other terminal emulators don't; you can either bind \e\[91\;5u to do what you want in Fish, or rebind ctrl+left_bracket to send the same sequence as Kitty.

Note: Fish since version 4 (which is still in beta right now) supports Fixterms, so it'll just work™ there.

[liaowang11]

Using fish beta, the problem is still there, while fish_key_reader returns the same.

❯ fish --version
fish, version 4.0b1

Ghostty:

Iterm2:

[00-kat]

Ohh apologies I misinterpreted.

So for some context: ctrl+[ and escape encode to the same thing, and Fish binds escape to exit insert mode but doesn't bind ctrl+[ to escape insert mode. But since they encode to the same thing, Fish reads ctrl+[ as escape. Ghostty implements fixterms, which changes the two encodings to something different. Fish since version 4 supports fixterms so it interprets that weird \e\[91\;5u thing (A.K.A. a CSI-u encoded key sequence) as ctrl+[, which is correct, but you actually want it to be interpreted as escape, and hence escape insert mode.

So the following solutions are possible:

• If you want to use Fish before version 4, add the following to fish_user_key_bindings or your config.fish¹:

  bind -M insert \e\[91\;5u 'set fish_bind_mode default; commandline -f backward-char repaint-mode'

• If you want to use Fish version 4 or later, add the following to fish_user_key_bindings or your config.fish¹:

  bind -M insert 'ctrl-[' 'set fish_bind_mode default; commandline -f backward-char repaint-mode'

• If you want ctrl+[ to work as escape in all applications (e.g. Fish, Vim, etc.), add the following to your Ghostty configuration:

  keybind = ctrl+left_bracket=text:\x1b

Footnotes

1. Note: I use set fish_bind_mode default; commandline -f backward-char repaint-mode here, while the default binding for escape is actually if commandline -P; commandline -f cancel; else; set fish_bind_mode default; commandline -f backward-char repaint-mode; end'; I'll let you decide which you want (the default binding for escape will escape the pager menu from completions and the like if it's open, and will otherwise switch to normal mode, while what I have will always switch to normal mode). ↩ ↩²

[liaowang11]

Sorry, but I still don't understand. On my machine without the tweaks you provided(maybe related to me using hybrid mode?), Ctrl-[ works in insert mode to escape to normal mode, the issue is the flash of content during the switch from insert mode back to normal mode.

Using the first two solutions, it's not solved, but the last one do. Maybe it's related to the difference between the actual escape commands and the commands in the tweak.

Anyway, thanks for your thorough explanation.

[liaowang11]

My mistake, I'm actually running fish 3.7 for system shell. This works like a charm. Thanks again.

[00-kat]

Based on your second message, I'm guessing that it did work?

flash of content during the switch from insert mode back to normal mode

Not sure why this is happening. Fish does this when you try to complete something like echo somethingthatdefinitelycantbecompleted, and there are no matches. There are probably a few other cases where it does it too, and I'm guessing that interpreting \e[91;5u as keypresses is causing it to trigger some action which makes it flash, before eventually switching to normal mode after a timeout or something (likely because of the \e, which is an escape).

[liaowang11]

Yes it works.