This adds some new special case handling for key sequences when an
unbound keyboard input is received. If the current keybinding set scope
(i.e. active tables) has a `catch_all` binding that would `ignore`
input, then the entire key sequence is dropped.
Normally, when an unbound key sequence is received, Ghostty encodes it
and sends it to the running program.
This special behavior is useful for things like Vim mode which have `g>g`
to scroll to top, and a `catch_all=ignore` to drop all other input. If
the user presses `g>h` (unbound), you don't want `gh` to show up in your
terminal input, because the `catch_all=ignore` indicates that the user
wants that mode to drop all unbound input.
Fixes#2127
This adds a UI similar to macOS to show the current state of key
sequences and/or key tables.
https://github.com/user-attachments/assets/4399d2af-a88c-4b70-922b-7727dc4d2053
**AI disclosure:** AI was used for various things, but I did write most
of the code myself, especially around the memory management of
properties since agents can't get that quite right yet. 😄
Add a unit test to prevent regressions in our failure state.
For example, we always want to set GHOSTTY_SHELL_FEATURES, even if
automatic shell integration fails, because it's also used for manual
shell integration (e.g. #5048).
- Display keybindings grouped by their source table, with table name as prefix
- Sort default bindings before table bindings, maintaining visual hierarchy
- Support keybindings defined in key tables alongside default bindings
- Enable users to discover all available keybindings across the entire config
Fixes#10020
This improves parsing key tables so that the following edge cases are
now handled correctly, which were regressions from prior tip behavior:
- `/=action`
- `ctrl+/=action`
- `table//=action` (valid to bind `/` in a table)
- `table/a>//=action` (valid to bind a table with a sequence)
Fixes#10020
This improves parsing key tables so that the following edge cases
are now handled correctly, which were regressions from prior tip
behavior:
- `/=action`
- `ctrl+/=action`
- `table//=action` (valid to bind `/` in a table)
- `table/a>//=action` (valid to bind a table with a sequence)
Fixes#9961
This implements chained keybinds as described in #9961.
```
keybind = ctrl+shift+f=toggle_fullscreen
keybind = chain=toggle_window_decorations
```
These work with tables and sequences. For tables, the chain is unique
per table, so the following works:
```
keybind = foo/ctrl+shift+f=toggle_fullscreen
keybind = foo/chain=toggle_window_decorations
```
For sequences, it applies to the most recent sequence:
```
keybind = ctrl+b>f=toggle_fullscreen
keybind = chain=toggle_window_decorations
```
## TODO
Some limitations to resolve in future PRs (make an issue) or commits:
- [x] GTK: Global shortcuts cannot be chained: #10019
- [x] Inspector doesn't show chained keybinds
- [x] `+list-keybinds` doesn't show chains
**AI disclosure:** AI helped write tests, but everything else was
organic. AI did surprisingly bad at trying to implement this feature, so
I threw all of its work away! 😄
Our automatic shell integrations require certain resource paths to
exist. If they're missing, the launched shell could end up in an
inconsistent and unexpected state.
For example, we temporarily set ZDOTDIR to our zsh shell integration
directory and then restore it from our .zshenv file, but if that script
isn't available, the user's shell environment will be broken.
The actual runtime logic change was simple: each shell integration
routine attempts to open its expected resource path and skips automatic
shell integration upon failure. The more complex change was reworking
our unit tests to run in a temporary resources directory structure.
See: #9941
Our automatic shell integrations require certain resource paths to
exist. If they're missing, the launched shell could end up in an
inconsistent and unexpected state.
For example, we temporarily set ZDOTDIR to our zsh shell integration
directory and then restore it from our .zshenv file, but if that script
isn't available, the user's shell environment will be broken.
The actual runtime logic change was simple: each shell integration
routine attempts to open its expected resource path and skips automatic
shell integration upon failure. The more complex change was reworking
our unit tests to run in a temporary resources directory structure.
Two unrelated changes to polish key tables:
1. Key tables should be reset (deactivated) when teh config is reloaded.
This matches the behavior of key sequences as well, which are reset
on config reload.
2. A maximum number of active key tables is now enforced (8).
This prevents a misbehaving config from consuming too much memory
by activating too many key tables. This is an arbitrary limit we
can adjust later if needed.
Part of #9963
This adds a new special key `catch_all` that can be used in keybinding
definitions to match any key that is not explicitly bound. For example:
`keybind = catch_all=new_window` (chaos!).
`catch_all` can be used in combination with modifiers, so if you want to
catch any non-bound key with Ctrl held down, you can do:
`keybind = ctrl+catch_all=new_window`.
`catch_all` can also be used with trigger sequences, so you can do:
`keybind = ctrl+a>catch_all=new_window` to catch any key pressed after
`ctrl+a` that is not explicitly bound and make a new window!
And if you want to remove the catch all binding, it is like any other:
`keybind = catch_all=unbind`.
This prevents a crash in our renderer when it is larger.
I will pair this with apprt changes so that our mac app won't ever allow
a default window larger than the screen but we should be resilient at
the renderer level as well.
Fixes#9957
Our `Page.getRowAndCell` uses a _page-relative_ x/y coordinate system
and we were passing in viewport x/y. This has the possibility to leading
to all sorts of bugs, including the crash found in #9957 but also simply
reading the wrong cell even in single-page scenarios.
