Related to https://github.com/mitchellh/ghostty/issues/1082
This fixes two separate issues to follow along with the new spec changes
Kovid pushed to Kitty:
1. When two modifiers are pressed and one is released, this shows up
as a proper release event with the correct side. Previously, the
correct side was shown but as a press event.
2. When two modifiers are pressed and one is released, the Kitty event
should not have that specific modifier set. For example, pressing
left ctrl, then right ctrl, then releasing right ctrl should encode
as "right ctrl released" but with NO modifiers still present.
Some control key combinations must be handled specially by Ghostty to
prevent undesirable behavior at the OS level. For now, this includes
only Ctrl-/, which makes a "beep" sound when processed by AppKit. It is
unclear why this beep occurs and no answer was found after extensive
searching.
This solution is inspired by iTerm2, which also handles certain control
key combinations (including C-/) manually before passing them on to
Cocoa/AppKit.
This enables shifted alt-prefixed keys, such as `shift+alt+.` on
US standard becoming `M->`. To do this, we needed to fix a few bugs:
(1) translation mods should strip alt even if other mods are set
(2) AppKit translation event needs to construct new characters with
the translation mods.
(3) Alt-prefix handling in KeyEncoder needs to allow ASCII utf8
translations even for macOS.
Fixes#872
In #867 we fixed macos-option-as-alt, but unfortunately AppKit ALSO does
some translation so some behaviors were not working correctly.
Specifically, when you had macos-option-as-alt set, option+e would
properly send `esc+e` to the pty but it would ALSO set the dead key
state for "`" since AppKit was still translating the option key.
This commit introduces a function to strip alt when necessary from the
translation modifiers used at the AppKit layer, preventing this
behavior.
This is a small cleanup change. This provides the Ghostty.AppState
object to any view that needs it within the TerminalSplit view hierarchy
without needing to explicitly define EnvironmentKeys.
This commit does two things: adds a weak reference to the parent
container of each SplitNode.Container and SplitNode.Leaf and moves the
"direction" of a split out of the SplitNode enum and into the
SplitNode.Container struct as a field.
Both changes are required for supporting split resizing. A reference to
the parent in each leaf and split container is needed in order to
traverse upwards through the split tree. If the focused split is not
part of a container that is split along the direction that was requested
to be resized, then we instead try and resize the parent. If the parent
is split along the requested direction, then it is resized
appropriately; otherwise, repeat until the root of the tree is reached.
The direction is needed inside the SplitNode.Container object itself so
that the container knows whether or not it is able to resize itself in
the direction requested by the user. Once the split direction was moved
inside of SplitNode.Container, it became redundant to also have it as
part of the SplitNode enum, so this simplifies things.
The resizeIncrements property is only modified when the cell size of the
focused window changes. If two splits have the same cell size then the
property is not modified when focusing between the two splits.
Fixes#519
The core issue here was that `mouseEntered` was called AFTER
`cursorUpdate` (by Cocoa) so we were messing up our NSCursor state. To
fix this more robustly, all cursor state should ONLY be handled by
cursorUpdate and mouseEntered/Exit goes through that system now.
Without this, non-first-responder views would not receive mouse
entered/exit events. This would break some of our mouse hiding state.
See comments for more info.
