Fixes#473
A new config value for `gtk-single-instance = desktop` is now valid.
This value uses GTK single instance mode only if it detects Ghostty was
launched from a desktop environment (i.e. clicking on the icon) through
the `.desktop` Freedesktop file. Otherwise, it disables single instance.
This enables Ghostty to work well with the CLI and accept all the
typical CLI flags. It will make Ghostty launch slower from the CLI but I
think this is an acceptable tradeoff.
Currently, the selection on a surface is cleared on any keypress that
generates a control sequence. For applications that enable kitty, this
can mean that every keypress clears the current selection thereby making
it impossible to copy text, since it will be cleared as soon as Control
is pressed. This can be very confusing to terminal users: some
applications I can shift+highlight and copy+paste, some I can't - and
unless I know the implementation of the application I won't know
why...very frustrating.
Only clear the selection when there is printable text, otherwise retain
the selection. From my testing, this makes text selection feel the same
between applications using kitty keyboard and applications that don't.
The macOS desktop menu-bar grows in total height by adding extra padding
to deal with the physical notch found on various MacBook displays.
When config `macos-non-native-fullscreen = visible-menu` we apply
`safeAreaInsets.top` to reduce frame height.
User input withe Alt modifier is typically ESC prefixed. Escape
prefixing a non-ascii character can cause bugs in some applications. For
example in bash, emitting an Alt+ф allows the user to backspace one
character into the prompt. This can be repeated multiple times.
When a character is outside the ASCII range (exclusive of 0x7F, this is
handled as a control sequence), print the character as is, with no
prefix.
Fix reporting of alternate keys when using the kitty protocol. Alternate
keyboard layouts were failing to report the "base layout" key. This
implementation now matches kitty's output 1:1, and has some added unit
tests for cyrillic characters.
This also fixes a bug where a caps_lock modified key would report the
shifted key as well. The protocol explicitly requires that shifted keys
are only reported if the shift modifier is true.
When converting keyval to the unshifted version, gdk_keyval_to_lower
returns a keyval type, not a unicode value. Convert this to unicode
before assigning to keyval_unicode_unshifted.
This fixes a bug where the incorrect keycode is sent in alternate
layouts with kitty keyboard on linux.
