This reverts commit 14134d61fb4b1bbf4ce80bb9b3ed849908bf9344, reversing
changes made to 6a876ef8ec3e2aeb3d15df0dfb0e07677e49ff03.
This causes translation failures, this should be reintroduced when the
CI check passes.
This is a large refactor of the keyboard input handling code in
libghostty and macOS. Previously, libghostty did a lot of things that
felt out of scope or was repeated work due to lacking context. For
example, libghostty would do full key translation from key event to
character (including unshifted translation) as well as managing dead key
states and setting the proper preedit text.
This is all information the apprt can and should have on its own.
NSEvent on macOS already provides us with all of this information,
there's no need to redo the work. The reason we did in the first place
is mostly historical: libghostty powered our initial macOS port years
ago when we didn't have an AppKit runtime yet.
This cruft has already practically been the source of numerous issues,
e.g.
https://github.com/ghostty-org/ghostty/issues/5558, but many other hacks
along the way, too.
This commit pushes all preedit (e.g. dead key) handling and key
translation
including unshifted keys up into the caller of libghostty.
Besides code cleanup, a practical benefit of this is that key event
handling on macOS is now about 10x faster on average. That's because
we're avoiding repeated key translations as well as other unnecessary
work. This should have a meaningful impact on input latency but I didn't
measure the full end-to-end latency.
A scarier part of this commit is that key handling is not well tested
since its a GUI component. I suspect we'll have some fallout for certain
keyboard layouts or input methods, but I did my best to run through
everything I could think of.
This also fixes one bug where preedit state didn't properly clear when
changing keyboard layouts. This now does and matches the behavior
of native apps like TextEdit and Terminal.app
This is a large refactor of the keyboard input handling code in
libghostty and macOS. Previously, libghostty did a lot of things that
felt out of scope or was repeated work due to lacking context. For
example, libghostty would do full key translation from key event to
character (including unshifted translation) as well as managing dead key
states and setting the proper preedit text.
This is all information the apprt can and should have on its own.
NSEvent on macOS already provides us with all of this information,
there's no need to redo the work. The reason we did in the first place
is mostly historical: libghostty powered our initial macOS port years
ago when we didn't have an AppKit runtime yet.
This cruft has already practically been the source of numerous issues, e.g.
#5558, but many other hacks along the way, too.
This commit pushes all preedit (e.g. dead key) handling and key translation
including unshifted keys up into the caller of libghostty.
Besides code cleanup, a practical benefit of this is that key event
handling on macOS is now about 10x faster on average. That's because
we're avoiding repeated key translations as well as other unnecessary
work. This should have a meaningful impact on input latency but I didn't
measure the full end-to-end latency.
A scarier part of this commit is that key handling is not well tested
since its a GUI component. I suspect we'll have some fallout for certain
keyboard layouts or input methods, but I did my best to run through
everything I could think of.
This PR implements a more lightweight alternative to #5326 that contains
features that I personally think Just Make Sense for the bell.
No configs, no GStreamer stuff, just sane defaults to get us started.
Fixes#7032
This introduces a syntax for `command` and `initial-command` that allows
the user to specify whether it should be run via `/bin/sh -c` or not.
The syntax is a prefix `direct:` or `shell:` prior to the command, with
no prefix implying a default behavior as documented.
Previously, we unconditionally ran commands via `/bin/sh -c`, primarily
to avoid having to do any shell expansion ourselves. We also leaned on
it as a crutch for PATH-expansion but this is an easy problem compared
to shell expansion.
For the principle of least surprise, this worked well for configurations
specified via the config file, and is still the default. However, these
configurations are also set via the `-e` special flag to the CLI, and it
is very much not the principle of least surprise to have the command run
via `/bin/sh -c` in that scenario since a shell has already expanded all
the arguments and given them to us in a nice separated format. But we
had no way to toggle this behavior.
This commit introduces the ability to do this, and changes the defaults
so that `-e` doesn't shell expand. Further, we also do PATH lookups
ourselves for the non-shell expanded case because thats easy (using
execvpe style extensions but implemented as part of the Zig stdlib). We
don't do path expansion (e.g. `~/`) because thats a shell expansion.
So to be clear, there are no two polar opposite behavioes here with
clear semantics:
1. Direct commands are passed to `execvpe` directly, space separated.
This will not handle quoted strings, environment variables, path
expansion (e.g. `~/`), command expansion (e.g. `$()`), etc.
2. Shell commands are passed to `/bin/sh -c` and will be shell expanded
as per the shell's rules. This will handle everything that `sh`
supports.
In doing this work, I also stumbled upon a variety of smaller
improvements that could be made:
- A number of allocations have been removed from the startup path that
only existed to add a null terminator to various strings. We now have
null terminators from the beginning since we are almost always on a
system that's going to need it anyways.
- For bash shell integration, we no longer wrap the new bash command in
a shell since we've formed a full parsed command line.
- The process of creating the command to execute by termio is now unit
tested, so we can test the various complex cases particularly on macOS
of wrapping commands in the login command.
- `xdg-terminal-exec` on Linux uses the `direct:` method by default
since it is also assumed to be executed via a shell environment.
This introduces a syntax for `command` and `initial-command` that allows
the user to specify whether it should be run via `/bin/sh -c` or not.
The syntax is a prefix `direct:` or `shell:` prior to the command,
with no prefix implying a default behavior as documented.
Previously, we unconditionally ran commands via `/bin/sh -c`, primarily
to avoid having to do any shell expansion ourselves. We also leaned on
it as a crutch for PATH-expansion but this is an easy problem compared
to shell expansion.
For the principle of least surprise, this worked well for configurations
specified via the config file, and is still the default. However, these
configurations are also set via the `-e` special flag to the CLI, and it
is very much not the principle of least surprise to have the command run via
`/bin/sh -c` in that scenario since a shell has already expanded all the
arguments and given them to us in a nice separated format. But we had no
way to toggle this behavior.
This commit introduces the ability to do this, and changes the defaults
so that `-e` doesn't shell expand. Further, we also do PATH lookups
ourselves for the non-shell expanded case because thats easy (using
execvpe style extensions but implemented as part of the Zig stdlib). We don't
do path expansion (e.g. `~/`) because thats a shell expansion.
So to be clear, there are no two polar opposite behavioes here with
clear semantics:
1. Direct commands are passed to `execvpe` directly, space separated.
This will not handle quoted strings, environment variables, path
expansion (e.g. `~/`), command expansion (e.g. `$()`), etc.
2. Shell commands are passed to `/bin/sh -c` and will be shell expanded
as per the shell's rules. This will handle everything that `sh`
supports.
In doing this work, I also stumbled upon a variety of smaller
improvements that could be made:
- A number of allocations have been removed from the startup path that
only existed to add a null terminator to various strings. We now
have null terminators from the beginning since we are almost always
on a system that's going to need it anyways.
- For bash shell integration, we no longer wrap the new bash command
in a shell since we've formed a full parsed command line.
- The process of creating the command to execute by termio is now unit
tested, so we can test the various complex cases particularly on
macOS of wrapping commands in the login command.
- `xdg-terminal-exec` on Linux uses the `direct:` method by default
since it is also assumed to be executed via a shell environment.
Fixes#6962
I believe this is an upstream bug
(https://github.com/ziglang/zig/issues/23454), where Zig is allowing
extern unions to be tagged when created via type reification. This
results in a CValue that has an extra trailing byte (the tag).
This wasn't causing any noticeable issues for Ghostty for some reason
but others using our pattern were seeing issues. And I did confirm that
our CValue was indeed tagged and was the wrong byte size. I assume Swift
was just ignoring it because it was extra data. I don't know, but we
should fix this in general for libghostty.
See #6957
We were not considering GTK's internal scale factor that converts between
"surface coordinates" and actual device coordinates, and that worked fine
until the scale factor reached 2x (200%).
Since the code is now dependent on the scale factor (which could change
at any given moment), we also listen to scale factor changes and then
unconditionally call `winproto.syncAppearance`. Even though it's somewhat
overkill, I don't expect people to change their scale factor dramatically
all the time anyway...
This was a hack, and is no longer required since #6877. Users can
explicitly override the GTK GSK renderer by setting the standard GTK env
var `GSK_RENDERER`.
Fixes#6872
This commit explicitly acquires the GL context in the `unrealize`
signal handler of the GTK Surface prior to cleaning up GPU resources.
A GLArea only guarantees that the associated GdkContext is current for
the `render` signal (see the docs[1]). This is why our OpenGL renderer
"defers" various operations such as resize, font grid changing, etc.
(see the `deferred_`-prefix fields in `renderer/OpenGL.zig`).
However, we missed a spot.
The `gtk-widget::unrealize` signal is emitted when the widget is
destroyed, and it is the last chance we have to clean up our GPU
resources. But it is not guaranteed that the GL context is current at
that point, and we weren't making it current. On the NGL GTK renderer,
this was freeing GPU resources we didn't own.
As best I can understand, the old GL renderer only ever used a handful
of GL resources that were early in the ID space, so by coincidence we
were probably freeing nothing and everything was fine. But with the new
NGL renderer uses a LOT more GL resources (make a few splits and the ID
space is already in the thousands, from GTK!), so we were freeing real
resources that we didn't own, which caused rendering issues. :)
I suspect the above also resulted in VRAM memory leaks (which would be
RAM memory leaks for unified memory GPUs). This potentially relates to
#5491.
The fix is to explicitly make the GL context current in the `unrealize`
handler.
[1]: https://docs.gtk.org/gtk4/method.GLArea.make_current.html
Fixes#6821
UTF8 translation using KeymapDarwin requires a buffer and the buffer was
stack allocated in the coreKeyEvent call and returned from the function.
We need the buffer to live longer than this.
Long term, we're removing KeymapDarwin (there is a whole TODO comment in
there about how to do it), but this fixes a real problem today.
1. Refactored Nix devshell/package to make it easier to keep
LD_LIBRARY_PATH & buildInputs in sync (plus make it easier to re-use in
other Nix environment).
2. Added a CI job to ensure that Blueprints are formatted correctly and
that they will compile using `blueprint-compiler` 0.16.0.
3. Reformatted all Blueprints with `blueprint-compiler format`.
1. Refactored Nix devshell/package to make it easier to keep
LD_LIBRARY_PATH & buildInputs in sync (plus make it easier to re-use
in other Nix environment).
2. Added a CI job to ensure that Blueprints are formatted correctly and
that they will compile using `blueprint-compiler` 0.16.0.
3. Reformatted all Blueprints with `blueprint-compiler format`.
As of Adwaita 1.5.0, the GTK Box is not being properly unref'd when the
parent window is closed. Update the conditional to account for this.
Also add a couple of missing unref()s in errdefers.
This fixes an issue where Ghostty would not properly quit after closing
the last surface.
https://github.com/ghostty-org/ghostty/discussions/3807 is related
(though I'm not sure it's the exact same problem).
As of Adwaita 1.5.0, the GTK Box is not being properly unref'd when the
parent window is closed. Update the conditional to account for this.
Also add a couple of missing unref()s in errdefers.
Fixes#6772
When typing Korean with the fcitx5-hangful input method, moving between
graphemes does not trigger a preedit end/start cycle and instead just
clears the preexisting preedit and reuses the started state.
Every other input method we've tested up until now doesn't do this. We
need to mark composing set to "false" in "commit" because some input
methods on the contrary fail to ever call END.
What is the point of start/end events if they are just ignored depending
on the whim of the input method? Nothing. That's what. Its all a mess
that GTK should be protecting us from but it doesn't and now its the app
developer's problem. I'm frustrated because I feel like the point of an
app framework is to mask this kind of complexity from the app developer
and I'm playing whack-a-mole with input methods.
Well, here's another whack. Let's see if it works.
1. Remove usage of C header imports for gtk x11/wayland.
2. Move X11 C header imports to winproto_x11.zig
3. Clean up long line by breaking it up into multiple steps.
Fixes#6633
For macOS, we set LANGUAGE to the priority list of preferred languages
for the app bundle, using the GNU gettext priority list format (colon
separated list of language codes).
This previously was inherited by the termio env. At first, this was by
design, but this has inherent flaws. Namely, the priority list format is
a GNU gettext specific format, and programs that use alternate gettext
implementations (like musl or Python) do not understand it and actually
do the wrong thing (not their fault!).
This change removes the inheritance of LANGUAGE in the termio env. To
make it extra safe, we only do set and unset LANGUAGE when we know we
launch from an app bundle. That was always the desired behavior but this
makes it more explicit.
Fixes#2384 on GTK
I'm not exactly sure how to deal with centered quick terminals so I
opted to make them similar to either top/bottom or left/right quick
terminals based on the monitor's orientation (portrait/landscape). This
may not be the right approach, so I'd like to hear more thoughts about
this.
Fixes#2384 on GTK
I'm not exactly sure how to deal with centered quick terminals so I opted
to make them similar to either top/bottom or left/right quick terminals
based on the monitor's orientation (portrait/landscape). This may not be
the right approach, so I'd like to hear more thoughts about this.
AdwAlertDialog is the recommended way to do alert/message dialogs
starting from libadwaita 1.5, and is much easier to manage than
GtkMessageDialog. (The latter is also deprecated since GTK 4.10, but
this PR does not migrate it to use GtkAlertDialog, mostly because of its
obtuse interface and that we'll remove the GtkMessageDialog code anyway
in 1.2 when we remove non-Adwaita builds.)
We also had two bugs where tabs with only one split would display the
"close surface" confirmation dialog, and windows would do the same when
closed via the "Close Window" menu item or by the `close_window` keybind
action. (The "close window" dialog only appears when the user clicks on
the close button on the titlebar.) Initially I was very confused by
this, but it turns out that we don't have any apprt action related to
closing a window, and it was simply closing surfaces...
As of now `gtk4-layer-shell` is unavailable on recent, stable releases
of many distros (Debian 12, Ubuntu 24.04, openSUSE Leap & Tumbleweed, etc.)
and outdated on many others (Nixpkgs 24.11/unstable, Fedora 41, etc.)
This is inconvenient for our users and severely limits where the quick
terminal can be used. As a result we then build gtk4-layer-shell ourselves
by default unless `--system` or `-fsys=gtk4-layer-shell` are specified.
This also allows me to add an idiomatic Zig API on top of the library
and avoiding adding even more raw C code in the GTK apprt.
Since we now build gtk4-layer-shell it should be theoretically available
on all Linux systems we target. As such, the `-Dgtk-layer-shell` build
option has been removed. This is somewhat of an experimental change as
I don't know if gtk4-layer-shell works perfectly across all distros, and
we can always add the option back if need be.
