Per review feedback, cover the four Windows branches added in the
parent commit:
- bare `cmd.exe` resolves via `%COMSPEC%` (with documented fallback)
- bare non-cmd shell (`pwsh.exe`) is passed through unchanged
- shell value with arguments (`wsl ~`) is split on whitespace
- direct command is passed through without modification
Co-authored-by: Claude <noreply@anthropic.com>
On Windows the shell value was always executed as `cmd.exe /C <shell>`.
For even a simple `command = wsl ~` this spawned two processes (the
cmd wrapper and the user's actual shell) and had visible side effects:
an extra cmd.exe in the process tree, and cmd AutoRun state (DOSKEY
aliases, `cd` in init.cmd, etc.) running in the wrapper rather than
the user's shell, since AutoRun is per-process.
Run the shell value directly. If it contains whitespace, split on
whitespace into argv. Bare `cmd.exe` is resolved via %COMSPEC% which
is the documented path to the current command processor; other bare
values are left to PATH resolution in Command.startWindows.
The simple whitespace split does not honor Windows CLI quoting rules.
Users who need quoted arguments should use the direct command form.
Also skip the termios focus timer on Windows since Windows has no
termios; the focusGained callback was starting a timer whose callback
would then do nothing.
Co-authored-by: Claude <noreply@anthropic.com>
Widens the existing `-fno-sanitize=undefined` gate from `abi == .msvc`
to `os.tag == .windows`. The same undefined `__ubsan_handle_*` link
errors from simdutf/highway also reproduce on Windows GNU ABI, and the
fix is identical.
Part of the Win32 apprt upstreaming series (see discussion #2563 /
mattn/ghostty#1).
`combine_archives` spawns `zig ar -M` to combine static archives via
an MRI script. It hard-coded the command name `"zig"` and relied on
the binary being on `PATH`, which fails on Windows when the build is
driven by an absolute zig.exe path (common in CI and in Scoop/winget
installs where PATH isn't populated at build time). The failure
surfaces as `error: FileNotFound` from `Child.spawn`.
Pass `b.graph.zig_exe` as the first argument so the tool always uses
the exact zig binary that is driving the build, matching how other
build tools in this repo spawn zig subcommands.
The existing `-fno-sanitize=undefined` flag was gated on `abi == .msvc`
to avoid undefined `__ubsan_handle_*` references from simdutf/highway.
The same linker error reproduces on Windows GNU ABI for the same
reason: the Zig-bundled libraries don't provide a matching UBSan
runtime for these C/C++ objects in our build configurations.
Widen the condition to `os.tag == .windows` so both MSVC and GNU
Windows targets skip ubsan for these C++ deps.
Per review feedback (#12373), fold the nested `if/else if/else` into a
single Windows-gated struct whose handler picks up the abi difference
via a comptime check. This removes the duplicated `const BOOL = ...`
block that the two per-abi structs shared.
Part of preparation for upstreaming a Win32 application runtime
(see discussion #2563). This is one of three small build-related
fixes that unblock the Windows GNU-ABI library build.
When targeting Windows with GNU ABI, the existing `DllMain` declaration
falls through to `void` (a type), which Zig stdlib's `start.zig` then
attempts to call as a function via `root.DllMain(...)` - producing the
compile error "type 'type' not a function".
Restructure the conditional so that:
- non-Windows builds keep `DllMain = void`
- Windows + MSVC keeps the existing CRT-init handler (unchanged)
- Windows + non-MSVC gets a no-op `BOOL` handler
This unblocks `zig build -Dtarget=native-native-gnu -Dapp-runtime=none`
on Windows.
The GhosttyKit xcframework previously shipped the entire include/
directory, which pulled in the libghostty-vt headers under
include/ghostty/. Because those headers are not referenced from the
ghostty.h umbrella, Clang's module system emitted "umbrella header for
module 'GhosttyKit' does not include header 'ghostty/vt/*.h'" warnings
in Xcode builds.
Stage only ghostty.h and module.modulemap via addWriteFiles so the
xcframework Headers directory contains exactly the GhosttyKit API,
mirroring the pattern used in GhosttyLibVt.xcframework.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Expose the foreground process PID and TTY device path as read-only properties on the AppleScript terminal class and App Intents TerminalEntity. This enables reliable process-to-terminal mapping for automation tools when multiple terminals share the same CWD.
Closes#11592Closes#10756
Session: 019d341c-a165-7843-a2f7-2f426114cf17
Stop trying to use POSIX shared memory functions such as `shm_open` on
Android as it's unsupported and the platform libc does not have those
symbols.
This avoids an error such as the below when trying to use
`libghostty-vt` on Android:
> dlopen failed: cannot locate symbol "shm_open" referenced by [..]
Stop trying to use POSIX shared memory functions such as
`shm_open` on Android as it's unsupported and the platform libc does not
have those symbols.
This avoids an error such as the below when trying to use
`libghostty-vt` on Android:
> dlopen failed: cannot locate symbol "shm_open" referenced by [..]
Maybe related to #12298?
When Screen resize forwards the active cursor into PageList reflow, a
history-pinned viewport can be remapped into the active area before the
preserved-cursor grow step finishes. The old code kept treating that
viewport as a history pin during the intermediate grow calls, which left
too few rows beneath the pin and tripped the viewport integrity checks.
Fix this by normalizing the viewport back to active as soon as reflow
moves the pinned row into the active area. Add a Screen-level regression
test that exercises the full resize path with bounded scrollback and
wrapped rows, and document the setup so the unwrap and viewport
transition are clear.
This updates simdutf to my fork which has a SIMDUTF_NO_LIBCXX option
that removes all libc++ and libc++ ABI dependencies.
From there, the hand-written simd code we have has been updated to also
no longer use any libc++ features. Part of this required removing utfcpp
since it depended on libc++ (`<iterator>`).
libghostty-vt now only depends on libc.
This updates our synthetic generator for UTF-8 to expose:
- Flags to change 1/2/3/4-byte UTF-8 character distribution
- Flags to have only printable characters so we can benchmark
pure UTF-8 vs our control sequence finder.
- Flags to have invalid characters so we can benchmark our error
handling.
This also adds an AGENTS.md to src/benchmark so agents can do the right
thing more easily.
* Don't alter Kitty keyboard protocol responses. Kitty does not support
DECBKM so KKP doesn't take DECBKM into consideration.
* Make better use of the function key lookup to control what sequence is
returned when backspace is pressed using the legacy encoding.
This mode allows programs to modify the code that the `backspace`
key (backarrow key in DEC parlance) sends. If this mode is
`off`/`false`/`reset` (the default, the same as before this PR), we
send the byte `0x7f`. If this mode is `on`/`true`/`set` we send the
byte `0x08`.
## Summary
Mirror the `libghostty-vt-static` pkg-config pattern from #12210 for the
internal library.
- Add `ghostty-internal.pc` (shared, `-lghostty`) and
`ghostty-internal-static.pc` (static, direct archive reference) so
consumers can discover either variant via pkg-config
- Named `ghostty-internal` to distinguish from the public
`libghostty-vt` API
- Static module points at the platform-correct archive name
(`ghostty-static.lib` on Windows, `libghostty.a` elsewhere)
- pkg-config files are generated during shared builds and installed via
`GhosttyLib.install()`
## Test plan
- [x] `zig build` succeeds (default target)
- [x] `ghostty-internal.pc` and `ghostty-internal-static.pc` appear in
`zig-out/share/pkgconfig/`
- [x] Static `.pc` points at `ghostty-static.lib` (Windows) /
`libghostty.a` (Unix)
- [x] Shared `.pc` uses standard `-L -l` flags
- [x] Existing `libghostty-vt` pkg-config files are unaffected
The vendored Highway package was being built with libc++ even though
Ghostty only uses its runtime target selection and dispatch support.
That pulled in extra C++ runtime baggage from upstream support files
such as abort, timer, print, and benchmark helpers.
Build Highway in HWY_NO_LIBCXX mode, only compile the target dispatch
sources we actually need, and compile Ghostty's SIMD translation units
with the same define so the header ABI stays consistent. Replace the
upstream abort implementation with a small local bridge that provides
Highway's Warn/Abort hooks and the target-query shim without depending
on libc++.
This keeps the Highway archive down to the dispatch pieces Ghostty
uses while preserving the existing dynamic dispatch behavior. The
bridge is documented so it is clear why Ghostty carries this small
local replacement.
Replace the ImageInfo and PlacementInfo sized structs and their
associated .info enum variants with a new _get_multi pattern that
batches multiple enum+pointer pairs into a single call. This avoids
struct ABI concerns (field order, padding, alignment, GHOSTTY_INIT_SIZED)
while preserving the single-call-crossing performance benefit for FFI
and Cgo callers.
Each _get_multi function takes an array of enum keys, an array of
output pointers, and an optional out_written parameter that reports
how many values were successfully written before any error. This
applies uniformly to all _get APIs: terminal_get, cell_get, row_get,
render_state_get, render_state_row_get, render_state_row_cells_get,
kitty_graphics_image_get, and kitty_graphics_placement_get.
The C example is updated to use compound-literal _get_multi calls,
and tests cover both success and error paths for every new function.
Switch the shared ghostty-internal.pc Libs: line from -lghostty to a
direct ${libdir}/<file> path, matching what the -static module already
does. The name-per-OS helpers now emit:
shared: ghostty-internal.dll (Windows) / ghostty-internal.so (other)
static: ghostty-internal-static.lib (Windows) / ghostty-internal.a
Direct paths sidestep the GNU-ld -l<name> search template, which
expects libghostty-internal.so/.a on Unix - we drop the lib prefix to
match the ghostty-internal pkg-config module name.
Also update the LipoStep out_name for the macOS universal static
archive to ghostty-internal.a for consistency.
Add three sized structs that let callers fetch all image, placement,
or rendering metadata in a single call instead of many individual
queries. This is an optimization for environments with high per-call
overhead such as FFI or Cgo.
GhosttyKittyGraphicsImageInfo is returned via image_get() with the
new GHOSTTY_KITTY_IMAGE_DATA_INFO data kind. It bundles id, number,
width, height, format, compression, data pointer, and data length.
GhosttyKittyGraphicsPlacementInfo is returned via placement_get()
with the new GHOSTTY_KITTY_GRAPHICS_PLACEMENT_DATA_INFO data kind.
It bundles image id, placement id, virtual flag, offsets, source
rect, columns, rows, and z-index.
GhosttyKittyGraphicsPlacementRenderInfo is returned by the new
ghostty_kitty_graphics_placement_render_info() function, which
combines pixel size, grid size, viewport position, and resolved
source rectangle. This one requires image and terminal handles so
it does not fit the existing _get() pattern and is a dedicated
function.
All three use the sized-struct ABI pattern with GHOSTTY_INIT_SIZED
for forward compatibility.
In C ABI builds, the Zig std.log default writes to stderr which is
not appropriate for a library. Override std_options.logFn with a
custom sink that dispatches to an embedder-provided callback, or
silently discards when none is registered.
Add GHOSTTY_SYS_OPT_LOG to ghostty_sys_set() following the existing
decode_png pattern. The callback receives the log level as a
GhosttySysLogLevel enum, scope and message as separate byte slices,
giving embedders full control over formatting and routing.
Export ghostty_sys_log_stderr as a built-in convenience callback that
writes to stderr using std.debug.lockStderrWriter for thread-safe
output. Embedders who want the old behavior can install it at startup
with a single ghostty_sys_set call.
Add a ghostty_vt_add_target() CMake function that lets downstream
projects build libghostty-vt for a specific Zig target triple. The
function encapsulates zig discovery, build-type-to-optimize mapping,
the zig build invocation, and output path conventions so consumers
do not need to duplicate any of that logic. It creates named IMPORTED
targets (e.g. ghostty-vt-static-linux-amd64) that work alongside the
existing native ghostty-vt and ghostty-vt-static targets.
The build-type mapping is factored into a shared _GHOSTTY_ZIG_OPT_FLAG
variable used by both the native build and the new function.
The static library targets now propagate c++ as a link dependency on
non-Windows platforms, fixing link failures when consumers use static
linking with the default SIMD-enabled build.
A new example/c-vt-cmake-cross/ demonstrates end-to-end cross-
compilation using zig cc as the C compiler, auto-detecting a cross
target based on the host OS.
Keep libghostty-vt.pc as the shared/default pkg-config module so
`pkg-config --static libghostty-vt` continues to emit the historical
`-lghostty-vt` flags. This preserves the old behavior for consumers
that still want it, even though that form remains ambiguous on macOS
when both the dylib and archive are installed in the same directory.
Add a separate libghostty-vt-static.pc module for consumers that need
an unambiguous static link. Its `Libs:` entry points directly at the
installed archive so macOS does not resolve the request to the dylib.
Update the Nix packaging to rewrite the new static module into the `dev`
output, use it in the static-link smoke test, and add a compatibility
check that covers both pkg-config entry points.
Fixes#11990
Previously only slashes were replaced with hyphens in the branch
name used as the semver pre-release identifier. Branch names
containing dots (e.g. dependabot branches like
"cachix/install-nix-action-31.10.4") would cause an InvalidVersion
error because std.SemanticVersion only allows alphanumeric
characters and hyphens in pre-release identifiers.
Replace all non-alphanumeric, non-hyphen characters instead of
only slashes.
Add iOS device and simulator slices to the xcframework, gated on
SDK availability via std.zig.LibCInstallation.findNative. Refactor
AppleLibs from a struct with named fields to an EnumMap keyed by
ApplePlatform so that adding new platforms only requires extending
the enum and its sdk_platforms table.
tvOS, watchOS, and visionOS are listed as not yet supported due to
Zig stdlib limitations (missing PATH_MAX, mcontext fields).
The xcframework now generates its own headers directory with a
GhosttyVt module map instead of reusing include/ directly, which
contains the GhosttyKit module map for the macOS app. The generated
directory copies the ghostty headers and adds a module.modulemap
that exposes ghostty/vt.h as the umbrella header.
A new swift-vt-xcframework example demonstrates consuming the
xcframework from a Swift Package. It creates a terminal, writes
VT sequences, and formats the output as plain text, verifying
the full round-trip works with swift build and swift run.
On Darwin targets, the build now automatically produces a universal
(arm64 + x86_64) XCFramework at lib/ghostty-vt.xcframework under
the install prefix. This bundles the fat static library with headers
so consumers using Xcode or Swift PM can link libghostty-vt directly.
Fix three categories of test bugs in the kitty graphics C API tests:
The placement iterator reset in getTyped was clobbering the
layer_filter field when reinitializing the iterator struct,
causing the layer filter test to see unfiltered placements.
Preserve layer_filter across resets.
The viewport position tests were not accounting for the default
cursor_movement=after behavior of the kitty display command,
which calls index() for each row of the placement before the
test scroll sequence. Add C=1 to suppress cursor movement so
the scroll math in the tests is correct.
The source_rect tests used an 88-character all-A base64 payload
which decodes to 66 bytes, but a 4x4 RGBA image requires exactly
64 bytes. Fix the payload to use proper base64 padding (AA==).
Add ghostty_kitty_graphics_placement_source_rect which returns the
fully resolved and clamped source rectangle for a placement. This
applies kitty protocol semantics (width/height of 0 means full
image dimension) and clamps the result to the actual image bounds,
eliminating ~20 lines of protocol-aware logic from each embedder.
Add ghostty_kitty_graphics_placement_viewport_pos which converts a
placement's internal pin to viewport-relative grid coordinates.
The returned row can be negative when the placement's origin has
scrolled above the viewport, allowing embedders to compute the
correct destination rectangle for partially visible images.
Returns GHOSTTY_NO_VALUE only when the placement is completely
outside the viewport (bottom edge above the viewport or top edge
at or below the last row), so embedders do not need to perform
their own visibility checks. Partially visible placements always
return GHOSTTY_SUCCESS with their true signed coordinates.
Add a placement_iterator_set function that configures iterator
properties via an enum, following the same pattern as other set
functions in the C API (e.g. render_state_set). The first settable
option is a z-layer filter.
The GhosttyKittyPlacementLayer enum classifies placements into three
layers based on kitty protocol z-index conventions: below background
(z < INT32_MIN/2), below text (INT32_MIN/2 <= z < 0), and above text
(z >= 0). The default is ALL which preserves existing behavior.
When a layer filter is set, placement_iterator_next automatically
skips non-matching placements, so embedders no longer need to
reimplement the z-index bucketing logic or iterate all placements
three times per frame just to filter by layer.
Add the inverse of ghostty_terminal_grid_ref(), converting a grid
reference back to coordinates in a requested coordinate system
(active, viewport, screen, or history). This wraps the existing
internal PageList.pointFromPin and is placed on the terminal API
since it requires terminal-owned PageList state to resolve the
top-left anchor for each coordinate system.
Returns GHOSTTY_NO_VALUE when the ref falls outside the requested
range, e.g. a scrollback ref cannot be expressed in active
coordinates.
The PlacementIterator, PlacementMap, and PlacementIteratorWrapper
types in the C API were unconditionally referencing
kitty_storage.ImageStorage, which transitively pulled in
Image.transmit_time (std.time.Instant). On wasm32-freestanding,
std.time.Instant requires posix.timespec which does not exist,
causing a compilation error.
Gate these types behind build_options.kitty_graphics, matching the
existing pattern used for KittyGraphics and ImageHandle. When
kitty graphics is disabled, they fall back to opaque/void types.
Add early-return guards to placement_iterator_new and
placement_iterator_free which directly operate on the wrapper
struct.
The test transmits an image with f=24 (24-bit RGB) but was asserting
that the format field equals .rgba (32-bit). Corrected the expectation
to .rgb to match the transmitted pixel format.
Expose Placement.pixelSize() and Placement.gridSize() as new C API
functions ghostty_kitty_graphics_placement_pixel_size() and
ghostty_kitty_graphics_placement_grid_size(). Both take the placement
iterator, image handle, and terminal, returning their results via
out params.
Rename the internal Zig method from calculatedSize to pixelSize to
pair naturally with gridSize — one returns pixels, the other grid
cells. Updated all callers including the renderer.
Expose Placement.rect() from the Zig kitty graphics storage as a new
C API function ghostty_kitty_graphics_placement_rect(). It takes the
terminal, image handle, and a positioned placement iterator, and
writes the bounding grid rectangle into a GhosttySelection out param.
Virtual placements return GHOSTTY_NO_VALUE.
Move all opaque handle typedefs (GhosttyTerminal, GhosttyKittyGraphics,
GhosttyRenderState, GhosttySgrParser, GhosttyFormatter, GhosttyOsc*)
into types.h so they are available everywhere without circular includes
and Doxygen renders them in the correct @ingroup sections.
