Zig's ubsan runtime emits /exclude-symbols linker directives that
are incompatible with the MSVC linker, causing LNK4229 warnings and
LNK1143 errors. Disable bundling ubsan_rt on Windows while keeping
compiler_rt which provides essential symbols like memcpy, memset,
memmove, and ___chkstk_ms.
The previous check used target.result.abi == .msvc which never
matched because Zig defaults to the gnu ABI on Windows.
Zig's ubsan instrumentation emits ELF-style /exclude-symbols linker
directives into the compiled object files, causing LNK4229 warnings
with the MSVC linker. The bundled compiler_rt also produces COMDAT
sections that are incompatible with MSVC, causing fatal LNK1143.
Disable sanitize_c entirely on the root module for MSVC targets and
skip bundling both compiler_rt and ubsan_rt since MSVC provides its
own runtime.
Zig's bundled compiler_rt and ubsan_rt produce object files with
ELF-style linker directives (/exclude-symbols) and COMDAT sections
that are incompatible with the MSVC linker, causing LNK1143 and
LNK4229 errors when linking the static library.
MSVC provides its own compiler runtime so bundling Zig's versions
is unnecessary. Skip bundling both runtimes when the target ABI is
MSVC.
Use writerStreaming() instead of writer() for stdout in helpgen and
main_build_data. The positional writer calls setEndPos/ftruncate in
end(), which fails on Windows when stdout is redirected via
captureStdOut() because ftruncate maps INVALID_PARAMETER to
FileTooBig. Streaming mode skips truncation entirely since stdout
is inherently a sequential stream.
Replace scandir with opendir/readdir plus qsort in framegen since
scandir is a POSIX extension not available on Windows.
The X11 color map parser in x11_color.zig uses @embedFile to load
rgb.txt at comptime, then splits on \n. On Windows, git may check
out rgb.txt with CRLF line endings, leaving a trailing \r on each
line. This caused color names to be stored as e.g. "white\r" instead
of "white", so all X11 color lookups failed at runtime.
Strip trailing \r from each line before parsing. Also mark rgb.txt
as -text in .gitattributes to prevent line ending conversion in
future checkouts.
Extract the platform-specific page backing memory allocation into
AllocPosix and AllocWindows structs behind a PageAlloc comptime
switch. Previously, POSIX mmap calls were inlined at each call
site. This adds a Windows VirtualAlloc implementation and routes
all allocation through PageAlloc.alloc/free, making the backing
memory strategy consistent and easier to extend.
The colors_get function used structSizedFieldFits to guard the
palette copy, which required the entire palette array to fit in the
provided size. This prevented partial palette writes when the caller
passed a truncated sized struct, since the guard failed even though
the inner code already handled partial copies correctly. Remove the
outer guard so the existing partial-copy logic applies.
The setopt_from_terminal test expected alt_esc_prefix to be false on
a fresh terminal, but the mode definition in modes.zig sets its
default to true. Update the test expectation to match.
The inline else switch in each C API getter expands the .invalid
case, which has OutType void. When called with .invalid and a null
out pointer, the @ptrCast(@alignCast(out)) panics before getTyped
can return early.
Handle .invalid explicitly in the outer switch of every getter to
short-circuit before the pointer cast. This affects build_info,
cell, row, terminal, osc, and render (three getters).
Previously WindowPaddingBalance was defined inside Config.zig, which
meant tests for renderer sizing had to pull in the full config
dependency. Move the enum into renderer/size.zig as PaddingBalance
and re-export it from Config so the public API is unchanged. This
lets size.zig tests run without depending on Config.
This is a continuation of the solid work done by @jcollie in PR #7864. I
checked with him if I could take over to continue the implementation.
His changes of last year have been adapted to be compatible with the
current GTK implementation. Aside from just "making it work", I also
dived into the portals and OpenURI implementation and made some
improvements there.
Notable improvements were:
- Improved lifecycle management of glib resources in the OpenURI
implementation
- More forgiving error handling in OpenURI implementation by adding more
fallbacks
- Fixed some memory leaks
- Less memory allocations in Portals implementation
- Added tests for building the Portals request path
Fixes#5991
Fixes#11705
Add bg_color and fg_color options to GhosttyRenderStateRowCellsData that
resolve the final RGB color for a cell, flattening the multiple possible
sources. For background, this handles content-tag bg_color_rgb,
content-tag bg_color_palette (looked up in the palette), and the style
bg_color. For foreground, this resolves palette indices through the
palette; bold color handling is not applied and is left to the caller.
Both return GHOSTTY_INVALID_VALUE when no explicit color is set, in
which case the caller should fall back to whatever default color it
wants (e.g. the terminal background/foreground).
Fixes#11705
Add bg_color and fg_color options to GhosttyRenderStateRowCellsData
that resolve the final RGB color for a cell, flattening the multiple
possible sources. For background, this handles content-tag bg_color_rgb,
content-tag bg_color_palette (looked up in the palette), and the
style bg_color. For foreground, this resolves palette indices through
the palette; bold color handling is not applied and is left to the
caller.
Both return GHOSTTY_INVALID_VALUE when no explicit color is set, in
which case the caller should fall back to whatever default color it
wants (e.g. the terminal background/foreground).
Fixes#11704
The RenderState empty initializer set both background and foreground
to the default RGB value of black (0, 0, 0), making text unreadable
when a caller has not explicitly configured terminal colors via
DynamicRGB. This is the common case for libghostty consumers.
Default the foreground to white so that the initial render state
provides readable white-on-black text out of the box.
Long term we also need to expose setting the default colors for a
Terminal instance but this is a workable fix in the mean time.
#11706
Add a new GHOSTTY_TERMINAL_DATA_MOUSE_TRACKING option to the
ghostty_terminal_get API. This returns true if any mouse tracking
mode is active (X10, normal, button, or any-event), replacing the
need for consumers to loop over four separate mode queries.
The static library was built without position-independent code,
which caused linker errors when consumers tried to link it into
PIE executables (the default on most Linux distributions). The
linker would fail with "relocation R_X86_64_32 against symbol
cannot be used when making a PIE object."
Enable PIC on the static library root module so it can be linked
into both PIE and non-PIE executables.
Refactor GhosttyLibVt to support both shared and static library
builds via a shared initLib helper that accepts a LinkMode. The
shared and static entry points (initShared, initStatic) delegate
to this common path.
For static builds, compiler_rt and ubsan_rt are bundled to avoid
undefined symbol errors. Debug symbols (dsymutil) are skipped for
static libs since they are not linked. The install artifact uses
a "-static" suffix internally but installs as "libghostty-vt.a"
via a new installLib method. Wasm is excluded from static builds
since it has no meaningful static vs shared distinction.
Previously, every call to vt_write created a fresh ReadonlyStream with
new Parser and UTF8Decoder state. This meant escape sequences split
across write boundaries (e.g. ESC in one write, [27m in the next)
would lose parser state, causing the second write to start in ground
state and print the CSI parameters as literal text.
The C API now stores a persistent ReadonlyStream in the TerminalWrapper
struct, which is created when the Terminal is initialized. The vt_write
function feeds bytes through this stored stream, allowing it to maintain
parser state across calls. This change ensures that escape sequences
split across write boundaries are correctly parsed and rendered.
Add GHOSTTY_BUILD_INFO_OPTIMIZE to query the Zig optimization mode
(debug, release safe/small/fast) the library was compiled with. This
reads directly from builtin.mode at comptime so it requires no build
system plumbing.
Add a new C API function ghostty_build_info() that exposes compile-time
build options to library consumers. This allows callers to query whether
SIMD, Kitty graphics protocol, and tmux control mode support were
enabled at build time.
Remove the dedicated `zig build lib-vt` step and replace it with a
`-Demit-lib-vt` build option. This fixes two problems:
1. We can default XCFramework, app, etc. steps to false if emit-lib-vt
is true, so that the lib-vt build doesn't pull in unrelated artifacts.
**Most importantly, lib-vt alone can be build without full Xcode
installations.**
2. We can build lib-vt as part of a bundle with other artifacts if we
really want.
Remove the dedicated `zig build lib-vt` step and replace it with a
`-Demit-lib-vt` build option. This fixes two problems:
1. We can default XCFramework, app, etc. steps to false if emit-lib-vt
is true, so that the lib-vt build doesn't pull in unrelated
artifacts. **Most importantly, lib-vt alone can be build without
full Xcode installations.**
2. We can build lib-vt as part of a bundle with other artifacts if we
really want.
Add two new CellData variants to extract background color values
directly from cells. color_palette (10) returns the palette index
as a GhosttyColorPaletteIndex and color_rgb (11) returns the RGB
components as a GhosttyColorRgb. Both reuse the existing color
types from color.h rather than introducing new ones.
These are only valid when the cell content_tag is
bg_color_palette or bg_color_rgb respectively; querying them
with a mismatched tag reads from the wrong union member.
This adds a function to the core surface to get process information
about the process(es) running in the terminal. Currently supported is
the PID of the foreground process and the name of the slave PTY.
If there is an error retrieving the information, or the platform does
not support retieving that information `null` is returned.
This will be useful in exposing the foreground PID and slave PTY name to
AppleScript or other APIs.
The GRAPHEMES_BUF data kind previously required a double pointer
(pointer to a uint32_t*) because the OutType was [*]u32, making the
typed out parameter *[*]u32. Change OutType to u32 so that callers
pass a plain uint32_t* buffer directly, which is the natural C
calling convention. The implementation casts the out pointer to
[*]u32 internally to write into the buffer.
The STYLE data kind read directly from the render state style array
without checking whether the cell actually had non-default styling.
The style data is undefined for unstyled cells, so this caused a
panic on a corrupt enum value when the caller read the style of an
unstyled cell. Now check cell.hasStyling() first and return the
default style for unstyled cells.
Expand the c-vt-render example to exercise dirty tracking, color
retrieval, cursor state, row/cell iteration with style resolution,
and dirty state reset. Break the example into six doxygen snippet
regions and reference them from render.h.
Expose the cursor fields from RenderState.Cursor through the C API
via new GhosttyRenderStateData enum values. This adds getters for
visual style, visibility, blink state, password input detection,
and viewport position (x, y, wide tail).
A new GhosttyRenderStateCursorVisualStyle enum maps the Zig
cursor.Style values (bar, block, underline, block_hollow) to
stable C integer constants. Viewport position getters return
GHOSTTY_INVALID_VALUE when the cursor is not visible within
the viewport.
Add individual color data kinds to GhosttyRenderStateData so callers
can query background, foreground, cursor color, cursor-color presence,
and the full 256-color palette through ghostty_render_state_get()
without using the sized-struct colors API.
COLOR_CURSOR returns GHOSTTY_INVALID_VALUE when no explicit cursor
color is set; callers can check COLOR_CURSOR_HAS_VALUE first.
Add next, select, and get functions to the render state row cells
API, mirroring the row iterator pattern. row_cells_next advances to
the next cell sequentially, row_cells_select jumps to a specific
column index with bounds validation, and row_cells_get queries data
for the current cell position.
The get function supports querying raw cell values (GhosttyCell),
resolved styles (GhosttyStyle), grapheme codepoint counts, and
writing grapheme codepoints into a caller-provided buffer.
Also add Cell.C and Cell.cval() to page.zig, matching the existing
Row.C/Row.cval() pattern, so the render state can convert cells to
the C ABI type without a raw bitCast.
Change row_iterator_new to only allocate with undefined fields,
matching the pattern used by row_cells_new. The iterator is now
populated via the render state get API with a new .row_iterator
data kind, which slices the row data and resets y to null.
This separates the lifetime of the opaque handle from the render
state it iterates, letting callers allocate once and re-populate
from different states without reallocating.
Add a new opaque RowCells type that wraps per-row cell data
(raw cells, graphemes, styles) for the C API. The caller
allocates a RowCells handle via row_cells_new, then populates
it by passing it to row_get with the new .cells data kind.
This queries the current row from the iterator and slices the
underlying MultiArrayList into the RowCellsWrapper fields.
The new type and functions are wired through main.zig,
lib_vt.zig, and the render.h C header.
Replace ghostty_render_state_row_dirty_get and
ghostty_render_state_row_dirty_set with generic
ghostty_render_state_row_get and ghostty_render_state_row_set
functions using enum-dispatched data/option kinds.
Replace the individual ghostty_render_state_size_get,
ghostty_render_state_dirty_get, and ghostty_render_state_dirty_set
functions with generic ghostty_render_state_get and
ghostty_render_state_set functions that use enum-dispatched data
kinds and option kinds, following the same InType/OutType pattern
used by the terminal and mouse encoder C APIs.
When some tools spawn subshells, PROMPT_COMMAND may be inherited as an
environment variable while the __ghostty_hook function is not (bash
doesn't export functions by default). This causes "command not found"
errors on every prompt in the subshell.
Add 2>/dev/null to the __ghostty_hook entry in PROMPT_COMMAND so that it
silently no-ops in subshells where the function isn't defined. This also
silences any errors from inside __ghostty_hook itself, but those are all
terminal escape sequences and non-actionable.
See: #11245
We previously used a readonly variable (__ghostty_ps0) to define the
best __ghostty_preexec_hook expansion for the current bash version.
This worked pretty well, but it had the downside of managing another
variable (#11258).
We can instead simplify this a bit by moving this into __ghostty_hook. I
didn't take that approach originally because I wanted to avoid the bash
version check on each command, but slightly loosening our guard check to
just look for "__ghostty_preexec_hook" (rather than the full expansion
expression) means we can bury the bash version check to the cold path.
One small gap here is that we may not update PS0 to the correct syntax
if we start switching between significantly different bash versions in
interactive subshells, but that seems like a pretty rare case to handle
given the benefits of this approach.
