feat: Add Windows support via ConPTY by deblasis · Pull Request #6 · ghostty-org/ghostling

deblasis · 2026-03-23T17:25:21Z

Adds Windows support using the ConPTY API for terminal emulation.

Shell spawn via ConPTY with threaded pipe reader, resize, and cleanup
Auto-detect shell (pwsh, powershell, cmd) with --shell override
GUI subsystem app so the child shell writes through the pipe
CMake workarounds for upstream ghostty DLL path and import library on Windows
CI job for Windows

All behind #ifdef _WIN32 in main.c, Unix codepath unchanged.

Tested on Win, Mac and Linux

deblasis · 2026-03-23T17:27:08Z

 functional terminal built on the libghostty C API in a
 [single C file](https://github.com/ghostty-org/ghostling/blob/main/main.c).

-The example uses Raylib for windowing and rendering. It is single-threaded


Removed the threading detail because with windows technically this is an incorrect statement since we have two threads. Happy to revert as needed

mitchellh · 2026-03-23T19:42:29Z

Lets wait for ghostty-org/ghostty#11756 and then I'll ask if you can rebase this and clean it up, since I think that'll maybe improve some of this.

# What PR #11756 added IMPORTED_IMPLIB pointing to the .lib import library, but the import library is not listed in the OUTPUT directive of the `add_custom_command` that runs zig build. The file is produced as a side-effect of the build. This works with the Visual Studio generator (which is lenient about undeclared outputs) but fails with Ninja: ninja: error: 'zig-out/lib/ghostty-vt.lib', needed by 'ghostling', missing and no known rule to make it The fix adds "${ZIG_OUT_DIR}/lib/${GHOSTTY_VT_IMPLIB}" to the OUTPUT list. No behavioral change. The file was already being built, Ninja just needs to know about it. ## but_why.gif I am cleaning up ghostty-org/ghostling#6 and I realise that in order to get rid of the CMake workarounds we had before #11756, this change is necessary. # POC I set up a branch pointing at my fork with a POC and it builds, this is the cleaned up CMakeList https://github.com/deblasis/winghostling/blob/test/cmake-implib-no-workaround/CMakeLists.txt

deblasis · 2026-03-24T04:10:30Z

The #ifdef _WIN32 blocks in main() could be consolidated behind a unified Pty struct with platform-specific implementations (eg: pty_drain(), pty_reap(), pty_resize()), eliminating all conditionals from main().

I kept inline #ifdef blocks to match the existing style in the codebase, but happy to refactor if you'd prefer that approach.

Anything else, fire away.

mitchellh

Not bad. I agree the ifdefs are pretty ugly. I'll have to think about it. I put a couple simple requests otherwise.

mitchellh · 2026-03-24T04:12:57Z

@@ -1,6 +1,55 @@
+// ---------------------------------------------------------------------------
+// Platform-independent headers
+// ---------------------------------------------------------------------------


We don't need this at the top.

mitchellh · 2026-03-24T04:14:12Z

-int main(void)
+int main(int argc, char *argv[])
 {
+    // Parse --shell <path> flag for explicit shell selection.


let's just change this to argv[1] (no parsing).

deblasis · 2026-03-24T04:40:01Z

Not bad. I agree the ifdefs are pretty ugly. I'll have to think about it. I put a couple simple requests otherwise.

Awesome! Done and done and tested.

res2k · 2026-04-20T22:56:56Z

I gave these changes a go, and I have some observations:

"Send an initial carriage return" - not necessary for me, I can see the initial process output just fine.
In fact, this is actually unwanted input to the shell and should probably be removed.
PROPERTIES WIN32_EXECUTABLE TRUE wasn't necessary, either; even without it, output was never mistakenly sent to ghostling's console, yet ghostling's own output can be inspected.
The process exit detection doesn't quite work; it appears the child process exiting doesn't automatically results in the PTY getting closed, and so a PTY read error can't be used to detect that event. Instead, the WaitForSingleObject() on the process handle will detect the child having exited.

Add a build-windows job that runs on windows-latest with Zig + Ninja. Set WIN32_EXECUTABLE TRUE so the app launches as a GUI subsystem process without spawning an extra console window. MSVC/Clang need an explicit /ENTRY:mainCRTStartup override; MinGW handles this automatically. Signed-off-by: Alessandro De Blasis <alex@deblasis.net>

Add a complete Windows backend using the ConPTY (pseudo-console) API, gated by #ifdef _WIN32 alongside the existing Unix pty code. Platform abstraction: - PtyHandle typedef (HANDLE on Windows, int on Unix) so pty_write, handle_input, handle_mouse, etc. share signatures across platforms - WIN32_LEAN_AND_MEAN + NOGDI + NOUSER to avoid symbol clashes with raylib (Rectangle, CloseWindow, ShowCursor, PlaySound) ConPTY implementation: - pty_spawn_win32(): creates pipes, pseudo-console, attribute list, and launches the child shell (pwsh > powershell > cmd fallback) - Threaded pipe reader to work around PeekNamedPipe unreliability - pty_buf_drain() called each frame from the main loop - pty_resize() delegates to ResizePseudoConsole - pty_cleanup() with 1s graceful timeout before TerminateProcess Integration: - AttachConsole(ATTACH_PARENT_PROCESS) + CONOUT$ redirect so logs are visible when launched from a terminal (no-op from Explorer) - WaitForSingleObject() for child exit detection (ConPTY pipes don't reliably close on child exit) - Shell detection via SearchPathW, --shell flag support - TERM=xterm-256color set via SetEnvironmentVariableA - placement_iter freed on both platforms Co-authored-by: Mitchell Hashimoto <m@mitchellh.com> Signed-off-by: Alessandro De Blasis <alex@deblasis.net>

deblasis · 2026-04-21T01:16:54Z

Hi @res2k, thank you for taking the time to look into this.

I gave these changes a go, and I have some observations:

"Send an initial carriage return" - not necessary for me, I can see the initial process output just fine.
In fact, this is actually unwanted input to the shell and should probably be removed.

You are right, removed.

PROPERTIES WIN32_EXECUTABLE TRUE wasn't necessary, either; even without it, output was never mistakenly sent to ghostling's console, yet ghostling's own output can be inspected.

Without WIN32_EXECUTABLE TRUE, launching from Explorer (clicking the icon) opens an extra console window alongside the raylib window. The child shell inherits the parent's console subsystem and opens its own. The STARTF_USESTDHANDLES with NULL handles prevents I/O from going to the wrong place, but it doesn't prevent the console window itself from appearing.

Launching from the terminal instead just swallows the output. I made a change that perhaps hits two birds with one stone (without harming animals). Now launching from terminal gets the output from the caller, same behaviour as linux/mac where there's no subsystem and logs go to the parent terminal.

The process exit detection doesn't quite work; it appears the child process exiting doesn't automatically results in the PTY getting closed, and so a PTY read error can't be used to detect that event. Instead, the WaitForSingleObject() on the process handle will detect the child having exited.

Good catch, addressed. The ConPTY pipe doesn't close on child exit, so we now poll the process handle with WaitForSingleObject() each frame to detect exit independently.

res2k · 2026-04-21T06:44:31Z

Now launching from terminal gets the output from the caller, same behaviour as linux/mac where there's no subsystem.

This matches macOS/Linux behavior where logs go to the parent terminal.

Alright, that's a sensible approach.

deblasis commented Mar 23, 2026

View reviewed changes

deblasis mentioned this pull request Mar 24, 2026

cmake: add import library to custom command OUTPUT ghostty-org/ghostty#11794

Merged

deblasis force-pushed the feat/windows-port branch 2 times, most recently from 39ccb3a to 224bcc4 Compare March 24, 2026 04:06

mitchellh requested changes Mar 24, 2026

View reviewed changes

deblasis requested a review from mitchellh March 24, 2026 04:40

deblasis and others added 2 commits April 21, 2026 03:12

deblasis force-pushed the feat/windows-port branch from 37a33db to 616c675 Compare April 21, 2026 01:16

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: Add Windows support via ConPTY#6

feat: Add Windows support via ConPTY#6
deblasis wants to merge 2 commits into
ghostty-org:mainfrom
deblasis:feat/windows-port

deblasis commented Mar 23, 2026

Uh oh!

deblasis Mar 23, 2026

Uh oh!

mitchellh commented Mar 23, 2026

Uh oh!

deblasis commented Mar 24, 2026

Uh oh!

mitchellh left a comment

Uh oh!

mitchellh Mar 24, 2026

Uh oh!

mitchellh Mar 24, 2026

Uh oh!

deblasis commented Mar 24, 2026

Uh oh!

res2k commented Apr 20, 2026

Uh oh!

deblasis commented Apr 21, 2026 •

edited

Loading

Uh oh!

res2k commented Apr 21, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

deblasis commented Mar 23, 2026

Uh oh!

deblasis Mar 23, 2026

Choose a reason for hiding this comment

Uh oh!

mitchellh commented Mar 23, 2026

Uh oh!

deblasis commented Mar 24, 2026

Uh oh!

mitchellh left a comment

Choose a reason for hiding this comment

Uh oh!

mitchellh Mar 24, 2026

Choose a reason for hiding this comment

Uh oh!

mitchellh Mar 24, 2026

Choose a reason for hiding this comment

Uh oh!

deblasis commented Mar 24, 2026

Uh oh!

res2k commented Apr 20, 2026

Uh oh!

deblasis commented Apr 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

res2k commented Apr 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

deblasis commented Apr 21, 2026 •

edited

Loading

res2k commented Apr 21, 2026 •

edited

Loading