| Ticket UUID: | a04f2a19812703cad0621338e0a7bcbe68327a11 | |||
| Title: | Wayland branch: wish crashes in glfwInit() on a compositor without a seat (headless Weston) | |||
| Type: | Bug | Version: | wayland | |
| Submitter: | serhiy.storchaka | Created on: | 2026-10-14 16:01:42 | |
| Subsystem: | (unused) | Assigned To: | aku | |
| Priority: | 5 Medium | Severity: | Minor | |
| Status: | Closed | Last Modified: | 2026-10-22 17:22:38 | |
| Resolution: | Fixed | Closed By: | serhiy.storchaka | |
| Closed on: | 2026-10-14 17:58:17 | |||
| Description: |
On the wayland branch ([438c60c73f]), wish crashes at startup with a segmentation fault when the compositor offers no wl_seat, as Weston 14 does with --backend=headless: #0 wl_proxy_get_version () from libwayland-client.so.0 #1 ?? () from libglfw.so.3 #2 glfwInit () from libglfw.so.3 #3 TkpOpenDisplay () at wayland/tkWaylandGC.c:118 #4 GetScreen () at generic/tkWindow.c:483 The crash is in GLFW 3.4 (Ubuntu 26.04, libglfw3 3.4-4): a program which only calls glfwInit() crashes the same way, also with libdecor disabled (GLFW_WAYLAND_DISABLE_LIBDECOR). But a headless compositor is what a test run on CI would use, and wish should fail to open the display rather than crash. TkpOpenDisplay could check that the registry has a wl_seat before calling glfwInit(). Headless sway (WLR_BACKENDS=headless WLR_LIBINPUT_NO_DEVICES=1) offers a seat, and the wish of the branch runs there. | |||
| User Comments: |
serhiy.storchaka added on 2026-10-14 17:57:17:
Thanks. Merged into wayland in [83fe22d94c]. kevin_walzer added on 2026-10-14 17:51:04: Confirmed the bug and fix, thank you. Can you please merge to wayland branch? serhiy.storchaka added on 2026-10-14 17:27:17: Even if Weston is not used for CI, wish should not crash where it cannot run. Proposed fix in branch wayland-require-seat, [afc5d6aed3], based on the current wayland: TkpOpenDisplay checks that the compositor offers a wl_seat before calling glfwInit(). Under headless Weston wish now reports couldn't connect to display "wl-test" instead of crashing; under headless sway it runs as before. kevin_walzer added on 2026-10-14 16:19:09: This is a known issue. Weston was our initial choice for CI, but these crashes confirm Weston is too limited for CI testing and is not worth trying to accommodate. The entire Wayland compositor environment is highly fragmented and we are largely relying on GLFW to address those discrepancies. The compositor we will use for the Github CI is headless mutter, which has proven to be robust and handles Tk Wayland with no issue. We are not yet ready to deploy this setup because the test suite failures are too noisy right now - there are hundreds of failures and we need community testing to determine which failures are directly related to Wayland, which should be constrained because they reflect invalid X11 assumptions, and so on. | |||
Home
Timeline
Branches
Tags
Forum
Tickets
Download
Wiki