[igouss]

I run a Wayland desktop (Hyprland). After its Xwayland socket went away,
DISPLAY=:0 still pointed at it, and vulkaninfo --summary started
segfaulting at exit. On a terminal the summary prints and then it
crashes; piped or redirected (vulkaninfo --summary | grep deviceName)
it prints nothing, because stdout is still buffered when it crashes.

Any setup where DISPLAY is set but XOpenDisplay fails takes the same
path. I hit it after Xwayland went away; a stale DISPLAY over SSH or a
container with DISPLAY set and no server should do the same, though I
haven't tested those.

Cause. AppCreateXlibWindow throws when XOpenDisplay returns null.
The main loop catches that and skips the extension (vulkaninfo.cpp,
the create_window try/catch), but the teardown loop still calls
destroy_window for every surface extension, whether or not its window
was created. AppDestroyXlibWindow then calls
XDestroyWindow(nullptr, ...). (The XCB path got the equivalent null
check in #292.)

Fix. SurfaceExtension records whether create_window succeeded,
and teardown only calls destroy_window for windows that were created.
I fixed it in the loop rather than in AppDestroyXlibWindow because the
same pattern applies to other platforms: from reading the code, DirectFB
and QNX Screen can throw partway through create_window and would then
run their destroy functions on fields that were never set. I could only
build and test the Linux XCB/Xlib/Wayland paths. If you'd rather keep
the change local, a null check in AppDestroyXlibWindow (matching the
existing one in AppDestroyXcbWindow) also fixes this crash; I tested
that too and it behaved identically here.

One trade-off: a create_window that throws partway through now leaks
whatever it had already allocated (e.g. the DirectFB dfb, or the QNX
context) instead of passing it to destroy_window. The process is about
to exit, so I preferred that to running destroy on half-initialized
state.

Repro (nothing listening on :99):

DISPLAY=:99 vulkaninfo --summary

• Expected: summary printed, surface creation errors on stderr, exit 0.
• Actual: Segmentation fault, exit 139 (stdout empty when
  redirected); backtrace
  XDestroyWindow ← AppDestroyXlibWindow.

Testing (Arch Linux, Mesa 26.2.5, libX11 1.8.14; Intel UHD 630 +
AMD Radeon R9 200 / HD 7900 Series (RADV TAHITI)), with VK_LAYER_KHRONOS_validation
enabled:

Case	main (1d3fbab)	this PR
DISPLAY=:99, no X server, --summary	segfault, exit 139	exit 0
DISPLAY=:99, no X server, full output	—	exit 0
DISPLAY unset	—	exit 0
live X server on :99 (Xwayland), full and --summary	exit 0	exit 0; XCB and Xlib surfaces listed (full output)

No validation messages in any run. With a live X server, output is the
same as main apart from the heap budget value, which varies between
runs. Also reproduced the crash with the Arch vulkan-tools 1.4.363.0-1
package.

🤖 Generated with Claude Code

[ci-tester-lunarg]

User igouss not on autobuild list. Waiting for curator authorization before starting CI build.

[ci-tester-lunarg]

User igouss not on autobuild list. Waiting for curator authorization before starting CI build.

[CLAassistant]

All committers have signed the CLA.