startcce: import VK_DRIVER_FILES into the systemd user manager

Apps spawned by cce-cloud (a user service) inherit the manager's
environment, not startcce's, so they ran without the session's Vulkan
pin. Houdini then loaded the NVIDIA Vulkan ICD alongside Mesa GL and
segfaulted inside libGLX_nvidia six seconds after every launch from the
app menu, while a compositor-spawned instance was fine. Import the pin
like RUSTICL_ENABLE so both launch paths see the same GPU setup; also
correct the rusticl comment, which still described the NVIDIA ICD as
having no hardware behind it.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

 scripts/startcce | 19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/scripts/startcce b/scripts/startcce
index 342db5c..91dcd31 100755
--- a/scripts/startcce
+++ b/scripts/startcce
@@ -20,12 +20,21 @@ export XCURSOR_SIZE=24
 export XCURSOR_PATH="${XDG_DATA_HOME:-$HOME/.local/share}/icons:$HOME/.icons:/usr/share/icons"
 export WLR_NO_HARDWARE_CURSORS=1
 export WLR_DRM_DEVICES=/dev/dri/card1
+# The session renders on the Intel GPU, so Vulkan is pinned to the Intel ICD.
+# This is not only a compositor concern: an app that also loads the NVIDIA
+# Vulkan ICD (libGLX_nvidia.so.0) and then asks NVIDIA OpenCL about its
+# Mesa GL context segfaults inside the NVIDIA driver — Houdini did, six
+# seconds after launch, every time it came from the launcher. It is
+# imported into the systemd user manager below for that reason: apps spawned
+# by cce-cloud (a user service) inherit the manager's environment, not this
+# script's, and without the import they ran under a different GPU setup than
+# anything the compositor spawned directly.
 export VK_DRIVER_FILES=/usr/share/vulkan/icd.d/intel_icd.json
 # Mesa's rusticl exposes NO OpenCL devices unless enabled per driver — without
-# this the session has zero OpenCL platforms (the other vendor ICD is a stale
-# opencl-nvidia with no NVIDIA hardware behind it), and every OpenCL consumer
-# fails quietly: the designer's kernel-generated geometry (the default
-# project's sphere) simply never appears.
+# this the session has zero OpenCL platforms from Mesa, and every OpenCL
+# consumer that wants the Intel GPU fails quietly: the designer's
+# kernel-generated geometry (the default project's sphere) simply never
+# appears. (The NVIDIA ICD is live too and serves the discrete GPU.)
 export RUSTICL_ENABLE=iris
 
 # The compositor binary this session runs, resolved once so the server launch
@@ -91,7 +100,7 @@ fi
 # literal '$CCE_BIN' and the client would never start.
 cat > "$CCE_LAUNCH" << LAUNCH_EOF
 #!/bin/sh
-systemctl --user import-environment DISPLAY WAYLAND_DISPLAY XDG_CURRENT_DESKTOP XDG_SESSION_TYPE XDG_SESSION_ID XDG_RUNTIME_DIR GNOME_KEYRING_CONTROL SSH_AUTH_SOCK RUSTICL_ENABLE${TERMINAL:+ TERMINAL}
+systemctl --user import-environment DISPLAY WAYLAND_DISPLAY XDG_CURRENT_DESKTOP XDG_SESSION_TYPE XDG_SESSION_ID XDG_RUNTIME_DIR GNOME_KEYRING_CONTROL SSH_AUTH_SOCK RUSTICL_ENABLE VK_DRIVER_FILES${TERMINAL:+ TERMINAL}
 systemctl --user start cce-session.target
 ${DEBUG_FLAG}exec "$CCE_BIN" client 2>"$CCE_CLIENT_LOG"
 LAUNCH_EOF