Environment: Batocera.linux 43.1, kernel 6.18.16, GTX 1050 Ti / proprietary driver 580.126.09, Steam flatpak 1.0.0.85 (flathub), launched via -oldbigpicture or via the Big Picture button from the desktop client.
Desktop mode works perfectly. Switching to Big Picture produces a permanent black screen, with no error message and no crash. Full live diagnostic performed on the machine (logs, pressure-vessel sandbox inspection, direct window pixmap read via XComposite):
• steamwebhelper and all its subprocesses (zygote, GPU, network, audio) stay alive and healthy the whole time — no crash.
• GL_VENDOR/GL_RENDERER confirm the NVIDIA driver is active throughout — no software fallback.
• The top-level X11 window is correctly created, positioned, focused, and topmost in the stacking order — not a WM/stacking issue.
• Direct read of the window's composite pixmap (manual redirect + XGetImage): every pixel is 0xff000000 — genuinely painted opaque black, not an empty/transparent buffer.
• Internal logs (steamui_html.txt) show the UI's JS loading and running normally, but routed to "ensure desktopui window" instead of "ensure gamepadui window exists" on switch — despite -oldbigpicture being explicitly passed.
Separately, and possibly related to the same family of container-bootstrap bugs: the pressure-vessel sandbox intermittently fails to capture 10_nvidia.json (host symlink into /var/run/nvidia/configs/, possibly not resolved by the capture logic) and never captures egl_external_platform.d for GBM at all — the fix in #485 (2022) appears to have only covered the Wayland external platform.
Happy to test any candidate fix on this setup. Full logs and diagnostic scripts available on request.
Environment: Batocera.linux 43.1, kernel 6.18.16, GTX 1050 Ti / proprietary driver 580.126.09, Steam flatpak 1.0.0.85 (flathub), launched via
-oldbigpictureor via the Big Picture button from the desktop client.Desktop mode works perfectly. Switching to Big Picture produces a permanent black screen, with no error message and no crash. Full live diagnostic performed on the machine (logs, pressure-vessel sandbox inspection, direct window pixmap read via XComposite):
• steamwebhelper and all its subprocesses (zygote, GPU, network, audio) stay alive and healthy the whole time — no crash.
• GL_VENDOR/GL_RENDERER confirm the NVIDIA driver is active throughout — no software fallback.
• The top-level X11 window is correctly created, positioned, focused, and topmost in the stacking order — not a WM/stacking issue.
• Direct read of the window's composite pixmap (manual redirect + XGetImage): every pixel is 0xff000000 — genuinely painted opaque black, not an empty/transparent buffer.
• Internal logs (steamui_html.txt) show the UI's JS loading and running normally, but routed to "ensure desktopui window" instead of "ensure gamepadui window exists" on switch — despite -oldbigpicture being explicitly passed.
Separately, and possibly related to the same family of container-bootstrap bugs: the pressure-vessel sandbox intermittently fails to capture 10_nvidia.json (host symlink into /var/run/nvidia/configs/, possibly not resolved by the capture logic) and never captures egl_external_platform.d for GBM at all — the fix in #485 (2022) appears to have only covered the Wayland external platform.
Happy to test any candidate fix on this setup. Full logs and diagnostic scripts available on request.