arch

In a Chrome Remote Desktop X session on PipeWire, apps were silent or failed with "Host is down". CRD puts PULSE_RUNTIME_PATH, PULSE_SINK and PIPEWIRE_REMOTE into the session, pointing at its own PipeWire, which had not started. Unset them before dbus-launch, bridge system PipeWire to CRD's FIFO, check sink volume.

faster-whisper fails at the first GPU compute with "Library libcublas.so.12 is not found" because the ctranslate2 wheel is built for CUDA 12 and the system has CUDA 13. Install nvidia-cublas-cu12 and nvidia-cudnn-cu12 into the venv and set LD_LIBRARY_PATH in a wrapper before Python starts.

claude auth login completes in the browser, then Claude Code says it is not logged in and gives up after a few retries. ~/.claude was owned by root after claude had once been run as root with the user's HOME, so the token file could not be written. chown the directory back to the user.

After a kernel and NVIDIA driver upgrade without a reboot, restarting X gave a black screen and "Failed to initialize the NVIDIA kernel module". The running kernel had only the old module loaded. Build the module for the running kernel with DKMS, swap the modules, recreate /dev/dri, then reboot properly.

CUPS jobs to an ipp://printer.local queue never printed and lpstat showed "Unable to locate printer". Avahi could see the printer, but /etc/nsswitch.conf had no mdns entry, so normal name lookups for .local failed. Adding mdns4_minimal [NOTFOUND=return] to the hosts line fixed it.

Clicking the folder icon in Chromium's downloads panel gave '"." is not a valid folder'. Baobab (GNOME Disk Usage Analyzer) was the default handler for inode/directory, so xdg-open launched it. Making a file manager the default in ~/.config/mimeapps.list fixed it.