The emulator in ~/Android/Sdk auto-updated to 37.2.12 on Oct 11 and since
then consistently fails its own KVM probe ("/dev/kvm is not found") and
segfaults on that failure path — while KVM is healthy: a manual
KVM_GET_API_VERSION ioctl returns the current API level 12, and both
/opt/android-sdk and the container-home emulator pass -accel-check on
the same kernel and device node. The script now prefers the documented
host SDK at /opt/android-sdk when no ANDROID_SDK_ROOT is set.
Once the home SDK is repaired (or the emulator regression disappears),
the fallback order can be revisited.
Co-Authored-By: Claude <noreply@anthropic.com>
The discrete card now runs continuous AI inference, so the emulator must
not claim GL/Vulkan contexts on it. In windowed mode the emulator uses
the host GL/Vulkan stack, and the default GLVND selection picks the
discrete card on this machine. The script now discovers the first
display device whose driver is not nvidia and pins the emulator process
to it (Mesa GLX/EGL + DRI_PRIME + the Intel Vulkan ICD).
Headless (the default) needs no change: it always ran on swiftshader,
strictly software, and never touched any GPU.
Co-Authored-By: Claude <noreply@anthropic.com>
The script's set -euo pipefail made it die without any visible error in
two places: an empty /tmp/jtube-adbkeys/ failed the glob cat (the error
itself was discarded, so the output just stopped mid-script — this hits
any host after a reboot), and the boot-wait loop was killed on its first
iteration because adb exits non-zero while the device is still booting.
Both were swallowing the whole && chain (adb connect, scrcpy), making it
look like the emulator was broken when it was fine or about to be.
The stale bridge-emulator.sh duplicate (different port, broken vendor
keys path) is removed in the same spirit — keeping it invites running
the wrong script. Docs updated to match the one script.
Document how the host emulator is driven over adb from the agent
container: the start script now boots headless by default and trusts
remote adb keys via ADB_VENDOR_KEYS at launch, since the TV system
image is a user build that blocks pushing keys to adb_keys after boot.
AGENTS.md points to the new doc and fixes the stale SDK path.
On launch and on every destination switch the first card of the
content gets focus automatically - invisible focus on chrome confused
'Do I need to press DOWN first?'. The search button shows a white ring
when focused, and the top bar gained a real account button (person
icon + logged-in username) replacing the cryptic monitor icon.
start-emulator.sh: headless-safe boot (no -gpuhost crash without a
display), whitelists adb public keys headlessly (drop them into
/tmp/jtube-adbkeys/), and starts the socat bridge itself.
scrcpy cannot talk to headless emulators inside the container: the
AOSP ATD image lacks framework pieces (KeyCharacterMap NPE) and there
is no GPU/KVM in the container at all. The script boots the android-tv
image image on the host with -gpu host and a window (so the RSA dialog
is clickable), installs the current debug APK and leaves the app
running; the container's socat bridge then exposes it as
rutherford:5556 for adb/scrcpy from anywhere.
