The same NativeScript conversion leak as the frame path (see
java-direct-buffer.ts) applied to every other ArrayBuffer handed to a
Java ByteBuffer parameter:
- Shell scenes (submitShellScene on the bridge and the preview display),
submitted on every shell render, which the frame-path fix predates.
These now go through a reused JavaDirectBuffer like frames do.
- Glyph, firmware-glyph and image atlas registration, Mapbox image
decode, the preview placeholder bitmap, firmware hash/write, binary
file writes and the EvenHub privacy-policy download, plus the Kotlin
protocol smoke test. These are rare or large, so they cross through a
new copyToJavaByteBuffer: a fresh Java direct buffer per call that is
freed normally, rather than one pinned for the life of the isolate.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
connect() registers window surfaces in a loop that reads the foreground
window once, before the loop, then awaits each surface's configure call
before queuing its visibility. On a fresh process restoreOpenApps runs at
the same time and focuses windows as it relaunches them. The bridge runs
Java calls in order, so a visibility decided before the await lands after
the focus change's own calls and undoes them: the old foreground can stay
visible and cover the new one, and a later window switch doesn't clear
it, because a switch only hides the window being left.
configureWindowSurface now also accepts visibility as a function, asked
after the configure call, when the visibility call is queued. The display-
attach loops (glasses and preview), the display-mode, per-app layout and
vertical-position changes, and an in-process window's height-mode change
pass one. Launch paths that pass a literal false are unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Passing a JS ArrayBuffer to a Java ByteBuffer parameter leaks it in
@nativescript/android 9.0.4 through 9.1.1: JsArgConverter wraps the backing store with
NewDirectByteBuffer and takes a JNI global ref that it never deletes. The
ref keeps the ByteBuffer reachable, so makeInstanceWeakAndCheckIfAlive
always answers "alive", the JS finalizer re-arms, and V8 never frees the
ArrayBuffer. submitSurfaceFrame passed a fresh copy of each frame this way,
so every frame leaked a full surface (576x260, 640x452 or 640x480 bytes)
plus one global ref, including frames that Java then discards as unchanged.
The draw-list buffer and buzzer payloads leaked the same way.
Each payload now goes through a reused Java direct ByteBuffer
(app/native/java-direct-buffer.ts): JS writes into it through an
ArrayBuffer.from view, and the Java object itself is passed, which takes
NativeScript's ordinary object path. Every consumer (SurfaceCompositor
applyAndComposite, playBuzzerSequence) copies before returning, so one
buffer per JS context is safe.
Covered: the communicator bridge (frames, draw lists, buzzer, which also
carries EvenHub tones), the preview-only display, and all eight worker
apps. Workers submit through the Android getActiveDisplay() (frames) and
worker-buzzer (tones), so the reused buffers live there rather than in
each worker; the worker sources are shared with iOS, whose active display
takes the ArrayBuffer itself. The
smaller one-off crossings (glyph and image atlas registration, the gray
preview helper, firmware and file writes) are left as they are.
tests/java-direct-buffer.test.cjs covers the helper against a fake
ByteBuffer, and the Android worker-buzzer test stubs it.
(Rebased onto main by James Babcock: worker changes moved into
active-display.ts and worker-buzzer.ts.)
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Ported to the shared Kotlin FlashPromptFlow after the Java was converted. The settings ack the battery read already parses carries silentMode; read it before writing the prompt page, and refuse only on a positive report. An arm that does not answer, or an ack without the field, falls through to the prompt as before.
The native session already clears silent mode on touchpad input and re-reads it on every battery poll, so the TypeScript-side clearing duplicated that. The display-target trigger was also wrong: the communicator stays up in silent mode, so it cleared the flag on the next shell render.
tests/ios-glanceboard.test.cjs requires .test-build/app/g2/events.js, but events.ts was not in tests/tsconfig.json, so a clean build never produced it.
An ack is not a render: the lens can accept the create-prompt page and display
nothing, so a null check on the reply throws away the only thing that can say
what it did. The page size going out and the ack payload coming back are both
logged now.
Silent mode blanks the display and stops the firmware dispatching input while
BLE and GATT stay up, so the create-prompt page is written, accepted, and never
rendered. There is no touchpad event available to answer it with, and the flash
ends after four attempts on "prompt page not acked", which tells the wearer
nothing about what is wrong or what to do about it.
readBattery was already performing a sid-0x09 settings read whose ack carries
silentMode next to the battery, and the field was being discarded. readSettings
is factored out so one read serves both callers, and requireNotSilent runs it
before the prompt page is written. A positive report throws, and the existing
error path turns the message into what the wearer sees: that the glasses are in
silent mode, and that long-pressing both touchpads leaves it.
It refuses only on a positive report. An arm that does not answer, and firmware
whose ack omits the field, both fall through to the prompt exactly as before, so
older firmware is unaffected and a missing answer can never block a flash.
This replaces the phone-side confirmation fallback that was here before. An
unacked page is not by itself evidence that the prompt was never displayed, and
the on-lens step is what proves the glasses being flashed are the ones in front
of you, so overriding it on a timeout gave up more than it was worth.
Silent mode is entered and exited only by long-pressing both touchpads. While
it is on the firmware ignores all input and blanks the display, so from the
phone the glasses look connected and simply answer nothing. The app cannot turn
it off either: the communicator exposes onSilentMode and a listener set, and no
setter.
The status read "Connected (Silent mode enabled)", which names the state but
not the way out of it, and someone who triggered the gesture by accident has no
reason to know the gesture exists. It now names it.
The belief can also go stale. silentMode is set by the firmware's push event
and cleared only when the phase goes to disconnected, on the assumption that it
is re-reported on the way back up. Toggle silent mode on the glasses without
the link ever dropping and that assumption does not hold: a missed "off" leaves
the phone claiming silent mode for as long as the connection lasts. Observed
with the status reading silent while the glasses were granting framebuffer
leases and rendering frames, which cannot happen in a mode whose whole
behaviour is to blank the display and ignore input.
So the belief is dropped whenever the glasses do something that mode makes
impossible: input from them, since the firmware discards input while silent, or
a shell render that gets past the display-target check, since silent mode does
not leave a display target standing. The second matters because it self-heals
without the user touching anything.
Not unit-tested. Both halves are in dashboard-controller.ts, which imports 56
modules, so a test there would mostly exercise mocks, and the stale-clear is
state management across events rather than a pure function. The refactor that
would make it testable is moving the silent-mode belief into a small state
holder taking the firmware's push events plus the two kinds of contradicting
evidence; glasses-display-state.ts next door is already exactly that shape.
renderShell paints first and checks for a display target last, and the
stale-data repaint re-requests before that check. requestShellRender runs
`do { renderShell() } while (shellRenderQueued)`, and because a render is
already in progress the re-request just sets shellRenderQueued. So with no
display target the two feed each other: paint, report stale, re-queue, discard
the work, round again, with no yield and no backoff. SHELL_REFRESH_INTERVAL_MS
is 60 seconds, so the timer is not the driver.
Measured on a device with the glasses not presenting, from the app's own
frame-timings export: 40 of 40 recorded frames were render:shell
target=shell-overlay, every one discarded, mean 55 ms and median 53 ms apart,
so about 19 full renders a second all thrown away. Main thread state=Runnable
with utm=22599, and seven ANR records over three days, each "Input dispatching
timed out ... Waited 10001ms". Afterwards the same state produced 5 renders in
100 seconds, and the export showed a normal mixed workload again instead of
one loop crowding everything else out.
Hoisting the check above the paint fixes both halves at once: a frame with
nowhere to go costs nothing, and the stale-data re-request lives downstream of
the early return, so it can no longer feed the loop. The copy of the check that
used to sit after the paint is removed rather than left in place. Nothing
between the two points can change the result, and the compiler agrees: the
early return narrows this.phase, which made the "charging" comparison
unreachable (TS2367).
submitWindowFrame, the sibling path for in-process window frames, already
checks the display target before doing any work.
Not unit-tested. The change is in dashboard-controller.ts, which imports 56
modules, so a test there would mostly exercise mocks, and it is an ordering
defect rather than a predicate, so extracting a shouldPresent() helper would
pass both before and after. The shape that would work is pulling the render
gate and its stale-data retry into a small module taking the display target and
the stale flag and returning whether to paint and whether to re-request.
Glasses running an older Faceclaw firmware revision cannot be upgraded at all.
The flash asks for confirmation on the lens first, and that firmware never
shows the prompt, so the one path off it is closed. Measured on revision 13
(base 2.2.9.22).
showPrompt writes the EvenHub create-prompt page to the right arm and waits for
an ack. The lens accepts the write at the GATT layer and then answers nothing
and displays nothing: four attempts, four onWriteCharacteristic status=0
length=165, no reply. buildCreatePromptPage documents its geometry as
stock-firmware-safe "so this works before flashing", and that does not hold
there.
Three faults, each only visible once the previous one is fixed.
An unacked prompt page was fatal. Flashing is the only way off that firmware,
so failing here leaves the glasses permanently unupgradable. An unacked page
now falls back to the phone-side confirmation that the low-battery retry
already uses, on a screen that says what is being given up: the on-lens step is
what proves the glasses are the wearer's.
connectArm made a single attempt. The lens is routinely unavailable for a
moment just after a previous link to it closed, which is exactly the state a
retry is in, so the fallback failed instantly with "could not connect to the
right arm". FaceclawFirmwareFlasher.connectLensResilient already retries within
a window for this reason; connectArmResilient mirrors it and reports each
attempt.
That retry then tripped the lost-connection latch. Disconnecting between
attempts fires onConnectionStateChange, which set rightLost AND finished and
counted down the selection latch, so a successful second attempt still aborted
with "Lost connection to the glasses". The callback now ignores drops while the
arms are still being established.
Logging, because a failed flash used to leave almost nothing to read. emitLog
reached logcat, but emitState, emitProgress and emitComplete on the flasher,
and emitState and emitResult on the prompt communicator, went only to the
in-app listener, so the line carrying the reason existed only on the phone
screen. Seven `catch (Exception ignored)` blocks across the two classes now log
with the Throwable, matching the call sites elsewhere that already do.
tests/flash-prompt-fallback.test.cjs covers the decision at its seam through
the existing load-typescript loader. The first test fails against the current
view model and passes with this change; the second is a regression guard and
passes either way, so an unrelated failure cannot quietly skip the on-lens
confirmation.
Verified on hardware: both lenses flashed through this path after it had been a
dead end.