refactor(camera): manage a device-wide feed without app resets by gwdp · Pull Request #148 · expo/serve-sim

gwdp · 2026-09-11T17:07:32Z

Why

The camera is a device-wide capability. Enabling or disabling its feed should behave like connecting or disconnecting camera hardware while apps and SpringBoard continue running.

How

Load the camera through the capability loader for all apps, including apps opened outside serve-sim.
Enable connects the selected feed; Disable disconnects it without restarting apps or changing camera permissions.
Split camera commands, helper lifecycle, media handling, and UI into focused modules.

Test Plan

CI
Tested image feeds and Enable/Disable on EAS Simulator at the tip of the stack.

linear-code · 2026-09-11T18:48:52Z

ENG-26327

ENG-26998

@public

gwdp pointed out that #148, #102, and #53 import setCapabilityEnabled. Restore it and its only callee with a @public tag that names those PRs, so knip stays clean and the first PR to land makes them live. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

@public

gwdp pointed out that #148, #102, and #53 import setCapabilityEnabled. Restore it and its only callee with a @public tag that names those PRs, so knip stays clean and the first PR to land makes them live. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

@public

* Run knip in CI and delete the dead code it found knip was configured but never ran in CI, so unused exports piled up. Tune the config so test files count as entry points, exports used only in their own file are not reported, and the Swift-only node-swift dependency is declared. Then delete what remains: five unused response helpers in middleware-utils, an unused capability toggle and its only caller, legacy device-frame constants, and three unused client helpers. Add a Knip job to the EAS Checks workflow and to `bun run check`. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * Say that check runs knip too Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * Keep setCapabilityEnabled and capabilityDefinition for open PRs gwdp pointed out that #148, #102, and #53 import setCapabilityEnabled. Restore it and its only callee with a @public tag that names those PRs, so knip stays clean and the first PR to land makes them live. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

@public

* Run knip in CI and delete the dead code it found knip was configured but never ran in CI, so unused exports piled up. Tune the config so test files count as entry points, exports used only in their own file are not reported, and the Swift-only node-swift dependency is declared. Then delete what remains: five unused response helpers in middleware-utils, an unused capability toggle and its only caller, legacy device-frame constants, and three unused client helpers. Add a Knip job to the EAS Checks workflow and to `bun run check`. * Say that check runs knip too * Keep setCapabilityEnabled and capabilityDefinition for open PRs gwdp pointed out that #148, #102, and #53 import setCapabilityEnabled. Restore it and its only callee with a @public tag that names those PRs, so knip stays clean and the first PR to land makes them live. --------- (cherry picked from commit df029c8) Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

gwdp · 2026-09-22T23:24:45Z

/devin review

devin-ai-integration

Note

Newer findings are available below. Devin Review posted a newer report on this PR, in addition to the findings presented here.

Devin Review found 5 potential issues.

devin-ai-integration · 2026-09-23T20:55:54Z

the review guide

🔍 Validation evidence needs follow-up

The native test evidence omits the simulator and Xcode version. The CLI and UI changes also lack command output and visual evidence required by the review guide.

Was this helpful? React with 👍 or 👎 to provide feedback.

devin-ai-integration · 2026-09-23T20:55:55Z

The contribution guidance

🔍 Change exceeds review scope guidance

This PR combines native lifecycle, wire format, CLI, ownership, and UI changes across roughly 2,700 lines. The contribution guidance requests one behavior change under roughly 400 lines.

Was this helpful? React with 👍 or 👎 to provide feedback.

devin-ai-integration · 2026-09-23T20:55:57Z

        const reply = await fetchCameraStatus();
        if (cancelled) return;
        const alive = !!reply?.alive;
-        const replyBundles = Array.isArray(reply?.bundleIds) ? reply.bundleIds : null;
-        const foregroundIsInjected =
-          !!bundleId && (replyBundles ? replyBundles.includes(bundleId) : injectedBundleIds.has(bundleId));
-        const replySource = reply?.source ?? null;
-        const replyHasRealSource = replySource && replySource !== "placeholder";
-        const attachedToCurrentHelper =
-          injected && alive && foregroundIsInjected && !!replyHasRealSource
-          && (attachedHelperPid == null || reply?.helperPid === attachedHelperPid);
-        setPillState((prev) => nextCameraPillState(prev, attachedToCurrentHelper));
-        if (!alive) {
-          setInjected((prevInjected) => {
-            if (!prevInjected) return prevInjected;
-            setInjectedBundleIds(new Set());
-            setAttachedHelperPid(null);
-            appliedMirrorRef.current = "off";
-            return false;
-          });
-        } else if (injected && attachedHelperPid != null && reply?.helperPid !== attachedHelperPid) {
-          setInjected(false);
-          setInjectedBundleIds(new Set());
-          setAttachedHelperPid(null);
-          appliedMirrorRef.current = "off";
-        } else if (alive && Array.isArray(reply?.bundleIds)) {
-          const next = reply.bundleIds;
-          setInjectedBundleIds((prev) => {
-            if (prev.size === next.length && next.every((b) => prev.has(b))) return prev;
-            return new Set(next);
-          });
-        }
+        setPillState((prev) => nextCameraPillState(prev, alive));
+        setEnabled(alive);


🔍 Active status tracks only helper liveness

The poll maps reply.alive directly to Active. It cannot distinguish a running helper from a published and connected camera capability.

Was this helpful? React with 👍 or 👎 to provide feedback.

…ether

…he published feed

devin-ai-integration

Note

Newer findings are available below. Devin Review posted a newer report on this PR, in addition to the findings presented here.

Devin Review found 4 new potential issues.

devin-ai-integration · 2026-09-23T22:39:57Z

🔍 Rollback plan is missing

The persisted ownership schema and shared-memory protocol both change, but the PR provides no rollback guidance required by AGENTS.md.

Was this helpful? React with 👍 or 👎 to provide feedback.

…nd confirm helper exit

devin-ai-integration

Note

Newer findings are available below. Devin Review posted a newer report on this PR, in addition to the findings presented here.

Devin Review found 4 new potential issues.

devin-ai-integration · 2026-09-23T23:21:30Z

🔍 Change scope lacks unaffected-device evidence

This PR combines camera, CoreMotion, lifecycle, CLI, and UI changes. It provides no required tap, drag, and scroll evidence for unaffected devices.

Was this helpful? React with 👍 or 👎 to provide feedback.

…ed, and fail a helper with no socket

…r a stale one

…reject unknown camera commands

…nd park early preview layers safely

devin-ai-integration

Note

Newer findings are available below. Devin Review posted a newer report on this PR, in addition to the findings presented here.

Devin Review found 1 new potential issue.

devin-ai-integration · 2026-09-24T00:30:26Z

simcam_addOutput

+    BOOL undecided = !SimCamSessionHasFakeCamera(self);
+    if (undecided) SimCamPromoteOpenConfigToNative(self);
    [self simcam_addInput:input];
+    if (undecided) [self simcam_handOverHeldOutputs];


🔴 Mixed camera sessions lose microphone samples

When a fake-camera session adds an audio output before its microphone input, undecided prevents handing the output to AVFoundation. simcam_addOutput only tracks that output, so its audio delegate receives no microphone samples.

Learn more

Outputs on a session without a native input are stored in simcam_addOutput instead of being added to AVFoundation. When a fake camera is present, the new gate leaves these outputs tracked after a real microphone input arrives. The fake output registry registers only video data delegates, so an audio data output cannot receive native microphone samples. The same gate exists in simcam_addInputWithNoConnections.

Example: An app adds a fake video input, an AVCaptureAudioDataOutput, and then an AVCaptureDeviceInput for its microphone. The audio output remains in SimCamSessionTrackedOutputs; recording receives no audio buffers, although the microphone input was added.

Recommended fix: When a real input joins a fake-camera session, move held audio outputs into AVFoundation while retaining fake video outputs in the synthetic session. Handle both real-input methods and preserve outputs added without connections.

Was this helpful? React with 👍 or 👎 to provide feedback.

…der to the e2e CLI

devin-ai-integration

Note

Newer findings are available below. Devin Review posted a newer report on this PR, in addition to the findings presented here.

Devin Review found 1 new potential issue.

devin-ai-integration · 2026-09-24T01:34:21Z

SimCamFakeConnectionForOutput

+        alive = [NSHashTable hashTableWithOptions:
+            NSPointerFunctionsStrongMemory | NSPointerFunctionsObjectPointerPersonality];
+    });
+    @synchronized (alive) { [alive addObject:connection]; }


🟡 Camera connections accumulate after output teardown

When apps repeatedly create camera outputs, SimCamKeepFakeConnectionAlive retains each connection forever. Releasing an output cannot remove its connection from alive, so memory grows throughout the app's lifetime.

Learn more

A fake connection is created once per output by SimCamFakeConnectionForOutput. The output retains it through an associated object, but this table retains it independently. After the output deallocates, the connection's weak output reference clears while the table keeps the connection for the rest of the app process. Repeated creation and disposal of capture outputs therefore grows the table even when no camera session remains.

Example: An app opens and closes a camera screen 1,000 times, creating a new video output each time. The 1,000 abandoned connections remain in alive; the expected retained count after teardown is zero.

Recommended fix: Give SimCamFakeConnection a safe teardown path that avoids the uninitialized superclass state, then remove the permanent strong table and let its associated output own it. Verify output deallocation and repeated capture-session creation on the supported simulator runtime.

Was this helpful? React with 👍 or 👎 to provide feedback.

…mments

devin-ai-integration

Devin Review found 3 new potential issues.

devin-ai-integration · 2026-09-24T02:45:26Z

simcam_capturePhotoWithSettings:

+    if (!SimCamSessionHasFakeCamera(self)) {
+        [self simcam_handOverHeldOutputs];
+        objc_setAssociatedObject(self, &kSimCamSessionFakeRunKey, @NO, OBJC_ASSOCIATION_RETAIN);
+        [self simcam_startRunning];
+        return;


🔴 Native photo captures become synthetic

After device-wide loading, simcam_capturePhotoWithSettings: intercepts photo captures without checking whether the output belongs to a fake session. Native photo outputs receive synthetic images, even while the fake camera is disconnected.

Learn more

The injector now loads into every eligible app, including apps using native AVFoundation sessions. simcam_capturePhotoWithSettings: always synthesizes a photo without checking whether its output is attached to a fake session. The new native-session branch does not protect photo capture, because capture calls go directly to the output swizzle.

Example: An app with a native photo output calls capturePhotoWithSettings: after the feed is disabled. Instead of using AVFoundation's native capture, the output delivers the synthetic black fallback photo.

Recommended fix: Gate photo-output interception on attachment to a fake session and forward other captures to the original implementation. Apply the same native-session gate to other globally swizzled output operations that manufacture fake results.

Was this helpful? React with 👍 or 👎 to provide feedback.

devin-ai-integration · 2026-09-24T02:45:28Z

hasProcessExited

+            if (gShmHeader && (!atomic_load_explicit(&gShmHeader->active, memory_order_acquire) ||
+                              kill((pid_t)gShmHeader->ownerPid, 0) != 0)) {


🟡 Zombie helper leaves camera connected

If the helper dies unreaped, SimCamRefreshDevice treats its PID as alive and never disconnects the old mapping. Apps keep the stale camera and cannot connect to a restarted helper.

Learn more

An unexpected helper exit cannot clear the shared header's active flag. A zombie process still passes kill(pid, 0) until reaped, so the monitor leaves its old shared-memory mapping attached and never checks the newly named region. hasProcessExited already handles this condition for host-side helper management.

Example: The helper is killed, remains a zombie, and a new helper opens the same shared-memory name. The injected app continues treating the old mapping as connected and misses the new feed.

Recommended fix: Detect zombie state as well as PID existence on the injected side, or use a liveness signal that does not depend on reaping; ensure the monitor closes the old mapping before retrying attachment.

Was this helpful? React with 👍 or 👎 to provide feedback.

devin-ai-integration · 2026-09-24T02:45:29Z

SimCamRegisterFakeVideoOutput

+        OBJC_ASSOCIATION_RETAIN);
    SimCamSetPosition(output, SimCamPositionOf(self));
    SimCamMarkOutputAttachedToFakeSession(self, output);
+    SimCamRegisterFakeVideoOutput(output);


🟡 Unstarted camera sessions receive samples

When an output already has a delegate, SimCamRegisterFakeVideoOutput registers it before startRunning. The pump ignores session running state and sends samples to an unstarted session.

Learn more

SimCamRegisterFakeVideoOutput adds outputs immediately when they already have a delegate. startPumpingIfNeeded starts on registration, and its delivery loop checks connection generation but not whether the owning session is running. The new registration path thus delivers samples before the app starts capture.

Example: An app assigns a sample delegate, adds the output, and pauses before calling startRunning. The delegate receives synthetic samples during the pause.

Recommended fix: Track the owning fake session and gate queued delivery on its running state; also stop or unregister delivery when that session stops.

Was this helpful? React with 👍 or 👎 to provide feedback.

gwdp added this pull request to stack #150 September 11, 2026 17:07

gwdp changed the title ~~gwdp/camera refactor~~ refactor(camera): manage a device-wide feed without app resets Sep 11, 2026

gwdp marked this pull request as ready for review September 11, 2026 17:15

gwdp force-pushed the gwdp/camera-refactor branch from 25777f8 to d791f67 Compare September 11, 2026 18:48

gwdp requested a review from krystofwoldrich September 11, 2026 19:01

gwdp force-pushed the gwdp/camera-refactor branch 2 times, most recently from 37a3d58 to 5e2a013 Compare September 11, 2026 19:37

gwdp force-pushed the gwdp/camera-refactor branch from 5e2a013 to d4507d6 Compare September 12, 2026 00:52

gwdp removed this pull request from stack #150 September 14, 2026 16:40

gwdp added this pull request to stack #160 September 14, 2026 16:40

gwdp force-pushed the gwdp/camera-refactor branch 2 times, most recently from cda5b25 to d09d1fa Compare September 21, 2026 22:07

gwdp force-pushed the gwdp/camera-refactor branch 3 times, most recently from 661f259 to 9ef6db4 Compare September 21, 2026 23:06

Base automatically changed from gwdp/trampoline-c-tests to expo September 21, 2026 23:07

gwdp force-pushed the gwdp/camera-refactor branch 2 times, most recently from 56a6098 to bb31f75 Compare September 21, 2026 23:42

gwdp mentioned this pull request Sep 22, 2026

Run knip in CI and delete the dead code it found #185

Merged

szdziedzic mentioned this pull request Sep 22, 2026

Run knip in CI and delete the dead code it found #193

Merged

gwdp requested a review from szdziedzic September 23, 2026 20:51

gwdp force-pushed the gwdp/camera-refactor branch from bb31f75 to 45a2030 Compare September 23, 2026 20:51

devin-ai-integration Bot reviewed Sep 23, 2026

View reviewed changes

gwdp force-pushed the gwdp/camera-refactor branch from 45a2030 to ccfade7 Compare September 23, 2026 20:57

gwdp added 9 commits September 23, 2026 15:36

[serve-sim] fail required E2E without a simulator

7eb7139

refactor(camera): manage a device-wide feed without app resets

f77fd32

fix(camera): harden capability and feed lifecycles

a9768e1

fix(launch): preserve session ownership during shutdown

671cfd5

[serve-sim] separate camera commands lifecycle and presentation

f454639

[serve-sim] remove unused camera return values and re-exports

d6ed832

refactor(camera): keep source validation and presentation cleanup tog…

d0446bc

…ether

[camera] clarify device-wide camera permissions

f763b74

fix(camera): share capability ownership, always push mirror, follow t…

80b6a8a

…he published feed

devin-ai-integration Bot reviewed Sep 23, 2026

View reviewed changes

gwdp force-pushed the gwdp/camera-refactor branch from 99bceac to 80b6a8a Compare September 23, 2026 23:07

fix(camera): leave other capture sessions and motion sensors alone, a…

e332436

…nd confirm helper exit

devin-ai-integration Bot reviewed Sep 23, 2026

View reviewed changes

gwdp added 2 commits September 23, 2026 16:36

fix(camera): install the motion prop only where the fake camera is us…

8f772f9

…ed, and fail a helper with no socket

fix(camera): bind the control socket before the first source and clea…

25cdede

…r a stale one

This comment was marked as resolved.

Sign in to view

fix(camera): keep native outputs and previews off the fake pump, and …

50d4cff

…reject unknown camera commands

This comment was marked as resolved.

Sign in to view

fix(camera): keep a fake session's outputs when a real input joins, a…

4da10d8

…nd park early preview layers safely

devin-ai-integration Bot reviewed Sep 24, 2026

View reviewed changes

gwdp added 2 commits September 23, 2026 18:19

fix(camera): keep fake connections alive, and pass the temp state fol…

43a2a4b

…der to the e2e CLI

test(camera): pass the temp state folder to the launch e2e cleanup

2ff9a95

devin-ai-integration Bot reviewed Sep 24, 2026

View reviewed changes

gwdp added 2 commits September 23, 2026 19:12

test(camera): look up fixture containers once, and trim review-fix co…

8612c70

…mments

chore(camera): keep the whys the comment trim dropped

1569532

devin-ai-integration Bot reviewed Sep 24, 2026

View reviewed changes

		if (gShmHeader && (!atomic_load_explicit(&gShmHeader->active, memory_order_acquire) \|\|
		kill((pid_t)gShmHeader->ownerPid, 0) != 0)) {

Conversation

gwdp commented Sep 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why

How

Test Plan

Uh oh!

linear-code Bot commented Sep 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gwdp commented Sep 22, 2026

Uh oh!

devin-ai-integration Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

devin-ai-integration Bot Sep 23, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Sep 23, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

devin-ai-integration Bot Sep 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Sep 23, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

devin-ai-integration Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Sep 23, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

devin-ai-integration Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Sep 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Sep 24, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot left a comment

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Sep 24, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Sep 24, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration Bot Sep 24, 2026

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

gwdp commented Sep 11, 2026 •

edited

Loading

linear-code Bot commented Sep 11, 2026 •

edited

Loading

devin-ai-integration Bot left a comment •

edited

Loading

devin-ai-integration Bot Sep 23, 2026 •

edited

Loading

devin-ai-integration Bot left a comment •

edited

Loading

devin-ai-integration Bot left a comment •

edited

Loading

devin-ai-integration Bot left a comment •

edited

Loading

devin-ai-integration Bot Sep 24, 2026 •

edited

Loading

devin-ai-integration Bot left a comment •

edited

Loading