RSDK-12758 — Support webcam unplug replug recovery on darwin

[hexbabe]

RSDK-12758

Uses upstream pion/mediadevices#670.

Will manually test before merging. Please make a first pass review.

Manual Tests

Mac, two webcams

• Initial state (webcam A and B both work)
• Unplug webcam A (webcam A disconnects)
• Replug webcam A (recovers)
• Unplug webcam B (should not affect webcam A)
• Verified debug log runs the right setup observer: 2026-01-06T19:25:51.254Z INFO entrypoint server/observer_darwin.go:26 DEBUG! darwin version!!!!!!

Linux, two webcams

• Same thing. Saw print: DEBUG! non-darwin no-op version!!!!!!!

[cheukt]

looks reasonable to me

[hexbabe]

windows build/tests were failing. It slipped past me that we do no_cgo for RDK windows, which makes the observer stub methods undefined for windows builds. Hence, I declared stubs for all the functions in this repo. This is a bit more verbose, but making another upstream change to the open source code would be difficult.

I thought about consolidating all the stubs into one file in videosource instead of separate stub files in main and in videosource, but that didn't work because if main.go explicitly imports videosource (to call the SetupCameraObserver wrapper), it forces videosource to compile on all platforms, including Windows no_cgo where query.go's reference to mediadevicescamera.Initialize() fails.

[hexbabe]

@-ing you @abe-winter because this ended up being really build-y

The context here is that I made an upstream change to add these "observer" functions you will see in the diff, but I didn't realize we do no_cgo for RDK windows, which made these functions undefined. So I had to add some build flags and extra file stub declarations to make it work.

[abe-winter]

lgtm + oliviamiller is a better reviewer than me for this, but I left a bunch of questions because this is a subtle system / thread thing

[abe-winter]

when this errors, is the system left in a bad state? (i.e. how do you know this is an ignorable error)

[hexbabe]

I don't think the system will be left in a bad state i.e. it shouldn't hinder the rest of viam-server from behaving. But it will make it so that webcams cannot react to hot unplug/replug. Should I add that to the error?

[abe-winter]

imo yes add to error -- better if errors say what to do / what to expect

[abe-winter]

what happens if you get the order wrong? crash?

[hexbabe]

StartObserver actually calls SetupObserver under the hood, but we call SetupObserver first in the entrypoint main.go to allow the objective-c NSRunLoop to set up on the main thread. It would be the "wrong" order if we only called StartObserver in the webcam code and didn't call SetupObserver in main.go

[abe-winter]

does this need to be cleaned up ever?

[hexbabe]

it should be cleaned up by DestroyObserver (see the deferred logic in main.go)

[abe-winter]

is it safe to call startCameraObserver multiple times?

[hexbabe]

yes, I wrote it so that we can make it work like this in RDK (idempotent)

[abe-winter]

what does 'producer' mean here?

[hexbabe]

The part of the OS that pushes the notification that a device was (dis)connected

[hexbabe]

your questions were helpful, I updated comments and error messages to be clearer @abe-winter