Fixed fader detach crash

[aleksandr-voitenko]

Description

Fixed a data race during fader/volmeter detachment and source destruction.

Stack trace:

w32-pthreads 	+0x005277	pthread_mutex_unlock (pthread_mutex_unlock.c:61)
obs 			+0x092c90	signal_handler_disconnect (signal.c:288)
obs				+0x003807	obs_fader_detach_source (obs-audio-controls.c:739)
obs64			+0x025657	osn::Fader::Detach (osn-fader.cpp:213)
...

Motivation and Context

The old detach logic used a raw obs_source_t * and resolved the signal handler from that source during detach. If source destruction won the race, detach could pass a stale/freed handler into signal_handler_disconnect.

Crash Scenario:

1. Frontend sequence is: SceneItem::Remove -> Source::Release -> Fader::Detach.

Note: technically it can be fixed in the frontend, but such a fix would be fragile.

2. Source::Release triggers source remove/destroy on another thread.
3. Fader::Detach runs concurrently and enters detach.
4. Old code reads raw fader->source, then calls obs_source_get_signal_handler(source).
5. Source context/handler may already be in destruction/free path.
6. signal_handler_disconnect touches invalid handler state and crashes in mutex unlock/read (0x8)

What Changed
All changes are made solely in obs-audio-controls.c:

• Weak + temporary strong ref pattern (common in other places) added where source access is required
• obs_fader and obs_volmeter now store weak source handles plus cached handler pointers
• Attach paths now:
  • get weak source (obs_source_get_weak_source)
  • connect callbacks via signal_handler_connect_ref
  • store weak source + handler
• Detach paths now:
  • atomically swap out and clear source/handler under mutex
  • disconnect using stored handler pointer
  • release weak source

Why This Fix Probably Works 😊

• Detach no longer depends on raw source lifetime.
• signal_handler_connect_ref guarantees handler lifetime until matching disconnect.
• Detach is idempotent/race-safe (fields cleared before disconnect).
• Temporary strong refs prevent transient use-after-free while operating on source internals.

Conclusion

• Objects' lifetime in OBS is mess

How Has This Been Tested?

Win/Mac, manually

Types of changes

• Bug fix (non-breaking change which fixes an issue) -->